Pentium 4 what socket. 3D modeling packages

[ Print version ]

In principle, a lot of what is in this review has already been previously considered in materials published by us not so long ago - these are the SiS 735 , VIA Apollo KT266 and VIA P4X266 chipsets, coupled with the new Intel i845 chipset (Brookdale). Besides, the Pentium 4 1.7 GHz present in this review was also covered in a separate large material. However, all these platforms this testing will no longer play the main role, but only the role of a background for the real "heroes of the day": Pentium 4 2 GHz "Willamette" (Socket 423) and Pentium 4 2 GHz "Willamette" (Socket 478).

Let's take a closer look at our heroes:

On the left is a Pentium 4 for Socket 478, and on the right is a Pentium 4 for Socket 423

Pentium 4 dimensions for Socket 478 are amazing, especially next to the good old i386DX :)

The new Socket 478 looks very elegant, as does the cooler for the new socket, compared to the cooling system for Socket 423

Although the Pentium 4 processor for Socket 478 is still significantly smaller than the cooler for it

It would seem that the event is not as epochal as, say, the release of the first Pentium 4, because the core in the new processors is still the same - Willamate (see). However, we have already written more than once that for the processors of this platform, each frequency increase is an event, because the frequency is their main trump card, it is the "alpha and omega" of the performance of the main Intel platform. However ... let's make a small lyrical digression. What is the essence of rivalry between Intel and AMD? What goals does each company pursue and by what means does it plan to achieve them? The next part of the review in some way claims to attempt an analytical approach to these issues. Although after reading it, it may seem to someone that, on the contrary, there are only more questions :)

Intel Pentium 4 and AMD Athlon: Confrontation Aspects

It is no longer a secret to anyone that not only the well-being of individual companies and their closest "friendly environment" depends on the outcome of the ongoing processor wars in the camp of the x86 platform, but also in many respects what development paths the entire platform will take in further. And, of course, processor giants clashed in a fierce battle and their products cannot differ only in the number of transistors in the next processor and the name written on its upper part. Everything is much more serious. From a conceptual point of view, the confrontation between Pentium 4 and Athlon can be divided into several relatively independent aspects. Which we will do.

1. Frequency vs Optimized Architecture. It's no secret that Athlon reacts much more calmly to "chaotic" code containing a large number of conditional jumps due to the shorter pipeline length. One could say that AMD is encouraging sloppy programmers, but we still refrain from such a bold statement :). The Pentium 4 pointedly ignores this way of increasing performance. For the Intel processor, the main thing is the frequency. Let them have to return, "flying" the next turn, the main thing is the speed on the straight line. In fact, by and large, both ways are good, and neither can claim to be the ultimate truth. Yes, in order to outperform Athlon, Pentium 4 must always outperform it in frequency. However, this is exactly what we have been seeing for the entire time since its release! Another thing is that it should be ahead of Athlon in terms of frequency. much. So far, there is a "balance on the brink" in this regard, and sometimes (the recent situation of simultaneous neighborhood on the market of Pentium 4 1.7 GHz and Athlon 1.4 GHz) - with an obvious shortage on the part of Intel. What happens next, only time will tell.

2. RDRAM vs DDR SDRAM. Many will remember the recently released VIA P4X266 review and say that this confrontation has already lost its relevance with its release. However… The fact is that Pentium 4 and RDRAM are by and large quite well suited to each other! For those who doubt the predisposition of the Pentium 4 (at least its current version) to Rambus memory, we will give a simple example: why, in fact, the resulting bus bandwidth of this CPU is equivalent to exactly 400 MHz? Why not 266 or 532? But because it was originally designed for PC800 RDRAM. All modern subtypes of DDR SDRAM (except, sorry, the "killed" variant in the form of PC1600 200 MHz DDR) will work with Pentium 4 in asynchronous mode. And asynchrony is the worst enemy of performance, any hardware professional will tell you that.

3. Manufacturability versus technological sophistication. The Athlon "Thunderbird" is still manufactured in 0.18 micron technology. Pentium III-S and Pentium III-M are already being produced according to the 0.13 micron process technology and are being delivered to the market. Already on the horizon is a new Pentium 4 (with the new "Northwood" core) that will be manufactured using 0.13 micron technology and a new Celeron that will also be manufactured using a 0.13 micron process technology. The Athlon "Palomino" (Athlon 4, Athlon MP) is again made using 0.18 micron technology, but with copper conductors. AMD squeezes everything it can out of the old process, Intel just switches to one that allows you to drive the frequency without unnecessary frills. In principle, it is understandable - in terms of the quantity and quality of equipment of factories, none of the manufacturers of modern processors can still compete with Intel.

4. Additional command sets. Intel embarked on this path even with the advent of MMX, and is further successfully moving along it - the release of Pentium III was marked by the arrival of SSE, Pentium 4 - SSE2. AMD tried to repeat the success of MMX with its 3DNow! set, but, despite the optimism that had flared up among software writers, still at the moment more numbers are again leaning in favor of multimedia command sets from Intel. Why? The author is not an expert in the field of programming, however, even a simple look at the list of commands themselves and their capabilities in the context of "SSE2 vs. 3DNow! Professional" (the latest degrees of perfection from both companies) is enough to make sure that the set from Intel is more advanced . By the way, this point of view was indirectly confirmed by AMD itself, including in 3DNow! Professional partial support for SSE, and in the next generation processor it is developing - and simply full support for SSE2 (fortunately, the license from Intel has been obtained quite a long time ago).

5. Marketing aspect. Is there anyone who wants to object? Marketing is the strongest weapon in the market, and it is the market that ultimately determines the winner. The user votes with his wallet, buying a computer based on a particular processor, he has no other way to influence the course of events. Intel's point of view is simple, uncomplicated... hmmm... a little does not correspond to the strict scientific approach :), but, nevertheless, it is quite consistent with the average position of an average computerized layman: they will buy a processor with a higher clock frequency. That is: if two processors show the same performance, but one operates at a frequency of 2 GHz, and the second at a frequency of 1.4 GHz, then they will buy the one that operates at a frequency of 2 GHz. Why? And because gigahertz, megamiles, kilopascals, and everything else with various "mega" and "giga" prefixes - flatters the buyer simply on an instinctive level. You can make a car with a speed of 200 km / h, and with an engine with a capacity of 200 horsepower. But 350 looks much cooler in the passport :)

6. Price wars. Having come to their senses after the disastrous failure of the first copies of the Pentium 4 "Willamette", which, with their exorbitant prices (including not only for the processor, but also for finished systems - let's not forget about the i850 and RDRAM) plunged the world computer community into a slight shock, more Intel no longer made mistakes. In the end, it's enough that RDRAM costs much more than PC133 and PC2100 DDR, and this is still the only type of memory for computers based on Pentium 4 (there are no boards based on VIA P4X266 or i845 "Brookdale" on the market yet). In the meantime, Intel is unable to do anything with the memory, it is steadily lowering prices for the Pentium 4 itself, and with the release of 0.13-micron Northwood, they will begin to fall even faster. In fact, now the company is forced to use all the methods that AMD has long used against it. What about us? :) The more "they" fight, the more money remains in our pocket. And if the slipping rumors that the Pentium 4 1.4 GHz will soon cost less than $100 are really true, then it should be recognized that regardless of the results of these wars and the final winner, users are the absolute winner from them.

Well, perhaps, quite a sufficient list. It might seem to someone that we mixed the sinful with the righteous, technology with business, programming and hardware in one stew. Yes, that's probably what it is. But what if this is the reality?

But enough theorizing. Today we have the opportunity to compare the performance of a wide variety of systems, which we unite according to one fundamental principle: they either claim the highest performance in their class, or they have recently been announced, and their performance is still rather little known, or both at once :) To begin with - about fees.

Board Specifications

Now more about each board:

Supermicro P4SBA

The motherboard from a company that rarely changes Intel was the first sign that appeared in our laboratory. We won't talk about Supermicro, it is very well known in its niche - boards for high-performance workstations and servers. The boards are highly reliable, but not always - high performance. However, on the field where the company plays, the first is much more important.

Now, having told the reader about the original mission of the company, let's move on to the board itself. It (which in this case works for the image of stability) is packed in a box of the usual design for the company. The box, it should be noted, is quite impressive in size - the reason for this was a fairly good set of accessories included with the motherboard, despite the fact that it has not yet entered mass production. From the standard package here you can find an ATA66/100 cable, a cable for connecting a floppy drive, a full manual for configuring and installing the board on English language and a CD with various software, a set of which we will cover a little later. And now I’ll tell you about what was unusual in the box - there was a bar with additional USB port on the back of the computer - a small thing often found with boards from other manufacturers, finally appeared in the box from Supermicro. Also in the box there is (which is a more pleasant and useful thing) a cooler from Foxconn. Earlier we also came across coolers in boxes from this company, and as we can see, this is becoming a good tradition. Now let's describe what we got on the CD.

As usual, there were drivers for different versions Supermicro boards based on the i8xx chipset series from Intel, their descriptions in .pdf format, which we first encountered on similar Adobe Acrobat discs Reader version 5.0 to read these files, a program for monitoring and remotely managing a computer - Intel LANDesk Client Manager 6 (LDCM) and DirectX 8.0. As you can see, there is nothing superfluous, the company is true to itself here too, without being distracted by the tasks of attracting customers by distributing free software - as a rule, the clientele of such companies is formed, and not high-profile PR campaigns, but private conversations of system administrators serve as advertising :)

The board itself does not impress with any abundance of functions - healthy asceticism in the performance of this company is known to everyone, because the emphasis is not on home users, but on another market sector, where this abundance is often harmful. However, the board is not so deprived of them, but we will talk about this a little later, when describing the BIOS. On the board, the inscription "Designed in USA" flaunts, which, apparently, should attract the buyer. Indeed, the board is mounted on high level, but the convenience of working with it suffered a little - the CD-in connectors are located in front of the PCI slots, which can make it difficult to access them. The board uses an ATX power connector with 24 pins versus 20 pins for the connector we are so familiar with. But this did not stop us from using a conventional power supply - we just need to note that it is advisable to choose a more powerful power supply. The northbridge (MCH, as Intel now calls it) houses a low-profile heatsink. Indeed, it makes no sense to erect giants on a microcircuit without a built-in video adapter. There are almost no switches on the board, there are only three of them, and all their functions are simple: clear CMOS, enable/disable the built-in audio codec, disable/enable the ability to start the computer from PS/2 devices. All possibilities for system configuration are hidden in the BIOS, to the description of which we are moving on.

Its choice is quite unexpected - the Medallion 6.00 version from AWARD is exotic for boards from this company, although we have seen boards with it. In the BIOS, we can adjust the memory timings, adjust the operation of the AGP bus and, which was interesting in our case, change the processor multiplier - the fact is that we received an engineering sample of the processor, where this function was not blocked. In principle, the choice of settings is not rich, but if you look at the box and see the name of the board manufacturer, then the tone with which we describe them will easily become clear.

A short summary about the board can already be made - more detailed conclusions, as usual, we will bring to the surface in the upcoming review of boards based on this chipset. And the summary is as follows - the board is of high quality, most likely not with the best speed indicators (however, who knows...) and amazingly stable - exactly what we used to expect from the boards of this company.

EPoX 4T2A

The popularity of the company is growing by leaps and bounds, and the reason for this is not only the advertising company, but also the really excellent quality of the products. Quality, expressed not only in the smooth operation of motherboards, but also in excellent functionality, also included in general concept"quality".

And this time, users who have chosen a blue box with the inscription "EPoX" will not go wrong, it seems. But more about that later - for now, about what lies in a nice blue box, and organically complements the board. In the box you can find the following: a user manual made in the same style as the box, 80- and 40-wire cables for connecting IDE devices, a floppy drive cable, a bracket for the back of the computer with two additional USB ports, and a compact disk with the software necessary for the operation of the board. Also on it you can find several often necessary programs- antivirus package from Trend Micro Inc. - PCCillin 2000, Symantec Norton Ghost 6.03 for working with hard drives and "reader" format.pdf - Adobe Acrobat Reader.

Also, some utilities were included in the number of programs: for overclocking the processor from Windows - Boostek, and several utilities for monitoring the system - actually from EPoX, from VIA, and from Winbond. It seems to be not much - other manufacturers have more programs on the disk (and sometimes on disks), but there is everything you need to work.

On the textolite of the traditional green color, there was no POST code indicator familiar on other boards of this manufacturer - no doubt it's a pity, especially since the dimensions of the board seem to allow installing it. Among the features of the board, one can note the installation of a sound chip from C-Media and the presence of a latch on the AGP slot. Initially, apparently, the connector was supposed to be an AGP Pro construct, as evidenced by a reminder on the board about the use of the rear part of the connector, but for some reason a regular connector was installed on the board. The CD-in and AUX-in audio inputs are located in front of the PCI and AGP slots - not the most convenient place, because with expansion cards inserted into these slots, access to it will be difficult, otherwise there are no complaints about the wiring. 12 LowESR capacitors are used in the power circuits of the processor core - 8 of 2200 uF and 4 of 1500 uF each, this is more than enough to ensure stable operation of the board in non-standard modes. There are several switches on the board - the most interesting of them for us contained the ability to set the base frequency of the processor - 100, 103, 105, 108, 110, 115, 118, 120, 122, 125 and 133. Strangely, in BIOS Setup this function does not exist (not strange in itself, but in relation to the boards of this company).

The BIOS itself, based on version 6.00 from AWARD, contains a fairly large selection of settings - there are several settings for memory operation, the AGP bus, etc. The board will also please overclockers who are looking for a board with a good overclocking potential - you can change the processor voltage in 0.025 V increments, the AGP bus voltage in 0.1 V increments, and the memory voltage in the same 0.1 V increments.

The board, which clearly confirmed the right of EPoX to be considered one of the best motherboard manufacturers. It will take its rightful place both in a high-performance computer and in a reliable server.

Intel M850MD

It is customary to start the description of the board with a couple of words about the manufacturer. But this time there is nothing to say - the name is so well known, and the history of the company has been studied more than once almost with a microscope, so we will not engage in plagiarism and go straight to the story about the features of this product.

Unfortunately, we won't be able to tell much, because the product came to our table "a goal like a falcon". That is, both the box and the entire standard kit were missing. But still, a pre-production sample came to our laboratory - therefore, the attitude towards it will not be too harsh. However, I think we will not be mistaken if we say that the design of the box will be ordinary - another confirmation of stability, and the set of accessories is minimal.

The board did not become an "engineering genius" - this was hard to expect, knowing about the board format, which limits the flight of that very thought. It should be noted that not only this has become a limitation, but also the company's orientation towards corporate customers, who do not need extra "spillikins" affects here. And now - closer to the body of the patient.

The board has a successful layout - the skill of the R&D department engineers affects, almost all functional elements located in easy-to-access locations - only the audio inputs can be covered by the PCI cards located on top of them. It seems unnecessary to talk about the quality of the installation - the company will not allow its name to be tarnished by such things. There are several unsoldered elements on the board - for example, there is no CNR slot on it, the place for which is reserved to the left of the 3rd PCI slot. However, the board does not leave the impression of a defect - everything is done neatly, as mentioned above. There are several jumpers on the board - and the most interesting of them is undoubtedly the one that allows us to enter the BIOS configuration menu. It's no secret that there are practically no options for system configuration on Intel motherboards. Here, with the help of this jumper, we can change the processor multiplier, reset passwords, etc.

The BIOS, on the other hand, is based on the Phoenix BIOS and has an interface very similar to the Medallion version from AWARD, and this is not surprising - the companies have merged into one, and, accordingly, use each other's developments. It has several settings - the most interesting thing is the ability to manually distribute interrupts across PCI slots (and this is on mATX board!), view the progress of the board loading in the form of a list of events that have occurred and enable or disable the proprietary technology "Intel Rapid BIOS Boot", which allows you to go through the POST procedure a little faster. Very poor, to be sure - but expecting something else from a company that dictates its terms to the industry is simply stupid.

The board is typical for Intel - not very fast, but has good stability. It is quite suitable for someone who would like to buy a computer and forget about problems with it.

VIA P4X266 Reference board

The board has already been described by us in an article talking about VIA chipset P4X266, but for the convenience of readers, we will give its description again. This is a reference board from VIA. Note that the company's engineers did their best to make their brainchild impress - the red PCB for the reference board is something new :). The rest of the product looks quite ascetic - 3 PCI, 1 AGP, 2 DDR DIMMs. The board has a connector for a "large" Pentium 4 (Socket 423), however, according to information received by us, motherboard manufacturers are planning models mainly with Socket 478.

The BIOS settings do not differ in any frills and it makes no sense to describe them separately. General impression from the board - a typical reference board i.e. a product whose main task is to demonstrate the performance and stability of the chipset. Such a board should pass all, even the most difficult tests, to the end (which it actually coped with "with a bang"), and not show outstanding performance.

ASUS A7V266

This first member of the ASUS motherboard family based on VIA's KT266 chipset has been delayed. Almost all competitors have long announced boards based on this chipset, but the largest motherboard manufacturer in the world was waiting. It is possible that the reason for this was the desire not to release a raw product - so let's see what came of it. The board that came to our lab was a pre-sale sample. And talking about any configuration would not be entirely true - after all, comparing raw meat and a cutlet is difficult and incorrect. But still it is worth noting that we found documentation in the kit, albeit printed on a printer - but still it was. Also in the red box were an ATA100 cable, a cable for a floppy drive and a CD-R disk. It contained a standard set of programs from ASUSTeK - drivers for the board, Adobe Acrobat Reader 4.05, antivirus program PC-Cillin2000v. 7.0 by Trend Micro, Power Player SE 5.0, PowerDVD Trial 3.0, VideoLive Mail 4.0 by CyberLink, 3Deep setup software colors in games, as well as programs from ASUS itself - a screensaver, ASUS Update (a BIOS update program from a Windows environment) and ASUS PCProbe - a system monitoring utility.

We did not make a reservation when we said "the first of the family" - the board has unsoldered microcircuits and connectors. Promise's IDE RAID controller with the accompanying ATA100 connectors should be in this location. Just like the boards by Chaintech, an external sound solution is used - a chip from C-Media with the ability to output sound over 6 channels. But there is no bracket for the rear panel in the kit, and through the drivers, the outputs of the center and rear channels become Line-In and Mic-In connectors. The board, despite its infantile (if I may say so) status, was not an object of study made on the knee - there are no attached elements on it that are often found in such products, the soldering was obviously done in the factory. The only drawback is the inconvenient position of the floppy cable connector - it is located behind the 4th and 5th PCI slots, and the cable stretching through the entire case does not provide the best air circulation. There is no point in considering the case with a full-sized PCI card now - they have almost disappeared. The efforts of the company's engineers to improve the power circuits of the processor core are nothing special in comparison with competitors - 4 capacitors of 3300 uF and 3 of 1500 uF each. The board also has SMARTCARD and AFPANEL connectors, which have recently become common for the products of this company - the first for connecting a smart card reader, the second for connecting the previously described ASUS iPanel. There are a lot of jumpers on the board, and the most interesting are 4 blocks of switches - with their help you can set the base frequency of the processor, the multiplier and the voltage of the processor core. With the help of the fourth one, we can choose the type of processor - Athlon, Duron or Athlon 4 (Palomino). Most of the settings are hidden in the board's BIOS Setup.

It is based on the Medallion 6.00 version from AWARD - a typical choice for ASUSTeK. And in it, the soul can roam to the fullest - a huge selection of memory operation modes, AGP and PCI buses, manual distribution of interrupts among slots, etc. etc. All the features of the "Jumper Free" technology are also present - this was the name once given to the ability to change a large number of system characteristics from BIOS Setup "a in the ASUSTeK version. There are such features as: changing the processor core voltage (1.525 - 1.85 V, step 0.025 V), changing the multiplier (x5 - x13), changing the FSB frequency (100 - 227 MHz (!) in 1 MHz steps) Very good, but still - it has already become a standard, especially for ASUS.

The board didn't dishonor the ASUSTeK name, joined the flow coming from the company's conveyors and became another benchmark for other manufacturers.

Abit KG7-RAID

Abit enough for a long time rests on the laurels of victories obtained in the past - there is nothing new at the moment in terms of technology. However, this does not prevent the company from releasing good boards. And there is confirmation of these words - a board that has been in our laboratory. She became another prominent representative of the AMD-760 chipset. Internal testing revealed that the board is one of the leaders at the moment, which is why we dared to include it in this review as a representative of this chipset.

In addition to the motherboard itself, the box, which has already become a familiar dark tone design, contained two ATA66 / 100 cables, a cable for connecting a floppy drive, a bracket for the rear panel of the computer with two additional USB ports, a CD with software, a floppy disk with drivers for IDE RAID and a manual user made using the new design. It is interesting to note that in addition to describing the setup and installation of the board, it also contains a large number of advertisements for the company's multimedia products - acoustic systems, video cards, etc.

There is not a very large selection of programs on the CD - there is a DVD player - WinDVD, Hardware Doctor for system monitoring, Adobe Acrobat Reader and a set of programs from Buzzsoft - SoftCard Manager, SoftCopier, SoftPostCard, SoftBulkEmail. There is also a utility for low-level formatting of hard drives from AWARD. Sparsely, but even a little funny - the set is original.

The board itself does not look so austere - it has IDE RAID from HighPoint and a fairly large selection of settings. But first things first. The layout is not the best - the IDE RAID and floppy drive connectors are located right behind the PCI slots, which, firstly, can create some difficulties when installing long cards in these slots, and secondly, it obviously does not improve air circulation inside the case. The board does not have a built-in sound controller, which we are slowly starting to get used to. But there is an IDE RAID based on the HPT370A chip - as a result, we have two extra ATA100 channels, and we can also build RAID levels 0, 1, 0+1, thereby increasing either performance or increasing the reliability of the disk subsystem. The northbridge of the chipset has a heatsink with a fan - there is no particular need for it, but it can still increase the stability of the board during extraordinary overclocking of the system. For the same purposes, there are 6 capacitors of 4700 uF and 2 of 2200 uF each. There are not so few switches on the board, we can say that there are almost none of them - only one for clearing the contents of CMOS. All functions for setting up the board are in the BIOS - what else could you expect from a board from this manufacturer?

It is based on version 6.00 from AWARD and contains a ton of settings. Naturally, it has a SoftMenu III sub-item - it allows you to change the FSB frequency from 100 to 200 MHz in 1 MHz steps, you can manually set the ratio of the FSB, SDRAM and PCI frequencies - 4:4:1 or 3:3:1, change the voltages applied to the processor core, chipset and memory. As in all motherboards based on this chipset, there is a huge choice of memory and AGP bus settings - only some have it a little smaller, while others, as in our case, have more. There is also just a huge submenu with chipset settings - it doesn't even fit completely on the screen! Not unimportant, on this board, you can manually distribute interrupts to PCI slots.

Needless to say - this board provides enormous opportunities for a user who likes to tinker with the settings in order to increase the speed of work. But don't get carried away - this may affect the stability of the board, which works very decently at nominal frequencies.

Chaintech 7SID

A few years ago, the company was ahead of the rest - almost every day its boards overthrew the former champions from the peaks of performance, conquering new peaks of speed. But the rise was followed by a decline - and now the boards of this company are ranked among the middle peasants. Let's take a look at the representative of the SiS 735 chipset from this company - is it as good as its fairly distant predecessors?

The board that came to our laboratory was not a serial sample - and therefore it cannot be fully treated as a finished product, nevertheless, it has proven itself well in tests, which does honor to the company's engineers.

In the box with the inscription "VIA Chipset Inside" - the inevitable costs of the fact that the board has not yet been released for sale :), we saw an ATA66/100 cable, a cable for connecting a floppy drive, a xeroxed page with a description of the switches and the location of various connectors on the board, as well as CD-R disc containing the necessary software. But the disk, to our surprise, contained a normal installer instead of just a set of drivers - apparently, the appearance of the board on store shelves is not far off. On the company's website, you can find a mention of the "Value Pack 2000" disk, which will be bundled with the board and contain a set of free programs.

Now to the board itself. The baby (namely, it seemed to us like that against the background of other boards) carries the usual set of functional components. In addition to such a familiar thing as an integrated sound controller with a front panel connector with audio outputs, there is also a connector for connecting a smart card reader. In other respects, the board is ordinary - moderately convenient (if you can call boards of this form factor that way), carries 11 capacitors with a capacity of 2200 uF each to ensure stable operation of the board, and has a small set of switches. With one of them we set the frequency of the processor and memory - 100/100, 100/133 and 133/133 MHz, with the other we allow the ability to turn on the computer using a mouse or keyboard with PS / 2 connectors, and with the help of the third we can clear CMOS if necessary.

All other functions are hidden in the BIOS. It is based on version 6.00 from AWARD and has a fairly large number of settings in its arsenal - you can fine-tune memory timings, work on AGP bus performance, and manually distribute interrupts across PCI slots. Unfortunately, nothing can be done for a "free upgrade" - overclocking the processor, but since one of the first, not fully perfected BIOS was flashed on the board, in the future we can hope that this shortcoming will be corrected by the company's programmers.

"Small, but daring" - this is how I want to characterize this product. There are several drawbacks - the form factor, the lack of overclocking capabilities. But as a board for a high-performance home computer, it will certainly do.

Abit TH7-II RAID

The board is very similar to its sister - TH7, so their descriptions are similar. The company is known primarily for its image as a pioneer in overclocking processors and worthy equipping of its products with overclocker's hands-helping tools for this. This board is no exception - against the background of other boards from this review, it looks quite advantageous, excluding only the board from EPoX.

The serial copy that landed on our dissecting table was packaged in the usual dark-colored box for Abit. In the belly of this cardboard shell, we found a user manual in English, an ATA66 / 100 cable, a cable for connecting a floppy drive, a bracket for the back panel of the computer with a Game port, a blank for the back panel of the computer, two C-RIMM modules, an additional thermistor and a disk with software.

The set of programs on the disk has changed somewhat compared to the one that was on them not so long ago - there is a DVD player - WinDVD, Hardware Doctor (the purpose is clear from the name - system monitoring), Adobe Acrobat Reader and a whole set of programs from Buzzsoft - SoftCardManager , SoftCopier, SoftPostCard, SoftBulkEmail and Award's low-level hard drive formatter. Well, not enough - but the choice of the user is mainly decided not by these programs, but by the board itself.

As for the board itself, we can say the following - it is convenient. Perhaps only audio inputs can prevent you from enjoying it - but, by and large, this is a trifle. The board became the first for the new technology "Abit Engineered", already widely advertised by the company. Let me explain - it provides for the presence of two seven-segment LEDs on the board to display the progress of the computer booting, the presence of power and reset buttons on the board itself in addition to the usual ones, and applying solder strips called "overclocking stripes" to the back of the board. In addition, there are four LEDs next to the buttons on the board - they indicate the presence of power on the board, the presence of 5 V voltage, the operation of the hard drive, and the last of them lights up when the computer is restarted. Also, in addition to all of the above, to increase the stability of work during overclocking, 5 capacitors of 4700 uF and 10 of 2200 uF are installed - according to this indicator, the board is the clear favorite of our review. On the board, you can find places for non-soldered chips for network and FireWire adapters - therefore, a non-standard plug for the rear panel and a bracket with a game port are supplied with the board. The use of IDE RAID on the board is a good and fairly cheap way to increase the performance of the disk subsystem. There are several switches on the board - including those for enabling / disabling the SoftMenu function and clearing CMOS. All other settings are hidden in the BIOS.

It is based on version 6.00PG from AWARD and carries the traditional SoftMenu technology - with the help of which the processor is overclocked. Here you can change the memory frequency (300 or 400 MHz), adjust the FSB frequency in 1 MHz steps (from 90 to 156 MHz), change the processor multiplier and change the voltage applied to the processor core. Also in BIOS Setup, you can configure the operation of the AGP bus and manually distribute interrupts among PCI slots - the rest of the functions are standard for most motherboards and there is no need to describe them.

The pronounced overclocking color of this motherboard, however, does not prevent it from working stably at nominal and slightly higher frequencies. However, "Only Sberbank can give a full guarantee."

Well, now let's move on to the most "delicious" - to the actual tests of the systems described above for performance in popular programs. But first, let's describe the conditions under which this testing took place.

Hardware and software

• Processors
  • 1.5 GHz Intel Pentium 4, Socket 423
  • 1.7 GHz Intel Pentium 4, Socket 423
  • 2.0 GHz Intel Pentium 4, Socket 423
  • 2.0 GHz Intel Pentium 4, Socket 478
• motherboards
  • EPoX 4T2A (i850, Socket 423)
  • Intel D850MD Sample (i850, Socket 478)
  • Abit TH7-II RAID (i850, Socket 478)
  • VIA P4X266 Reference Board (VIA P4X266, Socket 423)
  • Supermicro P4SBA (i845, Socket 478)
  • ASUS A7V266 (VIA KT266, Socket 462)
  • Abit KG7-RAID (AMD-760, Socket 462)
  • Chaintech 7SID (SiS 735, Socket 462)
• Memory
  • 2 x 128 MB PC800 RDRAM RIMM, Samsung
  • 256 MB PC133 SDRAM DIMM, Tonicom ACTRAM, CL2
  • 256 MB PC2100 DDR SDRAM DIMM, Mushkin, CL2
• Other
  • Video card Leadtek Winfast GeForce3
  • Winchester Seagate Barracuda ATA III, ST340824A, 7200 rpm, 40Gb
  • ASUS CD-ROM 50x
• Software
  • Windows 2000 Professional SP1
  • NVIDIA Detonator v12.40 (Vsync=Off)
  • BapCo & MadOnion SYSmark 2001
  • idSoftware Quake III Arena v1.17 demo001.dm3
  • MadOnion 3D Mark 2001
  • Ziff&Davis Business Winstone 2001
  • Ziff&Davis Content Creation Winstone 2001
  • SPECViewPerf 6.1.2
  • 3DStudio MAX 3.1
  • expendable, demo version
  • Unreal Tournament v4.36

In addition, "Pentium 4 2.2 GHz" is also present in some diagrams. It seems that pushing each new processor to the limit of its capabilities is becoming a tradition for us :) Naturally, the results of this system cannot be taken as testing of the Pentium 4 2.2 GHz, which has not yet been officially released - after all, overclocking was carried out by increasing the FSB frequency. Also, as you will probably notice, in some diagrams the corresponding columns remained empty - alas, the overclocked processor simply did not pass some tests.

Why are we doing this? Well, let's approach the question a little from the other side - why not, in fact, do it, if possible? These results still give some idea of ​​the performance of the upcoming Pentium 4, albeit an approximate one. Well, we never feel sorry for a little extra work for you :) Performance tests

3D Mark 2001

We only show in the charts the low detail test results for the Cars, Dragothic, and Lobby gaming tests, as well as the overall score that the 3DMark 2001 test itself assigns. The fact is that the results for the high detail tests and the Nature gaming test are simply are not suitable for comparing the performance of processors and chipsets. Thus, the tests with high details show the same picture, only with a reduced "contrast", while the Nature test simply refuses to react to a change in the processor frequency or a change in the chipset - apparently, in this case, the performance of the video system becomes decisive.

By the way, we will continue to act in some diagrams in a similar "voluntaristic" method - just so as not to litter the material with unnecessary details. If one of the elements of the diagram illustrates a sequence of identical results demonstrated by all systems without exception, then this subtest is of no interest from the point of view of studying the performance of processors and chipsets, right?

As you can see from the first diagram, the highest performance in 3DMark 2001 is demonstrated by systems based on i850 + Pentium 4 2 GHz. The configuration based on ASUS A7V266 (VIA KT266) + Athlon 1.4 GHz came closest to their performance. VIA P4X266 is only slightly behind the i850, while the i845 is just behind. The results of an overclocked Pentium 4 2.2 GHz are a bit incorrect to discuss, but they still give some idea of ​​the further scalability of systems based on this processor - yes, scalability is certainly there, and quite good. We can assume (assume!) that with the increase in the frequency of "standard" models, the performance of Pentium 4 will continue to grow. But the i845 is in the very tail. Well, PC133 SDRAM makes itself known...

The chart with fps in gaming tests gives a rather funny picture: if in Cars and Lobby the general scheme of performance distribution between processors and chipsets approximately corresponds to the results in the chart with a total score (3DMarks), then the results in Dragothic cause a little bewilderment - yes, systems on, Pentium 4 still ahead, but their performance is practically the same! One of the possible options from our point of view may look like this: in fact, Dragothic, like the subtests we excluded, does not depend on the performance of the processor and the memory subsystem, but it does depend on the fact that certain commands are supported by this processor. Recall that neither SSE nor SSE2 are supported by regular Athlons.

Quake III

An absolute victory for all systems based on Pentium 4... except for the configuration based on i845. Intel loves the test in Quake III to demonstrate the advantages of the Pentium 4 very much, but in the case of the i845 even this favorite gave an unfortunate puncture. Why? It seems to us, first of all, because "the best" in Quake III is not the Pentium 4 itself, but the Pentium 4 in combination with the performance of the memory subsystem adequate to the processor frequency. But P4X266, in principle, is quite good! - not as much as i850 + RDRAM, of course, but still, still... Especially if you remember the prices for RDRAM and PC2100 DDR :)

Ziff-Davis Winstone 2001

But in this test, the fastest Athlon-based system shows approximately the same performance as Pentium 4 2 GHz, but... in combination with VIA P4X266 + PC2100 DDR SDRAM! And in the Content Creation Winstone test, Athlon even outperformed any "regular" Pentium 4. So what? Meanwhile, everything is clear. Athlon, as an example of a significantly improved "old" architecture, loses much less than Pentium 4 from unoptimized, full of "hard to predict" code transitions. In addition, Athlon-based systems are equipped with PC2100 DDR. Apparently, the applications included with Winstone 2001 contain most of this code.

The victory of P4X266 over i850 is a continuation of the same "theme": PC2100 DDR, like any SDRAM, has a much lower latency than Rambus DRAM, and in such conditions provides Pentium 4 with better working conditions. Again, Brookdale (i845) has rather poor results, although it also uses low-latency PC133 SDRAM. Still, low latency is not everything, bandwidth also means something.

SPEC ViewPerf

We will not make a big deal out of winning one system over another by meager percentages. Therefore, we can say that in AWadvs and MedMCAD systems based on Athlon 1.4 GHz and Pentium 4 2 GHz show approximately the same performance. The DDR system based on VIA P4X266 lags behind its counterparts on the i850 almost everywhere, but this lag cannot be called significant.

But in IBM Data Explorer (DX-06), the Pentium 4 takes the lead. Data Explorer is focused mainly on mathematical calculations, and specifically on intensive computational mathematics, so the frequency of the processor core is of decisive importance here. As for the i845, even Intel agrees with SPEC's tests on this chipset - well, there are no plans to create high-end systems at Brookdale that will run serious applications, the embodiment of which is the ViewPerf test!

3DStudio MAX

The Pentium 4 2 GHz shows a "normal" result (that is, the result is not far behind the Athlon 1.4 GHz) due to the higher frequency. This is confirmed by the absolute victory of the overclocked Pentium 4 2.2 GHz in this subtest. In general, we can say that this test embodies the "conceptual confrontation", one of the aspects of which we mentioned at the beginning of the article: a higher frequency versus a well-optimized processor core architecture for this type of task. In this case, the stock 2GHz Intel processor did not have enough frequency advantage to outperform the fastest AMD Athlon at the current time.

But the almost identical results shown by systems based on i850 and VIA P4X266, and the system based on i845, which is not far behind them, clearly demonstrate that the memory subsystem in 3DStudio MAX is not heavily loaded, the computing power of the CPU is of the greatest importance.

Expendable

A wonderful, even we would say "classic" example of an application where the Pentium 4 has no chance of winning even in the foreseeable future. Even an overclocked Pentium 4 2.2 GHz doesn't change the overall picture: all it managed to achieve was a little closer to the results of Athlon 1.4 GHz. Even without knowing anything about Expendable (and we know that in fact everything is exactly like this), it is extremely easy to assume the main reason for the loss of the Intel processor: "chaotic" code, which is extremely disliked by a processor with a long pipeline. Well, what else can be said? Yes, there are such applications. Either they will "die out" if the software vendors begin to take into account the existence of such a processor as the Pentium 4 with all its specific features of the internal architecture ... or they will continue to appear as before if this CPU does not become very popular. In this case, making predictions is a difficult and thankless task, so we confine ourselves to simply stating a fact.

Unreal Tournament

Considering the main characters of this review to be new Pentiums 4 2 GHz, in the comments on the Unreal Tournament results, we can simply confine ourselves to stating a fact: the performance of the most powerful Intel and AMD processors is approximately equal, with a slight advantage of the latter. And again (as well as practically everywhere earlier) Brookdale "weaves" in the tail. Low-end chipset for low-end systems, what can you do... :)

SYSmark 2001

This is perhaps the only test where the i845, which we recently "chided" in one of the subtests, even overtook all systems based on an AMD processor. There is nothing to say about Pentium 4 on i850 and P4X266 - they win by an impressive margin. SYSmark 2001 is a prime example of high application concentration" new wave", in which SSE / SSE2 support is present almost as required parameter, and this, as it is easy to see from the diagram, gives a quite definite (and easily predictable!) result. Moreover, the overclocked Pentium 4 2.2 GHz shows - further to the "old" processor architectures in similar applications it will get worse :)

Scalability

As we promised in the previous article about the VIA P4X266 chipset, here we present the results of system scalability:

As they say - "nothing particularly terrible." In general, scalability in most cases is slightly better for i850-based systems, but that's not the point. The main thing is that it is present in both cases. It cannot be argued that systems based on VIA P4X266 + Pentium 4 + PC2100 DDR are significantly inferior to i850 + Pentium 4 + PC800 RDRAM in this aspect. Well, it's nice that our (and not only our) fears were not justified. conclusions

Performance

The miracle didn't happen, but an event significant in many respects happened - now we can already say that the Pentium 4 leads in terms of performance in more tests. Yes it is. Increasing the frequency on the one hand, and the favor of many software vendors on the other, are having an effect. Of course, not everything is smooth yet, and it is too early to talk about an unconditional victory. Rather, we can talk about the restoration of parity. However, given that Intel is already on the threshold and is waiting for permission to enter :) 0.13-micron process, we can assume even higher rates of increase in the frequency of Pentium 4 in the very near future, so it looks like AMD will soon have to catch up. In theory, only the release of Athlon 1.5 GHz or even 1.7 GHz can help this company restore the situation with the comparative performance of the top processor models that took place earlier.

Price aspect - present

Alas, if the performance of the Pentium 4 is basically great (meaning the two-GHz model), then there are still big problems with the price of the entire computer assembly at the moment. So far, the only platform that is actually on the market for the Pentium 4 is the Intel i850 chipset in combination with Rambus DRAM. Even if we don't take into account the current position of the parent company of this memory, the cost of RIMM, even taking into account certain changes towards cheaper prices, still significantly exceeds both PC133 SDRAM and PC2100 DDR. Therefore, now (we emphasize - right now) the purchase of a Pentium 4-based system is still problematic for an economical user.

Price aspect - the future?

However, everything is not so bad. After all, there is VIA P4X266, although it "is" with certain problems caused by Intel's attitude towards this chipset. In addition, something similar (Pentium 4 + DDR SDRAM) is promised to us by ALi (Aladdin-P4, samples are already available, delivery is planned for October 2001) and SiS (SiS 645, announced on August 9, 2001, and, it seems, already available everyone who wants it). Also, soon the company will start actively promoting the Pentium 4 "Nothwood" (Socket 478, 0.13 micron, L2 cache 512 KB) on the market, which, judging by the announcements from Intel, will have a very attractive price at even higher frequencies than now, and most importantly , with twice the size of the second-level cache. So, in terms of price/performance ratio (assuming RDRAM is abandoned and the cost of the processor itself is reduced), the situation with Pentium 4-based systems may change very drastically in the very near future. What we wish for this processor - because it already demonstrates quite good performance, and the presence of two strong players on the market is always better for the end user than the undivided dominance of one.

On the one hand, time in the IT industry flies so fast that you don't have time to notice new products and technologies, but on the other hand, let's remember - how many years have we not seen a new core from Intel? Not the old one with alterations: here the FSB frequency was raised, there virtual multiprocessing was transferred from the server processor to the desktop one (in fact, they just allowed the latter to honestly tell what it has), but is it really completely new? If not designed from scratch, then at least not patched, but re-sewn according to the same patterns, but with different ruffles and in the latest fashion? But two whole years, it turns out! Even with a small ponytail. And all this time, hotheads were discussing their favorite topic: what will it be like, the new core? What they just didn’t predict - up to the complete anathema to the NetBurst architecture and the reign of solid Banias on the desktop platform. The truth (as often happens) turned out to be less fabulous: the new core turned out to be an honest and consistent successor to Northwood. Of course, with some architectural innovations, but the desire "to the ground, and then ..." is not traced in it. Therefore, purely emotionally, Prescott can be assessed in different ways: someone will praise Intel engineers for consistency and dedication, someone, on the contrary, will complain about the lack of fresh ideas. However, emotions are a personal matter for everyone, but we will turn to the facts. Theory

Major Core Changes (Prescott vs. Northwood)

To begin with, we offer you a small table that summarizes the most significant differences between the Prescott and Northwood cores in everything related to "iron" (or rather, silicon, and other "mineral constituents").

It remains only to add that the new core contains 125 million transistors (where is the poor Northwood with its 55 million!), And its area is 112 square meters. mm (slightly smaller than the area of ​​Northwood - 146/131 sq. mm, depending on the revision). Having made a simple arithmetic calculation, we see that by increasing the number of transistors by ~2.3 times, Intel engineers managed to reduce the core area due to the new technical process. True, not so significantly - "only" 1.3 (1.2) times.

As for the technology of “stressed” (some people prefer the term “stretched”) silicon, it is, if you explain it on your fingers, quite simple: in order to increase the distance between silicon atoms, it is placed on a substrate with a larger distance between atoms. As a result, in order to "sit down well", silicon atoms have to stretch along the proposed format. It looks something like this:

Well, to understand why it is easier for electrons to pass through strained silicon, this simple picture will help you:

As you can see, the geometric association in this case is quite appropriate: the path of the electron simply becomes shorter.

Well, now let's look at much more interesting differences: in the logic of the kernel. There are a lot of them too. However, to begin with, it would be useful to recall the main features of the NetBurst architecture as such. Especially since we haven't done it that often lately.

A little background

So, one of the main differences between the kernels developed within the NetBurst architecture, Intel itself considers a unique feature, which is expressed in the division of the x86 code decoding process into internal instructions executed by the kernel (uops), and the procedures for their execution. By the way, this approach gave rise to a lot of controversy regarding the correctness of the calculation of pipeline stages in Pentium 4: if we approach this processor from a classical point of view (pre-NetBurst era), then decoder stages should be included in the general list. Meanwhile, Intel's official data on the length of the Pentium 4 processor pipeline contains information only on the number of stages of the executing unit's pipeline, taking the decoder out of its scope. On the one hand, “sedition!” On the other hand, this objectively reflects the peculiarity of the architecture, so Intel is in its right: it also developed it. You can argue, of course, until you're blue in the face, but ... what, in fact, is the difference? The main thing is to understand the essence of the approach. You don't like that the decoder is excluded? Well, add its stages to the "official" ones - and you will get the required pipeline size according to the classical scheme, together with the decoder.

Thus, the main idea of ​​NetBurst is an asynchronous core, in which the instruction decoder operates independently of the Execution Unit. From Intel's point of view, it is essential to about higher than competitors, the frequency of the core, can only be achieved with an asynchronous model. if the model is synchronous, then the cost of synchronizing the decoder with the executing unit increases in proportion to the frequency. That is why instead of the usual L1 Instructions Cache, where the normal x86 code is stored, the NetBurst architecture uses the Execution Trace Cache, where the instructions are already stored in decoded form (uops). Trace is the sequence of uops.

Also, in a historical digression, I would like to finally dispel the myths associated with an overly simplified formulation, according to which the Pentium 4 ALU operates at “double frequency”. This is true...and not true. However, to begin with, let's take a look at the conditional block diagram of the Pentium 4 processor (now Prescott):

It is easy to see that the ALU consists of several parts: it has Load / Store, Complex Instructions, and Simple Instructions blocks. So: at double the speed (0.5 cycles per operation), only those instructions are processed that are supported by the Simple Instructions executing blocks. The ALU Complex Instructions block, which executes commands classified as complex, on the contrary, can spend up to four cycles to execute one instruction.

That, in fact, is all that I would like to recall regarding the internal structure of processors designed on the basis of the NetBurst architecture. Well, now let's move on to the innovations in the latest NetBurst core - Prescott.

Extending the length of the conveyor

This change can hardly be called an improvement - after all, it is well known that the longer the conveyor, the better about More overhead is caused by an error in the branch prediction mechanism, and, accordingly, the average speed of program execution decreases. However, apparently, Intel engineers could not find another way to increase the overclocking potential of the core. I had to resort to the unpopular, but proven. Outcome? The Prescott pipeline has been increased by 11 stages, respectively, their total number is 31. To be honest, we deliberately put this “good news” at the very beginning: in fact, the description of all subsequent innovations can be conditionally called “and now we will tell you how Intel engineers struggled with the consequences of a single change, so that it would not completely ruin productivity :).

Improvements in the branch prediction engine

Basically, fine tuning touched the transition prediction mechanism when working with cycles. So, if earlier, by default, back transitions were considered a cycle, now the length of the transition is analyzed, and based on it, the mechanism tries to predict whether it is a cycle or not. It was also found that for branches with certain types conditional jumps, regardless of their direction and distance, the use of the standard branch prediction mechanism is most often irrelevant - accordingly, now it is not used in these cases. However, in addition to theoretical research, Intel engineers did not disdain bare empiricism, i. simply by tracking the efficiency of the branch prediction mechanism on the example of specific algorithms. For this purpose, the number of errors in the branch prediction mechanism (mispredictions) was studied using examples from the SPECint_base2000 test, after which, in fact, changes were made to the algorithm in order to reduce them. The documentation provides the following data (the number of errors per 100 instructions):

Speed ​​up integer arithmetic and logic (ALU)

A specialized block for executing shift and rotate instructions has been added to the ALU, which now allows these operations to be performed on a “fast” (two-speed) ALU, unlike the Northwood core, where they were executed in the ALU Complex Instructions block and required b about more cycles. In addition, the operation of integer multiplication (integer multiply), previously executed in the FPU, has been accelerated. In the new kernel, a separate block is allocated for this.

There is also information about the presence of a number of minor improvements that allow you to increase the processing speed of FPU (and MMX) instructions. However, we will better check it in the practical part - when analyzing the test results.

Memory Subsystem

Of course, one of the main advantages of the new core is the increased size of the L1 data cache (by 2 times, up to 16 kilobytes) and the L2 cache (also by 2 times, up to 1 megabyte). However, there is another interesting feature: special additional logic has been introduced into the kernel that detects page faults in software prefetch instructions. Thanks to this innovation, software prefetch instructions now have the ability to not only prefetch data, but also prefetch page table entries, i.e., in other words, prefetch can not stop on a loaded page, but also update memory pages in DTLB. Those who understand the issue will surely notice from this example that Intel is closely following the feedback of programmers, even if it does not publicly repent of every negative factor that affects performance that they have discovered.

New instructions (SSE3)

Among other things, Prescott adds support for 13 new instructions. This set is named, according to the established tradition, SSE3. Among them there are commands for data conversion (x87 to integer), work with complex arithmetic, video encoding (though only one), new commands designed for processing graphic information(arrays of vertices), as well as two instructions designed to synchronize threads (obviously the consequences of the emergence of Hyper-Threading). However, we will soon publish a separate article about SSE3, so we will refrain from considering the capabilities of this set in this material so as not to spoil a serious and interesting topic with excessive popularization.

Well, now, perhaps, enough theory and specifications from us. Let's try, as one well-known joke said, "to take off together with all this" :). Testing

Stand configurations and software

test stand

• Processors:
  • AMD Athlon 64 3400+ (2200 MHz), Socket 754
  • Intel Pentium 4 3.2 GHz "Prescott" (FSB 800/HT), Socket 478
  • Intel Pentium 4 2.8A GHz "Prescott" (FSB 533/no HT), Socket 478
  • Intel Pentium 4 3.4 GHz "Northwood" (FSB 800/HT), Socket 478
  • Intel Pentium 4 3.2 GHz "Northwood" (FSB 800/HT), Socket 478
• Motherboards:
  • ABIT KV8-MAX3 (BIOS version 17) based on VIA K8T800 chipset
  • ASUS P4C800 Deluxe (BIOS version 1014) based on Intel 875P chipset
  • Albatron PX875P Pro (BIOS R1.00) based on Intel 875P chipset
• Memory:
  • 2x512 MB PC3200 DDR SDRAM DIMM TwinMOS (timings 2-2-2-5)
• Video card: Manli ATI Radeon 9800Pro 256 MB
• Hard drive: Western Digital WD360 (SATA), 10000 rpm

Pentium 4 2.8A GHz "Prescott"
The only Prescott with 533 MHz FSB
and without Hyper-Threading support

Pentium 4 3.4 GHz "Northwood"
Just another Northwood

System software and device drivers

• Windows XP Professional SP1
• DirectX 9.0b
• Intel Chipset Installation Utility 5.0.2.1003
• VIA Hyperion 4.51
• VIA SATA Driver 2.10a
• Silicon Image Driver 1.1.0.52
• ATI Catalyst 3.9

At the end of the description, I would like to explain the algorithm for selecting test participants. On the one hand, it would be wrong to completely exclude AMD processors from tests, because this platform is Intel's main competitor, both now and in the foreseeable future. On the other hand, to combine in one article a comparison of two generations of Pentium 4 with processors from another manufacturer would mean not really comparing either one or the other. That is why we decided to make a certain compromise in the first article devoted to Prescott: firstly, to completely exclude all kinds of "extreme" variants in the form of the Pentium 4 eXtreme Edition and Athlon 64 FX, and secondly, to take only one , but the fastest of regular AMD desktop processors: Athlon 64 3400+.

And even then, by and large, its results are given here only as an option. In this material, we are most interested in comparing the new Intel cores with the old. If anyone wants to get a quick look at how Prescott's performance compares to its closest competitor - well, it's in the charts. Comments? Perhaps they are just redundant. You will see for yourself. Knowing the performance of Prescott and Northwood operating at the same frequency, and how the performance of Northwood and top AMD processors compare (and we have already covered this issue more than once), you know quite enough to draw all other conclusions on your own.

In addition, I would like to clarify the presence of two bars on the diagrams for Prescott 3.2 GHz. It's just that we decided to ... play it safe. Everyone knows that with the release of a processor on a different core, turmoil immediately begins among motherboard manufacturers with updating the BIOS, all kinds of microcode updates, and other "hardware-oriented" software. It seemed logical to us to use such a resource of our test lab as "officially Prescott-ready" motherboards to the fullest extent possible in order to protect ourselves from the possible consequences of incorrect operation of a particular model. However, as you will see below, the fears turned out to be groundless: in most cases, the new processor behaved in exactly the same way on both boards.

All Specs Prescott 2.8A GHz Program
CPU-Z defines quite correctly:
both the presence of SSE3 and the 533 MHz bus

Of course, she was not mistaken in the case of
Prescott 3.2E GHz

Low-level benchmarks in CPU RightMark

To begin with, we decided to test the functioning of the new core in two modes - traditionally the best for Pentium 4 processors and the worst: SSE/SSE2 and MMX/FPU. Let's start with the computational block (Math Solving).

The results are disappointing. The new core is slower than the old one, moreover, in MMX/FPU mode its lag is even greater than when using SSE/SSE2. We draw the first conclusion: if something was “twisted” in the FPU, then other commands are clearly used in CPU RightMark. Well, what about rendering?

First, let's consider the options for the rendering module to work in single-threaded and dual-threaded modes with maximum performance (SSE/SSE2). The picture is quite interesting: if one stream is used, the advantage of Prescott is minimal, and Northwood, which has a higher frequency, easily overtakes it. However, as soon as we use Hyper-Threading, Prescott immediately takes a sharp lead, and so much so that it overtakes all other participants. It seems that some work on the kernel in terms of improving the processing of parallel threads was carried out, and it consisted not only in expanding the instruction set. Now let's see how the same processors behave in MMX/FPU mode.

Absolutely the same picture. Moreover, if we compare it with the previous one, it is clearly seen that the thoroughness of the analysis justified itself: if, for example, we limited ourselves to considering the best (two-thread) result, we could erroneously conclude that the Prescott core is faster in terms of executing instructions, and even in MMX/FPU mode. Now it is clearly seen that the performance has increased solely due to the optimization of the use of virtual CPU resources.

Tests in real applications

Before we start looking at test results in real applications, let's make a small introductory clarification. The fact is that the Pentium 4 processor on the Prescott core with a frequency of 3.4 GHz, unfortunately, is still not available to us, so what you see on the diagrams called "Virtual" Prescott 3.4 GHz is nothing more than an approximation of the Prescott 3.2 GHz results, calculated from ideal conditions of performance growth proportional to frequency. One might notice that this is too clumsy an approach. Say, it would be much more correct, for example, to overclock the existing Prescott 3.2 GHz by setting a higher FSB frequency, or at least build an approximation curve over three points: Prescott 2.8 GHz -> 3.0 GHz -> 3.2 GHz. Of course, that would be more correct. However, "for every sage, it's quite simple", and ... just pay attention to what corrections the presence of even the "ideal" Prescott 3.4 GHz on the diagrams makes in the overall picture (and the real one will be either the same or slower - there is no third). At the risk of incurring the urge to reveal the secret prematurely, let's say right away: yes, practically none. Where the Prescott core wins, you can see it there. And where he loses, even the idealized 3.4 GHz does not help him

Working with graphics

The most predictable results are for Northwood 3.4 GHz (slightly better than Northwood 3.2 GHz) and Prescott 2.8 GHz (lack of Hyper-Threading support immediately made it an outsider). The Prescott 3.2 GHz tries to be at least on par with the single frequency Northwood, but it fails even that. Well, our "virtual Prescott 3.4 GHz", in turn, could not overtake the real Northwood 3.4 GHz - which is also natural. On the other hand, you can see that all processors except Prescott 2.8 GHz are almost equal. This is unlikely to be an argument for upgrading to Prescott, but at least it will not become a significant argument against buying it for those who are thinking about purchasing a new system.

In Lightwave, the situation is similar, only Prescott lags even further behind. Here it would be appropriate to recall that Lightwave (judging by the comparison of the results of the 6th branch with the 7th), was sharpened for the Pentium 4 very carefully and scrupulously. It can be assumed that this is why it turned out to be so sensitive to the slightest architectural changes in the core. We also note that the Athlon 64 3400+ tested by us for the first time in this program demonstrates quite a decent result, if not the best.

For Photoshop in modern processor architectures, apparently, the most important parameter is the cache size. We have already repeatedly drawn attention to the fact that this program is very "cash-loving", and the results of Prescott confirm this.

Media encoding

In general, since we are testing a new (or significantly modified, if you prefer) architecture, any application can become a small discovery for us. In fact, now the number of even more important than quality, because we just need to collect as much data as possible about how old (not yet optimized for Prescott) programs behave with the new processor core. Here is the same LAME: it turns out that Prescott is a new processor for it in all respects - the results do not fit at all with what we previously knew about Northwood. In fact, they got worse. Well, it happens. We continue to collect

Ogg Encoder demonstrates an almost identical picture: Prescott significantly loses to all other processors without exception, despite the doubled L1 data cache and L2. It remains to be assumed that the increase in the length of the pipeline is to blame with the volume of Trace Cache remaining unchanged.

Even the DivX codec, which gravitates toward the NetBurst architecture, did not like the new core. Not that it was very strong, but still he did not like it. However, there is some hope for SSE3 - DivX developers simply love various optimizations (at least, judging by the announcements), so there is a very good chance that the one and only instruction designed to speed up video encoding will find its place in a future release of this codec . However, this is all in the future, but for now - alas

But we again do not give the results of XviD due to the completely unimaginable "trick", which once again was thrown out by this dearly beloved program. The fact is that the increase in productivity of Prescott in relation to Northwood in it amounted to 232% ! Such tests, sorry, we simply refuse to use. It seems that their results can depend on anything at all

Well, here is the first victory. However, returning to the topic of preferences for various software, you can see that Windows Media Video 9 supports Hyper-Threading quite well, and low-level test data showed that the efficiency of using virtual CPUs in the case of a new core increases. This appears to be the first positive result achieved through a qualitative rather than quantitative change at Prescott. In all previous cases, he "left" solely due to the large amount of cache

Very, very interesting result. Mainconcept MPEG Encoder, which we blamed for the “clumsy” work with Hyper-Threading when encoding to MPEG1 format, works quite adequately with virtual processors, if they are emulated by Prescott and not by Northwood! It’s time to even think: maybe the programmers are not to blame, just a “plug” was in the processor core, which incorrectly parallelized threads? It is quite possible, at least, looking at the results of Prescott, you understand that this assumption has the right to life. On the other hand, Prescott 2.8A GHz showed itself quite well, about Hyper-Threading and never heard of it. Funny situation. Perhaps we are on the verge of an interesting discovery: it suggests that the whole “optimization of Hyper-Threading in Prescott” comes down to just the fact that this technology in Northwood did not have enough cache to unfold in full force!

And again, we can be glad for the new core: in Mainconcept MPEG Encoder, not only did the "glitch" with MPEG1 encoding disappear, but the conversion to MPEG2 began to work much faster. Bearing in mind the results of previous tests, we can almost unequivocally state that the main hero of the occasion is the improved work of Hyper-Threading (and do not forget about what could make it better - if our assumptions are correct). What is most interesting - they did not even need special teams to manage flows from the SSE3 set, the processor itself figured it out perfectly (support for SSE3 in this version of the encoder is not to be expected - it was released quite a long time ago).

But Canopus ProCoder just noticed almost nothing. In principle, there is a slight difference in performance, and it is even in favor of Prescott. But, in fact, this is a penny, a trifle. Considering ProCoder's "cache-loving" nature, one can even say this: all the large cache, apparently, went to compensate for other shortcomings of the new core. He simply pulled Prescott to the same height as Northwood, but, alas, no more.

Archiving

As usual, we've tested 7-Zip both with and without multithreading enabled. The expected effect was not achieved in this program: it is not noticeable that multithreading on Prescott gave much more effect than on Northwood. And in general, there is no particular difference between the old and the new kernel. It seems that we are seeing the effect mentioned above: all that Prescott's quantitative indicators (L1 Data and L2 cache sizes) could do was compensate for its lengthened pipeline.

By the way: one of the few tests where the difference between the boards is at least somehow visible. Otherwise, the picture is the same: Prescott and Northwood of the same frequency go side by side, practically the same in speed. Pessimists will say: “bad”, optimists: “it could be worse” :). We just keep silent

Games

The picture in all three games is similar, so there is no need to undersign especially: Prescott is still slower. True, not much.

Summarizing the results

Well, if we draw any conclusions based on the tests presented in the article, then the situation is as follows: the Prescott core is generally slower than Northwood. Sometimes this can be compensated for about more cache, bringing performance up to the level of the old core. Well, if the program is especially sensitive to L2 volume, Prescott can even win. In addition, the efficiency of Hyper-Threading has improved slightly (but it seems that the reason again lies in the increase in the size of the L2 cache). Accordingly, if a program can use both strengths of the new core - a large cache and virtual multiprocessing - then the gain is tangible. In general, the performance of Prescott is about the same as that of Northwood, and with respect to old, unoptimized software, even lower. The expected revolution, alas, did not happen. On the other hand ... was there a boy? But more on that below.

As for the Prescott 2.8A GHz with a 533MHz system bus and no Hyper-Threading support, everything is very clear here. Firstly, for Intel, this is just a very good way to make at least something from those instances that simply did not work in the “real Prescott” mode. A sort of “Celeron among Prescotts” (although it will, apparently, based on this core and the official Celeron). Secondly, the absence of Hyper-Threading most likely indicates Intel's fundamental unwillingness to see HT on an outdated, low-speed bus. Indeed, the only representative of 533 MHz FSB + HT is the first processor supporting this technology - Pentium 4 3.06 GHz. And even then, for a completely understandable reason that excused him: at that time there was no CPU with an 800-MHz bus.

So, may the Intel engineers forgive us this liberties, Pentium 4 2.8A GHz is “not like a Prescott”. And just a relatively inexpensive one (you can't produce it for others - no one will buy it ...), but a high-frequency Pentium 4. And it doesn't matter what core it is made on, that's not the point. To be honest, there was a temptation not to include him in this material at all, but then we decided to do the opposite: let him “light up” once, and more to this hour. at do not return to the bottom processor. From a simple comparison of Prescott and Northwood single-frequency cores, it is clear that without Hyper-Threading Prescott 2.8 GHz it will not be able to compete even with Pentium 4 2.8C (800 MHz FSB + HT) in terms of average performance indicators. Versions

Yes, namely “versions”, not “conclusions”. This material turned out to be too ambiguous. It would be easier to confine ourselves to the analysis of diagrams and draw a conclusion that suggests itself, lying on the surface: “if the new is not faster (or even slower) than the old, then it is worse.” Write off, so to speak, in expense. However, the simplest answer is not always the correct one. Therefore, we decided to touch on the analytics and consider the exit of Prescott in the historical market perspective. It turned out that the answers to the question "what is the point for Intel to release the Pentium 4 on the Prescott core?" in fact, several, and each of them can be logically argued.

Version One or Big Mistake

Why not? Once upon a time there was an Intel company, and it came up with an idea: to make a processor core focused not on maximum efficiency (if we consider efficiency as a ratio of performance to frequency), but on easy scalability. Say, if our 2000 MHz lose 1000 MHz from a competitor - it does not matter, we will catch up with the frequency to 4 GHz and leave everyone behind. By the way, from a purely engineering point of view, this is quite an adequate solution. Isn't it all the same? The user (literate) is still interested not in megahertz, but in performance, what difference does it make to him, due to what it is achieved? The main thing is that the scalability is exactly what it was supposed to achieve. And now, it turns out that big problems began with scalability. We caught up to 3.4 GHz, stopped ... and had to come up with a new core, whose efficiency is even lower ... and it is not known at what pace its frequency will increase ... and so on. Recall that this is a version. Let us consider it more closely in comparison with real facts.

A fact that testifies in favor of this version is the increase in the frequency of Pentium 4 over the past 2003. Still, 200 MHz, and even in relation to such a "frequency-loving" architecture as NetBurst, is clearly not enough. However… as is well known, it is not good practice to consider any fact in isolation from others. Was there any point in actively increasing the frequency of Pentium 4 last year? Apparently not... The main competitor solved other issues - it has a new architecture, a new core, it needs to organize mass production of processors based on this core, provide them with appropriate binding in the form of chipsets, motherboards, software, after all! Therefore, one of the answers to the question "why did the frequency (and performance) of the Pentium 4 practically increase in 2003" sounds simple: there was no particular point in increasing it. Neither catch up nor overtake - like there is no one. So, you don't have to be too hasty.

Unfortunately, we still cannot get an answer to the main question: how will the new core be “chasing”? So far, judging by external signs, there are no facts confirming the good scalability of Prescott. However, as well as those who refute it. 3.4GHz versions of both Prescott and Northwood have been announced. Northwood 3.4 GHz will probably be the last processor on this core (although there is no official confirmation of this assumption). And the fact that Prescott started at 3.4 GHz and not 3.8 or 4.0 is also easy to explain: why jump over steps? To sum it up: version " Big mistake”, in principle, has the right to exist. But if the frequency (more precisely, the performance) of Prescott grows rapidly, this will definitely confirm its failure.

Version Two or Transition Core

It's not a secret for anyone that sometimes a manufacturer needs to release a certain device that is quite ordinary in itself (in another situation, it does not deserve the title of a release product at all). But the fact of the matter is that the release this device necessary to promote others to the market, announced simultaneously with it or a little later. Such was the Pentium 4 Willamette, hardly worthy of the title of "good and fast processor", but clearly indicating the fact that one of the largest players in the processor market has switched to a new core, and at the end of its existence, it replaced the "intermediate" Socket 423 with a "long-playing" one. Socket 478. What if a similar role is in store for Prescott?

Everyone already knows that with the release of Grantsdale-P, we will see another processor socket for Pentium 4 (Socket T / Socket 775 / LGA775), and at first it will be CPUs based on the Prescott core that will be installed in it. Only later Pentium 4 "Tejas" will gradually replace them. And here it is quite logical to ask the question: how fast will this replacement take place? Since we are just putting forward versions anyway, we will not limit our imagination, and assume that Intel wants to speed up this process as much as possible. With using what? Most likely - leaving Socket 478 to rest peacefully in the bottom lines on performance charts, and making Socket 775 a symbol of an updated, powerful and speed platform for Pentium 4. Then everything becomes clear: Prescott is needed in order to have a processor on the market that can work both in motherboards with Socket 478 and with the new Socket 775. Tejas, if our assumptions are correct, will be installed only in Socket 775, and thus become a gravedigger for both Prescott and the outdated Socket 478 platform. Logical? We think so. In this case, the following assumption also looks plausible: Prescott’s life is destined for a very short time.

Version three or "Who will come to us with a sword ..."

It's no secret that the rivalry between the two main competitors, Intel and AMD, has almost always been based on the opposition of two main arguments. Intel: "our processors are the fastest!", AMD: "but our better ratio price and performance! The rivalry is old, the arguments too. Moreover, they have not changed even with the release of AMD processors based on K7/K8 cores, despite the fact that the latter have much better performance than K6. Previously, Intel did not make exceptions to its main rule: to sell their CPUs with performance similar to competitor processors for a little more. The market is very simple in places, so the reason for this behavior is understandable: if they are already being bought, then why lower the price? Again: although Intel had to participate in price wars, AMD always unleashed them, this has already become a tradition. The third version is based on the obvious assumption: what if this time Intel decided to be more aggressive than usual and unleash a price war first?

The list of advantages of the new Prescott core includes not only novelty, cache volumes, and potentially good (though not yet confirmed) scalability, but also ... the price! This is a relatively cheap core to manufacture: if, using 90-nanometer technology, the yield of suitable chips is at least the same as that of Northwood, then without losing any absolute profit, Intel will be able to sell its processors for a much lower price. Recall one obvious dependence: such a CPU characteristic as "price / performance ratio" can be improved not only by increasing performance, but also by reducing the price. In fact, no one bothers even lowering the speed (!) - the main thing is that the price drops even more :). Judging by the unofficial price announcements for the Pentium 4 Prescott appearing on the Web, they will cost much less than the Pentium 4 Northwood. Thus, we can assume that Intel decided to implement a kind of “flanking”: while the main competitor, in the old fashioned way, is chasing and chasing performance, it will be hit in the middle-end sector, where users carefully analyze just such indicator as price / performance.

Version Four or Secret Weapon

Here we should make a small lyrical-historical digression for those who "in those days" did not actively monitor various small nuances in the processor sector. So, for example, we can recall that immediately after the appearance of the first processors with Hyper-Threading support (and they were not the Pentium 4 "Northwood" + HT at all, but the Xeon "Prestonia"), many asked the question: "if the cores of Prestonia and Northwood are so similar that they practically do not differ in their main characteristics, but Prestonia has Hyper-Threading support, while Northwood does not - is it not logical to assume that Northwood also has it, just artificially blocked? Subsequently, this assumption was indirectly confirmed by the announcement of the Pentium 4 3.06 GHz based on the same Northwood core, but with Hyper-Threading. Moreover, the most daring put forward a completely seditious idea: Hyper-Threading was even in Willamette!

And now let's remember: what we have recently known about the new technological initiatives of Intel. Two names immediately pop up: "La Grande" and "Vanderpool". The first is the technology of hardware protection of applications from outside interference, which can be briefly described by the words "make it so that one software cannot interfere with the operation of another." However, you can read about La Grande on our website. There is less information about Vanderpool, but based on the scraps available today, we can conclude that it is a variation on the theme of full PC virtualization, including all hardware resources without exception. Thus (the simplest, but also the most spectacular example), two computers can work in parallel on one computer. Operating Systems, and one of them can even be rebooted - but this will not affect the work of the other at all.

So: there are very strong suspicions that both La Grande and Vanderpool have already been implemented in the Prescott kernel, but (as it was before with Hyper-Threading) they have not yet been activated. If this assumption is true, then much about the core itself becomes clear. In particular, why it is so big, why it took so long to develop, but, despite this, it does not outperform the previous one in speed. Based on the “Secret Weapon” hypothesis, it can be assumed that the main resources of the development team were directed not at all to achieving speed, but to debugging new functions. Partially, this version echoes the second - one way or another, but we are dealing with a transitional core. Accordingly, it does not have to be perfect at all, because that is not its main purpose. By the way, the second and fourth versions are also successfully complemented by the third: the low price in this case is exactly the candy that will sweeten the “transitivity” pill for the end user.

Summing up

We called this article "half a step forward" for a reason. Prescott turned out to be more complex and ambiguous than the expected "Northwood with more cache and higher frequency" (as many perceived it). Of course, you can blame the manufacturer for the fact that the speed increase is close to zero on average (and in some places even negative), for another leapfrog with support for processors based on the new core by motherboards... And, by the way, it is quite fair to do this. These, after all, are not our problems - but meanwhile, it is we who will face them. Therefore, we simply put a “bold ellipsis” at the end of the article. The freeze-frame shows only the beginning of the step: a leg hovering in the air, or, if you like, a liner taking off. What's next for us? Whether the “landing” (Tejas?..) will be favorable, so far one can only guess.

For more than 5 years, many Pentium 4 cores and models based on them have been released. Moreover, with the release of a new model, either new letter, or some other numbers, and sometimes both; all this significantly confuses the identification of a particular model.

The Pentium 4 processor is built on a completely new architecture - NetBurst. Below are some distinctive features the original NetBurst architecture (some of which have since been changed).

• . The length of the pipeline was increased to 20 steps, that is, it took the processor 20 cycles to complete one instruction. This step made it much easier to increase the clock frequency, in addition, in the long term it made it possible to significantly increase performance, but the performance per 1 MHz was less than that of previous processors. This partly explains the low performance of the Pentium 4, which operates at low frequencies. Also, as a result of this innovation, the waiting time has also increased.
• Transition (branch) prediction module. To compensate for the disadvantages of using a long pipeline, Intel engineers improved the branch prediction scheme, as a result, the correctness of the branch was predicted with a probability of up to 95%.
• System bus. The Pentium 4 uses a completely new 128-bit processor with two 64-bit lanes. The frequency of the new bus () is 100 MHz (for the latest, then, Pentium III models, it was 133 MHz), however, due to the transmission of 4 packets per 1 cycle simultaneously (QPB - Quad Pumped Bus), the effective bus frequency was 400 MHz, and the bandwidth the bus capacity was 3200 Mb/s.
• Arithmetic Logic Unit (or ALU). The ALU processes integer instructions. In the new processor, the ALU runs at twice the core frequency (the Pentium 4 has a 1.5GHz ALU running at 3GHz by using both signal edges). Thus, some instructions are executed in half a cycle. The Pentium 4 uses two ALUs.
• first level (L1). As before, the L1 cache is divided into two parts: for instructions and for data. The cache now stores decoded commands and arranges them in the order they are executed (Trace Cache technology), which increases performance.
• Math(). The math coprocessor contains two modules for floating point operations. But only one module performs real computational work - these are addition (FADD) and multiplication (FMUL) operations, the second module performs exchange operations between and memory (FSTORE). For a 1.4 GHz Pentium 4 processor, the coprocessor delivers 1.4 times the performance. For example, the processors use a coprocessor consisting of three modules (one for FSTORE operations, two others for FADD and FMUL operations) and provides 2 GFLOPS performance (for a 1 GHz Athlon processor).
• SIMD extensions. A new set of SIMD extensions (SSE2) was added to the Pentium 4 processor, which added 144 new instructions (68 integer instructions and 76 floating point instructions).

In general, the architecture was aimed to work on high frequencies, where a long pipeline could work in full force.

Willamette

For the first time, this core "lit up" in the Intel roadmap () back in 1998. It was assumed that it should come to replace and conquer the frequency of 1 GHz. But processors based on this core were announced only in 2000 as Pentium 4. Released years, the processors were installed in the Socket 423 socket and were produced in the FC-PGA2 package. Processors for Socket 423 were not popular, as Intel immediately declared that this socket was a transitional socket, and besides, Pentium 4-based systems were very expensive (the processors themselves at the time of the announcement cost $644 and $819 for Pentium 4 1.4 and 1.5 GHz, respectively). Since the processor was manufactured using a 180 nm process technology, only 256 KB of L2 cache could be placed on the chip. Most experts regarded the 1.4 and 1.5 GHz versions as intermediate ones - the Athlon processor was gaining more and more popularity, and was faster than the Pentium III, and further improvement of the Pentium III architecture was not yet possible at that time. Lose your share Intel market was not going to, so she released these processors (the “raw” production technology did not allow the release of faster models at that time). Despite the unpopularity of the 1.4 and 1.5 GHz versions, Intel of the year announced a 1.3 GHz version of the Pentium 4, which cost $409. In various benchmarks, these processors outperformed both Pentium IIIs and Athlons (and in some cases even Athlons) running at lower frequencies. However, already in April 2001, Pentium 4 with a frequency of 1.7 GHz was released, and in August of this year a 2 GHz version was released, as well as "new-old" processors for Socket 478, which lasted more than 2 years, in the same month there is a new chipset from Intel (i845). The new chipset now supported PC133 SDRAM memory, which made it possible to significantly reduce the prices of systems based on Intel Pentium 4, but the use of this type of memory slightly reduced the system performance (sometimes quite significantly). Intel actively promoted this processor to increase sales - its advertising could be seen both on TV and in newspapers / magazines. Pentium 4 sales increased, the processor began to enjoy more and more popularity. Soon, many chipsets manufacturers introduced their chipsets for the Pentium 4 with memory support, and at the beginning of the year, Intel releases its chipsets with support for this type of memory. The processor begins to replace the Pentium III, and in terms of performance it is actually on par with the Athlon. Intel, which held the palm for 16 years, and then quickly lost it, is now starting to win back its own again. And the beginning of problems with the lack of production capacity at AMD and the release of the Pentium 4 on the Northwood core secured Intel's leading position, though not for long.

Northwood

The first processors on this core were announced years ago. The core is not much different from its predecessor, except for the use of a more advanced tech process - 130 nm, which made it possible to place 512Kb of L2 cache on a chip and reduce processor heat dissipation. The transition to a new process technology made it possible to further increase the clock frequency (up to 3.4 GHz). In order to distinguish processors based on the Northwood core from similar models based on the Willamette core, it was decided to add the letter "A" to the end of the names of new processors (for example, the Pentium 4 2.0A is based on the Northwood core).

Prescott

Mobile Pentium 4

The first versions of Pentium 4 for notebooks and laptops were announced a year ago, they were built on the Northwood core and were called Mobile Pentium 4-M. These processors differed from desktop versions in their low supply voltage (1.2-1.3 V) and support for . The system bus frequency for all processors was 400 MHz. Models with frequencies of 1.4 were released; 1.5; 1.6; 1.7; 1.8; 1.9; 2.0; 2.2; 2.4; 2.5; 2.6, the TDP of the latest model is 35W.

Prescott 2M

The first rumors about the new Prescott 2 core appeared in early 2005. It was supposed to have 2 MB L2 cache and FSB frequency equal to 266 MHz (1066 MHz effective frequency). Processors based on this core have been announced. This core differs from the Prescott core only in the presence of 2 MB of L2 cache. The new processors also received a new marking: 6x0. February 21, 2005 were announced Pentium models 4 630, 640, 650, 660 with frequencies 3.0; 3.2; 3.4; 3.6 GHz, later the model 670 was introduced, operating at a frequency of 3.8 GHz.

Cedar Mill

The announcement of processors built on the new core is scheduled for the second half of January. The Cedar Mill kernel is a single-core modification of the kernel known by the code name . Cedar Mill is manufactured using the latest 65nm technology. In fact, the core is a Prescott 2M core, the processor series has not even changed, processors without Vanderpool virtualization technology support are labeled 6x1, with Vanderpool support they are labeled 6x3. The processors will initially be clocked at 3.0 GHz to 3.8 GHz. It is recommended to read more about the reason for the release of Cedar Mill here.

The Cedar Mill core is the last in the Pentium 4 line. The next generations of processors, in particular Conroe, will be promoted under a new brand, the name of which has not yet been announced.

Tejas, Jayhawk and others

Intel pinned on the NetBurst architecture great expectations. In 2001-2003, Intel roadmaps featured cores such as Tejas, which was supposed to use a 1066 MHz bus and operate at frequencies from 4.4 to 9.2 GHz and should have gone on sale in the second half of 2004 and be called Pentium 6. Nehalem, this processor was supposed to use a 1200 MHz system bus and operate at frequencies above 10 GHz, and was supposed to go on sale in 2005. Jayhawk, a Xeon-series processor that was supposed to have an L1 cache for 24 KB of data and 16,000 µops. However, all these processors were canceled in 2004.

Intel intended to reach 10 GHz with processors based on the NetBurst architecture, but before reaching 4 GHz, this architecture faced thermal problems that were still (and probably never) unsolvable. This problem prompted Intel to develop a new architecture and to close all projects for the development of kernels based on the NetBurst architecture.

Looking back, the Pentium 4 leaves an ambivalent impression. On the one hand, it was one of the most popular processors, its promotion in, and, as a result, its huge popularity among the people, allowed Intel to occupy a large part of the market for a long time. On the other hand, the Pentium 4 did not have the most successful architecture. It never consolidated its leading position in terms of performance, in terms of TDP (heat dissipation) it almost always lost to competitive AMD Athlon processors, however, as well as in terms of cost. And the Pentium III architecture, which Intel once considered less promising than NetBurst, has reappeared in .

Specifications of various cores

Data related to all models

• Bit depth : 32
• External bus width: 128

Willamette

• Date of announcement of the first model: November 20, 2000
• Clock frequencies (GHz): 1.3; 1.4; 1.5; 1.6; 1.7; 1.8; 1.9; 2.0
• Effective System Bus (FSB) Frequency (MHz): 400
• L2 cache size (KB): 256
• Supply voltage: 1.7V or 1.75V
• Number of transistors (million): 42
• Crystal area (sq. mm): 217
• Maximum TDP (calculated heat dissipation): 75.3 W
• Process technology (nm): 180
• Connector: Socket 423, later Socket 478
• Package: 423-pin FC-PGA2 or 478-pin mPGA
• Supported technologies: IA32, SSE2

Northwood

• Date of the announcement of the first model: August 7, 2001
  • Processors with FSB frequency equal to 400 MHz: 1.6; 1.8; 2.0; 2.2; 2.4; 2.5; 2.6; 2.8
  • Processors with FSB frequency equal to 533 MHz: 2.26; 2.4; 2.53; 2.67; 2.8; 3.06
  • Processors with FSB frequency equal to 800 MHz: 2.4; 2.6; 2.8; 3.0; 3.2; 3.4
• Effective system bus frequency (FSB) (MHz): 400, 533, 800
• L1 cache size: 8KB (for data) + 12k operations
• L2 cache size (KB): 512
• Supply voltage: 1.475-1.55 (depending on the model)
• Number of transistors (millions): 55
• Crystal area (sq. mm): 146, later 131
• Maximum TDP (calculated heat dissipation): 89 W
• Process technology (nm): 130
• Connector: Socket 478
• Package: 478-pin mPGA
• Supported technologies: IA32, MMX, SSE, SSE2, HT (not all models)

The series of Intel Pentium 4 processors is the most successful when compared with other modifications of the developer, since over the years of work the right to its existence has been proven. In the presented article, you can find out how these processors differ, get acquainted with their technical characteristics.

Generations of processors are constantly changing one after another due to the race of developers for frequencies. Of course, new technologies also appeared, but they were not in the foreground. Thus, not only users, but also manufacturers were well aware that one day the effective frequency of the processor would be reached. This happened after the release of the fourth generation of Intel Pentium.

The operating frequency of a single core at 4 GHz has become the limit. This happened for the reason that the crystal needed a lot of electricity to work. Thus, the dissipated power in the form of enormous heat dissipation called into question the functioning of the entire system. Further modifications of Intel processors and analogues of rivals began to be produced in the region of 4 GHz. We should also mention technologies that used multiple cores, as well as the introduction of special instructions that can optimize data processing work.

The first pancake is lumpy

In the area of high technology monopoly in the market did not lead to anything good. This is confirmed by numerous electronics manufacturers who have been able to verify this on their own experience. But Intel and Rambus decided to make good money. As a result, a promising joint product was released. Thus, the first Intel Pentium 4 processor, running on Socket 423, saw the light and communicated with Rambus RAM at a fairly high speed. As a result, many users wanted to become owners of this fast computer. True, these two companies did not become monopolists in the market.

This was hindered by the opening of the dual-channel memory mode. The results of the testing showed a high performance increase. Thus, all developers of computer components immediately became interested in the new technology. As for the first Pentium 4 processor, it and socket 423 became history, since the manufacturer did not provide the platform with the possibility of upgrading. To date, components for this platform are in demand. It turns out that several state-owned enterprises have purchased ultra-fast computers. Thus, the replacement of components is somewhat cheaper than a full upgrade.

A step in the right direction

Most owners of personal computers who play games and prefer to work with documentation and watch multimedia content have Intel Pentium 4 (Socket 478) installed. Many tests that have been carried out by professionals and enthusiasts indicate that the power of this platform is quite enough to perform all the tasks assigned to the average user. Such a platform uses two kernel modifications:

Willamette;
Prescott.

Their characteristics indicate that the differences between the two processors are small. The latest modification provides support for 13 new instructions designed to optimize data, which were briefly called SSE3. The frequency range of the crystals is in the range of 1.4-3.4 GHz, which fully satisfies the requirements of the market. The developer took a risk and introduced an additional branch of processors for socket 478. These devices were supposed to attract the attention of game lovers and overlockers. New series became known as the Intel Pentium 4 CPU Extreme Edition.

Pros and cons of 478 socket

Feedback from IT professionals indicates that the Intel Pentium 4 processor, which operates on the socket 478 platform, is still considered in demand. Not every user can afford an upgrade that requires the purchase of three basic components. It is worth noting that for many tasks designed to improve the performance of the entire system, it is worth simply installing a more powerful crystal. It's good that the secondary market is full of them, since the processor is even more durable than the motherboard.

When designing an upgrade, priority should be given to the most powerful members of this category, the Extreme Edition, which currently perform well in the performance test. One of the minuses of processors for Socket 478 is the dissipated power, which requires decent cooling. Thus, the need to purchase a decent cooler is added to the user's expenses.

Processors at low cost

Surely, many users have come across models of Intel Pentium 4 processors on the market. They have the inscription Celeron in the marking. These devices are the younger line of units that have less power due to the reduction of instructions, as well as the disabling of internal memory blocks of the microprocessor (cache). Intel Celeron is designed for users who are primarily concerned with the cost of the computer, and not its performance. Many owners of such devices express the opinion that the junior line of processors is considered a rejection during the production of Intel Pentium 4 crystals.

This assumption appeared on the market in 1999, when some enthusiasts proved that the Pentium 2 and its junior Celeron model are the same processor. However, the situation has changed a lot in recent years. Now the developer has a separate line for the production of a relatively cheap device intended for undemanding buyers. In addition, it is worth remembering that there is still a competitor to AMD, which claims to oust Intel from the market. Thus, all price niches should be occupied by high-quality products.

A new round of evolution

Most experts working in the field of computer technology are of the opinion that it was the appearance on the market of the Intel Pentium 4 Prescott processor that marked the beginning of the era of devices with multiple cores, and also completed the race for gigahertz. With the introduction of new technologies, the developer needed to switch to socket 775, which made it possible to unleash the potential of personal computers in working with programs and dynamic games that require large amounts of resources.

Statistics show that more than 50% of all devices that exist on the planet are able to work on the legendary Socket 775 provided by Intel. The release of the Intel Pentium D processor caused a stir in the market, as the developer managed to run two instruction streams on one core, thereby creating a prototype of a dual-core device.

This technology became known as Hyper-threading. Today it is an advanced solution in the production of crystals with high power. Intel did not stop there and presented Dual Core, Core 2 Duo and Core 2 Quad technologies, which have several microprocessors on a single chip at the hardware level.

Two-faced processors

If we take a benchmark for the "price-quality" criterion, then processors with two cores turn out to be an advantage. They differ as important characteristics as low cost and high performance. Micro Intel processors Pentium Dual Core and Core 2 Duo are considered the most sold in the world. The main difference is that the latter has two physical cores that work independently of each other. As for the Dual Core processor, it is made in the form of two controllers installed on one chip, teamwork which are inextricably linked.

Truth, frequency range devices with two cores is slightly underestimated and is in the range of 2-2.66 GHz. The main problem is the power dissipation of the crystal. It gets quite hot at higher frequencies. An example is the eighth line of Intel Pentium D (D820-D840). They were the first to receive two separate cores, as well as operating frequencies exceeding 3 GHz. The power consumption of these processors reaches about 130 watts.

Busting with four cores

The four-core enhancements with Intel(R) Pentium(R) 4 cores were aimed at consumers looking to stock up on components for the future. But the software market suddenly stopped. Thus, the development, testing, and implementation of applications is carried out for equipment that has one or two cores at most. What to do with systems that have 6, 8 or more microprocessors?

This is an ordinary marketing ploy, which is aimed at potential buyers who want to purchase a computer or laptop of the highest power that exists in the world. You can draw an analogy with megapixels on a camera - the best is not the one that says 20 megapixels, but a device with a larger matrix and focal length. In processors, what matters is the set of instructions that are processed by the application's program code. They give the result to the user.

Thus, programmers must optimize this move so that the microprocessor can process it without problems and at high speed. It is worth noting that weak computers there are many on the market, so it becomes profitable for manufacturers to develop non-resource-intensive programs. From this we can conclude that a large computer power at this stage of evolution is not required.

Upgrade Tips

Owners of the Intel Pentium 4 processor (775 socket) who want to upgrade at minimal cost are advised to look towards the secondary market. First you need to familiarize yourself with the technical characteristics of the motherboard installed in the system. It is easy to do this on the official website of the developer. There you should find the "processor support" section. Then, in the media, you need to find a table of processor performance, and then compare it with the characteristics of the motherboard, selecting several optimal options. It is also necessary to study the reviews on the selected devices.

Then it is proposed to start searching for the required processor, which was already in use. For most platforms that support quad-core microprocessors, it is desirable to install the Intel Core Quad 6600. When the system is capable of running only dual-core chips, you should find an Intel Xeon server option or a tool designed for the Intel Extreme Edition overlocker. Their market price is in the range of 800-1000 rubles, which is much cheaper than any upgrade.

Mobile device market

In addition to desktop computers, Intel Pentium 4 processors can be installed on laptops. For this, the developers provided a separate line, which contained the letter "M" in its own marking. As for the characteristics of mobile processors, they were similar desktop computers. True, an underestimated frequency range was observed. Thus, Pentium 4M 2.66 GHz has the highest power among notebook processors. Although, with the development of platforms in mobile versions, everything is so confused that even the developer Intel himself has not provided a processor development tree on its official website until today.

With the use of the 478-pin platform in laptops, the company changed only the technology for processing processor code. As a result, it turns out to breed many processors on one socket. The most popular, as evidenced by statistics, is the Intel Pentium Dual Core crystal. It is worth noting that it is the cheapest device in production, and its power dissipation is quite small when compared with analogues.

The race to save energy

It should be noted that for computers, the power consumed by the processor is not considered critical for the system. In the situation with a laptop, the situation is somewhat different. In this case, Intel Pentium 4 devices have been supplanted by less volatile microprocessors. If the user gets acquainted with the tests of mobile processors, he will be able to make sure that the performance of the old Core 2 Quad, which is part of the Pentium 4 line, is not far behind the modern Core i5 chip. As for the energy consumption of the latter, it is 3.5 times less. Thus, the difference is reflected in the autonomy of the device. If you follow the market for mobile processors, it is easy to see that the developer has returned to the technologies that were popular in the past decade.

By the beginning of 2004, Intel managed to successfully transfer its processors to the new Prescott core. True, the core itself cannot boast of improved characteristics. In particular, in terms of performance in most applications, it is inferior to the Northwood core (in some - up to 15%), and in terms of heat dissipation it significantly exceeds it. But the problem of increased energy consumption is inherent in C0 stepping. And recently, Intel switched to the release of processors on a new stepping - D0, in which this problem is partially solved. And finally it will be solved in the next stepping - E0, which will introduce a mechanism for reducing the frequency during processor idle. But for now, the main stepping is D0, on which processors of both Socket478 and Socket LGA775 form factors are produced.

What caused the need for a new socket? The main version is a more even distribution of power consumption between different blocks of the processor core. In addition, in the near future Intel time will introduce several new technologies such as EM64T (64-bit command extension), NX-bit (additional information security features), as well as an improved power saving mechanism. It is possible that additional contacts will be needed to support them. By the way, according to preliminary information, all these technologies are already present in today's Prescott processors, but in a blocked form.

Another new technology that should appear in the near future (tentatively - in the E0 stepping) is SpeedStep. Thanks to it, the processor during idle time will reduce the clock frequency, and as a result, generate less heat. And if the decrease in frequency is serious (for example, by 2 times), and is accompanied by a decrease in the Vcore voltage, then a radical decrease in the typical level of heat generation is possible. Let me remind you that AMD Athlon64 processors already support a similar technology - Cool "n" Quiet, which, by reducing the frequency and voltage, reduces the heat dissipation by more than 2 times (35W vs. 89W for details in the AMD Athlon64 review).

And again we return to the problem of energy consumption. Intel experts estimate the technological potential of the Prescott core - 4GHz. And at this frequency, the maximum heat dissipation can reach 150W. Therefore, the use of a new socket, a new design of the power module and a new design of the cooling system is intended to realize this potential.

Intel decided not to limit itself to a simple change processor socket. In fact, the public is presented with a completely new platform: support DDR2 memory, support PCI bus Express, as well as advanced options for connecting peripherals. For this, the i925X and i915P chipsets were released. We will not dwell on them in detail, because we have already carefully analyzed the i925X capabilities in the review of the Abit AA8 DuraMAX board.

Back to the processors - for the LGA775 socket, Intel announced the following processors:

Bold type indicates "processor number", which is intended for a clear division of processors into classes. In fact, this means a departure from the outdated system of classifying processors by clock frequency.

After the transition of Pentium4 processors to a faster 1066 MHz bus, the corresponding models will most likely make up the "sixth" series, and will occupy an intermediate position between the "fifth" and "seventh" series (the "seventh" series includes Pentium4 Extreme Edition processors with 2 MB L3 cache).

As for the Celeron processors, it is worth noting their increased characteristics. In particular, the amount of L2 cache has increased from 128 to 256 KB, and the system bus frequency has increased from 100 to 133 MHz (QPB: from 400 to 533 MHz, respectively).

So, let's see what the Pentium4 540 processor is.

The CPU-Z utility correctly determined all processor parameters, including stepping (D0). As for the appearance, there are no surprises for regular readers.

Socket478 on the left, LGA775 on the right

And for those who see the LGA775 processor for the first time, please pay attention to complete absence legs.

Now the legs are directly on the processor socket (you can see all the stages of installing the processor in the preview of the LGA775 platform). By the way, almost immediately after the appearance of the first samples of motherboards with LGA775, many observers began to complain about the fragility and unreliability of the processor socket. The most common problem is that after a few CPU socket installs, the legs become deformed (or bent).

Naturally, after receiving the LGA775 platform, I installed the processor with great care. However, no difficulties were identified during the installation process. Moreover, in my opinion, the problem with the unreliability of the socket is somewhat exaggerated (on the other hand, "crooked" hands can break anything :). In any case, as soon as the first "budget" board with LGA775 gets to us, we will conduct a kind of "stress testing" of the LGA775 socket for multiple processor installations.