Description of the test platform.

The first AMD64 architecture processors appeared in April of this year. At that time, AMD introduced the Opteron 200 series server models. They could be used in single and dual processor configurations. Unfortunately, the frequencies of the presented processors (1.4-1.8 GHz) did not please users very much at first. However, thanks to its unique architecture, Opteron performed well. By autumn, the Opteron lineup expanded with both new frequencies and new series. Today, AMD already offers three series of processors for use in one- (series 100), two- (series 200) and four- or eight-processor (series 800) systems. The maximum frequency for Opteron processors is currently 2 GHz (XX6 models).

However, “not by single servers”, and the market was waiting and even demanding to show something really new, massive, inexpensive - for everyone. A lot of rumors and assumptions about the frequency, socket, L2 cache size and even the name of the new desktop processors excited the imagination. And in the last third of September, AMD finally revealed its plans to conquer the market.

• AMD Athlon 64 3200+
• AMD Athlon 64FX-51

In addition, the release of processors for notebooks (DTR class (DeskTop Replacement), desktop PC replacement class) with 3000+ and 3200+ ratings has been announced, but since they differ from Athlon 64 only in the absence of a die-covering cover, there is not much to talk about them yet. we will, but just a little later we will publish an article about such a processor. We only note that the Cool "n" Quiet mobile technology of dynamic frequency and voltage change can be used in all processors of the AMD64 architecture, it's only a matter of supporting such functionality by the motherboard. And of course, for now, Mobile Athlon 64 processors can only be used in DTR systems: they consume up to 89 watts - for example, the 3000+ version consumes 81 watts. By the way, this figure for Opteron is 85 W for junior models and 89 W for 2.0 GHz and higher (the same applies to Athlon 64/Athlon 64 FX) - for AMD64 architecture processors of all lines, the power consumption is determined solely by the frequency.

So, now let's try to put everything in its place. To get started, we advise you to read our past materials on the AMD64 architecture:

• Testing Athlon 64 and Opteron Processors in Real Applications

Since a lot has already been said and written about Opteron processors, we will describe the new products in the form of differences from them, since the cores are almost the same for all.

The processor called Athlon 64 uses Socket 754 and has a single-channel integrated memory controller with support for DDR400 (not registered!). It replaced the Athlon XP, which will be gradually forced out of the market. Despite the fact that the performance index of the new processor is the same as that of its predecessor (and the frequency is even lower), significant differences in the architecture allow us to hope that it will outperform Athlon XP 3200+ in speed.

It's even easier with Athlon 64 FX - at the time of the announcement, it differed from Opteron only by the frequency, which for the FX-51 model is 2.2 GHz. Of course, formally there is also a difference in support for DDR400 memory, however, as we will see later, this does not count :). AMD is positioning this processor as a high-end desktop model. Although, given its complete interchangeability with the Opteron (in single-processor systems), it becomes clear that the "positioning" is very shaky, and can easily be ignored by particularly smart buyers. :)

Despite the fact that both sockets have the same spacing of 1.27 mm between pins in the grid, Socket 754 is not a subset of Socket 940, since its pins are located in a 29 by 29 mm square versus 31 by 31 mm for the 940. Therefore, unlike, for example, the well-known i865/i875 and i848 pair, manufacturers will have to create different board designs for these products.

However, both sockets use the same cooling device mounting system.

The base, on which the cooler is actually attached, consists of two parts: a metal substrate and a plastic frame, which are located with different parties motherboard and fastened with two screws. The cooler itself is attached to the frame with two powerful latches.

The coolers we used had a copper base and welded copper fins. The design is similar to the well-known Thermaltake Volcano 7+/11+ models. By the way, according to the abundance of signs of this trademark on different parts boxed cooler, it can be assumed that it was this company that helped AMD in the development of cooling systems for new processors. The sizes of different models are slightly different. The boxed version from the Opteron 240 (working seamlessly with faster processors, including the Opteron 146) used a 55x75x5mm base and 46 12cm2 fins. The 70x70x15 mm Delta fan model AFB0712HBB had a built-in temperature sensor for speed control (the maximum value is 4300 rpm). The variant from Thermaltake had different parameters: a 65x60x4 base and 36 18 cm 2 fins, the same fan, but without a sensor. In addition to all-copper versions, there was one aluminum version with a copper cylinder inside. In addition, it is possible to use Zalman CNPS7000-Cu (however, it is fastened with screws and therefore is not very convenient for frequent replacements).

In principle, the design of the cooler assumes that it also slightly blows the memory modules located next to the processor, however, one of the versions used had ribs oriented along the long side of the socket and therefore (at least on the tested boards) is unsuitable for this purpose.

In terms of noise, all fans are very quiet (Delta has a noise rating of 38.5 dBA at maximum speed). So, from this point of view, the new AMD products are all right, despite the fact that they have almost twice as many transistors in the core as the Athlon XP (105.9 million versus 54.3).

Here is a summary table of the parameters of old and new processors that claim a place in system block desktop PC. The Opteron looks a bit alien here, of course, and is presented rather for visual comparison with Athlon 64 FX. However, the price of the 100 series models is not so terrible - from $250.

* determined by chipset

Despite the fact that this table contains official data, there is an inaccuracy in it - in fact, Opteron processors (we checked both the models of the early revision - B3, and the latest - C0) work fine with DDR400 memory too! The point, it turns out, is only that there were no registered modules with such a speed in April. And memory validation for server systems is a slow process. Let's assume that AMD just played it safe.

As for the company's future plans, only one thing can be assumed here - the frequencies will increase. The previous architecture (Barton core) reached 2.2 GHz, and the Athlon 64 FX starts from there. So hopefully there will be next, faster processors, but the revolutionary part is over. The next big step is the transition to 90 nm technology.

Externally, the processors practically do not differ from each other. Only the Athlon 64 has a case similar to the latest "green" Athlon XP with an organic base, while the Athlon 64 FX and Opteron have a ceramic case. And of course, they are all closed with a metal lid.

As for the labeling, one sentence will not do here :), but we will try, based on the current information, to decipher at least something. Note that this information is not strictly official, so changes and additions are possible in the future.

We dealt with the following processors:

• Opteron 240: OSA240CCO5AH
• Opteron 244: OSA244CEP5AL
• Opteron 146: OSA146CEP5AK
• Athlon 64 FX-51: ADAFX51CEP5AK
• Athlon 64 3200+: ADA3200AEP5AP

So, the first letter speaks about the brand: O - Opteron, A - Athlon 64. The second - about the application: S - Server, D - Desktop. Of course, for now we only have combinations of OS and AD, but who knows, maybe AMD will also release a server Athlon 64? :-)

The third letter, according to some sources, defines a certain "Power Limit". However, there are no detailed explanations yet, and all the tested processors have the letter “A” here, so you can’t distinguish them by this parameter yet.

Finally, the fourth item we have is the model number. For Opteron, these are three digits, the first is the series number, the second is still four, and the last, always even, determines the frequency: from “0” for 1.4 GHz to “6” for 2.0 GHz. For Athlon 64, we see here the performance index in the form of four digits, which correspond to the name of a particular model. The situation is similar with Athlon 64 FX.

The package version follows: A - 754-pin OuPGA with a cover (for Athlon 64), B - 754-pin OuPGA without a cover (mobile Athlon 64) and C - 940-pin CuPGA also with an iron cover for Opteron and Athlon 64 FX .

The next letter shows the core voltage. For the first Opteron model we tested, it is 1.55V (letter C), and for all others it is 1.50V (letter E). Letters through one to Y are provided, which corresponds to a value of 1.00 V.

The seventh indicator determines the operating temperature of the processor. "O" corresponds to 69°C, "P" to 70°C. The next letters in alphabetical order are b. about high temperature, up to "Z" - 105 degrees Celsius.

The last figure shows the amount of L2 cache of the processor: 1 - 64 KB, 2 - 128 KB, 3 - 256 KB, 4 - 512 KB, 5 - 1 MB. As you can easily see, representatives of the AMD64 architecture do not yet have less than one megabyte of cache.

And finally, the last two letters define the stepping, revision, socket, number of coherent HT buses and all that. The main thing to remember is that if the letters are older than AI, then this is a C0 stepping or higher.

In general, the most important (and easy to remember :-)) are the first three letters that identify the server or desktop processor, and of course the model index, which shows the performance in units known only to the at manufacturer. :-)

Since performance is not the only thing that interests buyers, we will also announce the prices at which the company plans to sell new products: $417 for Athlon 64 3200+ and $733 for Athlon 64 FX-51 (mobile processors will go for $417 and $278 for models 3200+ and 3000+ respectively). In general, prices are at the level of high-end desktop processors, but up to the coveted "$64 for 64 bits!" still very, very far away. On the other hand, this is only the beginning, and we can expect a significant reduction in prices in the coming months, but now all this is only for the very impatient. Well, the number of processors sold will be determined by the results they show in performance tests.

As you remember, during the introduction of Athlon XP, AMD published a list of applications that it used to assign ratings. But using not even a rating, but a code name (FX-51) for a desktop processor, the company once again emphasized its original approach to the concept of "performance".

The modern version of the list of applications used to evaluate speed looks like this:

Of course, compared to the previous version, it has become a little better - popular tasks such as encoding media data and archiving have been added. On the other hand, abundance synthetic tests like SYSmark and Winstone is a bit confusing. Since it has long been known that any modern processor with a frequency of about 2 GHz is able to provide decent work in modern office applications. Of course, there are examples of receiving 1000 emails with packed attachments per day and constant verification all this (including a two-gigabyte mail database) with an antivirus, but in this case it is not hardware that needs to be upgraded :-), and the indicated synthetics do not lose such a situation.

We also throw out 3DMark tests with "D3D Software T&L" there, because if a person already spent money on such a processor and did not buy a decent video card, then, apparently, he will not play on a computer.

With some games like QuakeIII, it's also not very clear - is it worth buying a new processor to increase the number of fps from 220 to 290? :-) Yes, and in the test guide from AMD, it sometimes slips "Select "Preferences" to "Speed". On the one hand, of course, it is clear that we do not want to test a video card, but

In general, what remains is encoding to MP3 (although ... and so it takes 5-10 minutes per disc, why is it faster? :-)), conversion to MPEG2 (but it’s also not clear why this should be done from RAW AVI? All discs are large and fast to store more than one and a half gigabytes per minute?), but “MPEG2 to MPEG4” definitely continues to unnerve with its slowness.

Clearly, there are not enough rendering class tasks and computational tasks. Apparently, the company refers these applications to workstations. In general, perhaps, this is correct, because, according to numerous surveys, powerful PCs at home are usually used for you know what :-). However, the positioning (again, this is a suspicious word :-)) of the Athlon 64 FX processor can easily be corrected towards "entry-level workstations", if it shows a decent speed in these applications.

64-bit applications and Windows XP for AMD64

We want to warn you in advance that despite the numbers "64" in the name, we will not really use 64-bit extensions on desktops soon. Of course, enthusiasts can already try them using the corresponding versions of Linux, but the real mass distribution of the 64-bit mode will only begin with the release by Microsoft of its Windows OS for this platform. AT this moment the company is working on two versions of the OS - server and desktop. Both of them already exist in the form of beta versions. We had an opportunity to get acquainted with the pre-release of Windows XP for AMD64.

As you can see in the screenshot, running the usual Microsoft Office XP, VirtualDub programs with DivX codec, FAR file manager were successful. What cannot be said about graphic applications. Despite "full compatibility", an attempt to run the QuakeIII and Return to Castle Wolfenstein games ended in failure (the games could not be configured graphics system). While Serious Sam: The Second Encounter and Unreal Tournament 2003 Demo worked without problems. As for speed, its performance in 3D applications, which are games, is very big influence provide video card drivers. In this case, detonators NVIDIA versions 50.30 from May of this year missed the stars from the sky and showed a 30 percent drop in speed compared to Windows XP Pro with the 45.23 driver. Apparently, it is the porting of drivers for a new system (which is mandatory, since the drivers in it must be 64-bit) that will be the main problem at first. Note that the OS hides them so that you can only find the actual driver files manually in Explorer. Trying to find them by searching in the explorer or file FAR manager ended in failure. There are also doubts about the version of the NVIDIA driver being used, since the number 50.40 and the date August 8 this year appear in the properties of the driver file.

Of course, and most console applications should also not have problems running under this version of the OS. The exceptions are programs that use 16-bit code (for example, in libraries), and those that run special system drivers for their work, for example, to access hardware resources (one of these programs is a utility for obtaining information about the processor, motherboard and memory, CPU-Z - could not show all the information in full under Windows XP for AMD64). Well, the fact that the performance of win32 applications (not graphical ones) in the new OS is at least no worse than in the 32-bit version is also evidenced by the fact that the SPEC CPU2000 test results, some subtests of which are very sensitive to memory speed , practically do not change when working in Windows XP for AMD64.

Chipsets

Chipsets for processors of the AMD64 architecture differ in that in the case of a desktop application, they practically do not affect the speed. Judge for yourself: the memory in such systems is connected directly to the processor, and the only formally "fat" consumer of information - the video card - has long acquired its own voluminous and fast memory. So the main streams of information circulate outside the chipset. Yes, of course, there is network and storage, but standard 100BaseTX requires only about 10 MB / s, and hard drives although they are improving the interface in the direction of 150 MB / s, but (also for desktops) they themselves are only approaching reading speeds from the surface of the order of 70-80 MB / s.

Of course, for workstations we also have gigabit network controllers and RAID arrays on hard drives, but that's a completely different story.

Another interesting property of chipsets is their versatility and scalability. Since they communicate with the processor(s) exclusively via the standard HyperTransport bus, taking into account the positive experience with Socket A, manufacturers can count on a long life of their developments. Well, the fact that any chipset (at least formally) can work with one or two or more processors allows you to position one product on several markets at the same time.

However, the first generation of desktop chipsets have a common drawback - they support only one HT bus. As you may remember from previous posts, the AMD8000 chipset has excellent expandability, since most chips have two HT buses and can be connected in series (although the “output” bus is only eight-bit). Since the current edition of HT supports transfer rates up to 6.4 GB / s, this eliminates the bottleneck for six PCI-X buses, twelve PCI 2.2 64 bit / 66 MHz, or 48 regular PCI 32 bit / 33 MHz.

Unfortunately, existing non-AMD solutions lack these capabilities and are limited to conventional PCs, and to move to the next level, manufacturers will have to come up with something new.

Note that in addition to the products from NVIDIA () and VIA () considered today, products from ALI () and SiS () have also entered the chipset market for new AMD processors. Now these are two-chip solutions, but single-chip products are also in the plans. In addition, chipsets with PCI Express and 3GIO bus support are expected to appear in the future. By this time, ATI also promises to present its chipsets, including the variant with integrated graphics.

NVIDIA

One of the first third-party chipsets for AMD processors was NVIDIA nForce3 Pro 150. This single-chip solution combines both a bridge to support AGP and PCI buses, and all controllers standard for the south bridge:

• 2 PATA/IDE channels with UltraATA 133 and RAID support
• Fast Ethernet network controller
• 6 USB 2.0 ports
• AC "97 sound controller with support for 5.1 and digital output

The next version of the chipset - with an index of 250 - is planned to include a gigabit network controller, 2 PATA ports and 4 SATA port. Well, today's boards are used for SATA and gigabit ethernet external chips.

Motherboards based on this chipset participate in testing today: ASUS SK8N for Socket 940 and Gigabyte K8NNXP for Socket 754.

Since the main topic of the article is new processors, here we will only give brief characteristics of the boards, and leave a detailed comparison until next time.

Note that Athlon 64 processors have some limitations in terms of speeds and memory size, caused by the use of non-registered modules. In particular, only 2 modules can be used at 400 MHz, which limits the maximum amount of RAM in this case to 2 GB.

As it usually happens, the manufacturer tries to fill the first products for the new architecture to the maximum, believing that the first buyers have a lot of money and can afford to spend a significant amount. So it happened with SK8N and K8NNXP. Now they can be purchased for about $200. Of course, this is too much for the mass market. Of course, we will soon see versions without FireWire and SATA controllers, which will be cheaper. And daily announcements from other manufacturers indicate future competition in the market for boards for new AMD processors, which will also lead to lower prices.

VIA

VIA also couldn't refuse such a fresh market and released its own chipset for new AMD processors - VIA K8T800. By the way, according to the first reviews of Athlon 64 on the Web, you should also remember the phantom called K8T400M (or even K8M400 - with an integrated video controller), which did not reach mass production of motherboards. While AMD was postponing the release of their desktop processor, VIA released a new version of their chipset :-) (though most likely just renamed the old one).

Unlike the nForce3 chipset, it is made in an almost classic version - with a north and south bridge, which are connected by an 8X V-Link bus with throughput 533 MB / s (some sources indicate a figure of 1 GB / s). The high-end southbridge uses the VT8237 chip (already known from KT600 boards), which supports:

• eight USB 2.0 ports
• two Parallel ATA133/100/66 ports supporting up to 4 devices
• sound solutions from VIA: VIA Vinyl 5.1 & Vinyl Gold 7.1
• two SATA ports with RAID support (V-RAID: RAID 0, RAID 1, RAID 0+1, JBOD)
• integrated 10/100 BaseT network controller
• Gigabit Ethernet companion controller connection

As one of the advantages of its chipset, the company presents Hyper8 technology, behind the beautiful name of which lies support for the HyperTransport bus mode between the processor and the chipset 16 bit/800 MHz in both directions.

Indeed, for nForce3 boards these parameters are "only" 8 bit/600 MHz in one direction and 16 bit/600 MHz in the other. However, this is formal big difference plays virtually no role today, since the only serious consumer of data in any AMD64 chipset is the video controller on the AGP bus, which is currently almost unloaded with real work. Perhaps, in the future, for workstations and servers with PCI-X and PCI Express buses, this will be important, but now it is somewhat premature. Because the Board BIOS on the K8T800 allows you to adjust the bit depth and frequency of the HT bus, we conducted express testing in Return to Castle Wolfenstein and SPECviewperf and did not reveal any differences in speed when working in these modes.

ASUS K8V Deluxe and MSI K8T Neo motherboards for Socket 754 took part in the tests. The test results of the boards are almost the same. For definiteness, the diagrams show indicators of the board from ASUS. But we advise you to treat the results with caution, since beta versions of the BIOS were used, and a lot can change with the release.

As you can see from the table, both models are typical examples of high-end motherboards. Both use external gigabit network adapters, 5.1 sound controllers allow you to connect speakers through optical and coaxial digital outputs. The possible number of drives is also impressive - 6 are connected only to the south bridge and an external ATA / RAID controller is still left in stock.

Note that the ASUS board has a special slot for connecting own card wireless radio access (included with the Deluxe version) standard 802.11b (11 Mbps).

Configurations

Processors:

• AMD Athlon XP 3200+
• AMD Athlon 64 3200+
• AMD Athlon 64FX-51
• AMD Opteron 146
• Intel Pentium 4 3.2 GHz

Motherboards:

• Athlon XP (Socket A): Albatron KX18D Pro II (nForce2 Ultra 400)
• Athlon 64 (Socket 754): Gigabyte K8NNXP (nForce3 Pro 150), ASUS K8V Deluxe (K8T800)
• Athlon 64 FX, Opteron (Socket 940): ASUS SK8N (nForce3 Pro 150)
• Pentium 4 (Socket 478): ASUS P4C800 Deluxe (i875P)

• two 256 MB Kingmax DDR400 modules (2-3-3-5) for Athlon 64, Athlon XP and Pentium 4 systems
• two 512 MB modules from Legacy Electronics DDR400 ECC Registered (2.5-3-3-5) for systems based on Athlon 64 FX-51 and Opteron (also used as DDR333 with the same timings), ECC control was disabled in the BIOS.

Video card:

• ATI Radeon 9800 Pro 256MB

HDD:

• Western Digital WD360 (SATA, 10000 rpm)

Software and drivers:

• Windows XP Pro SP1
• DirectX 9.0b
• set of drivers for NVIDIA nForce3 version 3.44
• chipset driver Intel version 5.0.2.1003
• ATI CATALYST 3.7 video driver

Test results

First, we note that the methodology for testing systems in this article differs from that used earlier. So the results cannot be directly compared. Moreover, we also changed the video card.

Of course, we did not use the entire list of applications proposed by AMD. This time we'll look at gaming, media encoding, and archiving as the most CPU-intensive desktop applications.

To improve accuracy, all tests on real applications were run at least three times, and the median was chosen for the report.

Games

The following applications were used for gaming performance testing:

• Return to Castle Wolfenstein 1.41, id Software/Activision
• Serious Sam: The Second Encounter 1.07, Croteam/GodGames
• Unreal Tournament 2003 Demo 2206, Digital Extreme/Epic Games

The demo scenes recorded in these programs (checkpoint, Grand Cathedral, botmatch-antalus, flyby-antalus) were played in different resolutions with the optimization of the "Quality" settings set in the game itself. No changes were made to the graphics card drivers other than disabling VSync.

Note that the results showed a high dependence of speed on resolution and, consequently, on the video card. Only the number of fps in the botmatch-antalus scene practically did not decrease with an increase in resolution. 1024x768 results are selected for the report. When playing at 800x600, the gap between the participants will be larger, while at 1600x1200 it will noticeably decrease. And if you use the anti-aliasing and anisotropy modes, then it may turn out that there will be no difference in the results at all.

In this one, enough old game have always been favorite processors by Intel. However, with the release of 64-bit processors from AMD, the situation has changed dramatically. The new processors with a frequency of 2 GHz are on a par with the Pentium 4 3.2 GHz, and the Athlon 64 FX increases its result by almost 10% in proportion to the frequency and takes the lead.

This game already loves AMD products more. And if earlier we had parity between Athlon XP 3200+ and Pentium 4 3.2 GHz, now the new processors are taking the lead. Like last time, the leader is Athlon 64 FX-51.

Let's also look at the dependence of the results on the resolution. The following two charts only show data for the Athlon 64 FX-51 and Pentium 4 3.2 GHz.

We see that RtCW is an easy task for the ATI RADEON 9800 Pro, and the results are almost independent of the resolution. The advantage of Athlon 64 FX is from 10 to 6% depending on the resolution.

For Serious Sam: The Second Encounter, the situation is different - at 1600x1200 the results of the systems are almost the same, but at 800x600 the difference is almost 30%.

In this game, the results generally repeat the data for Serious Sam: The Second Encounter. However, the spread of indicators in the flyby test is smaller and amounts to only 10%, while in the botmatch demo, which is more difficult for the processor, the leader outperforms the competitor by 25%.

For comparison, we also tested the two most fast systems and with a video card NVIDIA GeForce FX 5900 Ultra (driver 45.23).

On the whole, the alignment of forces remains the same in this case: Athlon 64 FX-51 outperforms Pentium 4 3.2 GHz from 7.5% in RtCW to 26.7% in UT2003 botmatch.

media coding

As before, two popular tasks are used: encoding music to MP3 format and video encoding to MPEG4(DivX) format. However, this time different settings and versions of the programs are used.

For the first task, we took the Lame 3.93 codec and used three settings:

• --preset standard -m s
• --preset 192 -ms
• --preset cbr 192 -m s

All of them create files of approximately the same size with an average bitrate of 192 Kbps. The original was a 71-minute WAV file (rewritten from CD-DA).

In this test, we see a clear dependence of the encoding speed on the frequency, and the Athlon XP 3200+ easily overtakes all new AMD processors with a frequency of 2.0 GHz and even slightly outperforms the Athlon 64 FX-51. And the product from Intel takes the lead with its 3.2 GHz. Its separation from the nearest pursuer is about 10%.

Video encoding in DivX (version 5.1 codec) was made from a movie trailer in MPEG2 format (length 2:25, resolution 720x576) in VirtualDub(with support for reading MPEG2 format, version 1.5.4) using crop, deinterlace and resize filters.

And again the Pentium 4 3.2 GHz is in the lead, but this time the Athlon 64 FX-51 almost caught up with it. But the Athlon XP 3200+ failed badly on this task. In principle, we can assume that the problem is the lack of SSE2 in the latter, but we have practically no information about SIMD support for the DivX codec, so we cannot say that this is the case. Just like with Lame, it is noticeable that the results are practically independent of the memory speed.

Archiving

Two programs were used for archiving: the console version of RAR (version 3.20) and 7-Zip (version 3.09.01 beta). Settings on maximum compression: -m5 for RAR and -mx9 for 7-Zip.

The following were used as input files:

• source code Linux kernels(approx. 150 MB)
• drivers for NVIDIA graphics cards(approx. 100 MB)

We have already used the 7-Zip archiver before. It shows one of the best results in terms of compression ratio, but you have to pay for this with a long time. As an example, the table shows the efficiency in the maximum compression mode (the ratio of the volumes of the input and output files) and the running time of archivers in seconds. The zip format is the win32 console version of the pkzip archiver version 2.50 from PKWARE.

By the way, this table shows why we excluded archiving to the zip format from the tests - its speed is determined more by the parameters hard drive than the processor. And the compression ratio is noticeably lower than that of competitors.

The only test where we see a noticeable difference in Athlon 64 performance on different chipsets. Moreover, its speed on nForce3 is the best among all participants. The difference between this configuration and the rest is the use of a Sil3512 SATA controller. Perhaps this is the case, or maybe there is some other secret in the NVIDIA chipset.

If we compare Pentium 4 3.2 GHz and Athlon 64 FX-51, then the latter is slightly ahead this time.

Here we have a different situation. The test shows a dependence on both memory speed (which is not surprising, since 7-Zip takes more than 300 MB of RAM when archiving test files) and processor frequency. And it seems that he likes the integrated controller in AMD processors more because of the lower delays. And again in this test Athlon 64 on nForce3 shows good result and almost catches up with the leader.

conclusions

Let's look at the final table of results:

So, we see that the new AMD Athlon 64 FX-51 processor in gaming applications shows excellent performance, 10 percent or more ahead of its direct Intel competitor Pentium 4 3.2 GHz. However, let's not forget that the results strongly depend on the video card used, and if your 3D accelerator is not of the highest class, then ... you need to go to the store and buy it as soon as possible :-), otherwise you may not notice the effect of the money spent on the processor.

In MP3 encoding, Intel's product is unbeatable - high frequency the kernel solves everything in this task. Tests show that the memory subsystem in this case has almost no noticeable effect on the result.

MPEG2 to DivX encoding is a more complex task, both core speed and processor-memory bus performance are important here. So Athlon 64 FX is practically catching up with Pentium 4. Other AMD processors show better results than their predecessor Athlon XP.

In archiving tasks, the Athlon 64 FX is also ahead of the competition. And for 7-Zip, this is the merit of the integrated memory controller, which provided low memory access latencies.

As for the comparison NVIDIA chipsets and VIA for Athlon 64, then in all tests, except for archiving to RAR, their results are practically the same. However, please consider the results of K8T800 as preliminary.

In general, our previous assumptions about the performance of the new AMD processors came true. Yes, they are good, but not as good as everyone would like. Of course, the potential of architecture is also visible on these samples, but buyers are usually not interested in abstract reasoning, but in real results. It's hard to say whether the Athlon XP core has exhausted itself, but AMD really needed to introduce something new and original. And I think she succeeded.

Of course, today we have not reviewed all the tests of the new processor, but for a start it is quite enough. Ahead we have a discussion of the results of tests on professional applications, as well as numerous synthetics.

And finally, let's try to figure out why AMD suddenly found such an interesting processor as the Athlon 64 FX-51 - in all respects it is very reminiscent of the delayed Opteron 148. As one of the scenarios, and quite plausible, we propose the following.

Starting from April, the development of the Opteron line went on as usual - the frequency increased, new series were released. At the same time, the operation of the Athlon 64 processor was also tested, which, unlike the Opteron, used a single-channel memory controller, and it is probably impossible to say that it was “developed separately from the Opteron”. And the use of non-registered modules also seems natural for a desktop processor. It is not very clear why, but the frequency of the first Athlon 64 was 2.0 GHz. This was obviously not enough to compete with the Pentium 4 3.2 GHz. In addition, having a single-channel memory controller, the processor was outperformed by its competitor on this formal basis. And this is despite today's results - in games the Athlon 64 3200+ still outperforms its competitor, in archiving too, only the encoding speed in MP3 and DivX let us down.

However, AMD needed a bright and unconditional victory. So, using a version of, in general, a server processor with a frequency of 2.2 GHz and a dual-channel memory controller, and making sure that register modules with a frequency of 400 MHz were already produced in sufficient volumes, she introduced a new brand - Athlon 64 FX, the first representative which differed from other models in two parameters at once: frequency (cores) and memory speed from Opteron and frequency (cores) and a dual-channel controller from Athlon 64.

This will not hurt sales of the Opteron line, especially since no one bothers to release these processors with a frequency of 2.2 GHz soon. Well, having set a price slightly higher than the cost of the Pentium 4 3.2 GHz, AMD has remained in the field of desktop processors.

True, there remains a slight ambiguity associated with the use of registered memory modules with this processor. Many expected AMD's high-end desktop to use conventional modules. But if this happened, then, firstly, it would be possible not to delay the announcement so long, and secondly, the processor could compete with the Opteron 100 series, having a higher frequency and working with cheaper memory. Of course, for most users, register modules (which, in fact, are needed to support large amounts of memory) are associated with the workstation and server market. However, it is strange to assume that the memory controller of the Athlon 64 FX and Opteron needs to be heavily modified to work with conventional modules - after all, Athlon 64 has no problems with this. So we again observe distant and inexplicable for common man market games.

The fate of the Athlon 64 FX is shrouded in mist. On the one hand, it is impossible to stop increasing AMD megahertz, on the other hand, the Opteron lineup is almost finished: x48 models will follow after x46 models, and then we will have to expand existing system designations. And the FX-51 will most likely be followed by the FX-53 with an increased frequency. Releasing a desktop processor that is completely similar to a server one, but with a higher frequency (and the ability to work only in single-processor configurations) means slowing down the pace of conquering the workstation market.

It would be strange to assume that AMD has technical problems with the release of processors with a high core frequency and two or three HT buses for operation in multiprocessor configurations. But to expect that the mass market will switch to registered memory is also not serious.

So under these conditions, AMD is likely to release 2.2 GHz Opteron models, which will remain the company's fastest server processors until the transition to 90nm technology. The Athlon 64 FX will clock up to 2.6 GHz or a little higher and will be AMD's flagship desktop processor. At the same time, given the need to use register memory, it will not be supplied in large quantities. Although if this restriction is suddenly canceled next year :-), then its chances of becoming widespread will greatly increase. Well, Athlon 64 will successfully replace modern Athlon XP.

Introduction

The first processors with AMD64 architecture began to appear in April 2003. These were the 200-series Opteron processors, which showed a pretty good level of performance due to their excellent architecture. The server market itself has a rather small percentage, from common market processors, so the announcement of desktop processors with the AMD64 architecture for high-performance PCs was not long in coming, so on September 23, 2003, the following models were officially presented: AMD Athlon 64 3200+ and AMD Athlon 64 FX-51, and then AMD Athlon 64 3400+. On the eve of the new year, fans of AMD products were also in for a surprise: without any excitement, the world saw the new Athlon 64 3000+ processor, which is aimed at the mass market, and which will be discussed in today's material.

Processor lineAMD 8th generation

AMD releases several models of 8th generation processors designed for specific market sectors.

• Athlon 64FX-51

For a better presentation and ease of perception of the material, we present a small plate in which the technical characteristics of the above processors are collected.

AMDAthlon 6430 00+

Rumors that AMD is preparing a new Athlon 64 processor with a rating of 3000+ quickly spread around the network. Most reviewers, testers and experts assumed that the new budget model would differ only in clock speed. No one has any doubts that it could be otherwise, especially if you look at the range of Athlon 64 mobile processors, so the model with a rating of 3200+ has a frequency of 2000 MHz, and 3000+ - 1800 MHz. The company's official release dotted the I. The new AMD Athlon 64 3000+ desktop processor has the same clock speed as the more expensive 3200+ rated model. The changes affected the second-level cache memory, the volume of which in Athlon 64 3000+ decreased by half compared to Athlon 64 3200+ and amounted to 512 KB against 1 MB in the 3200+ model. We present to your attention a screenshot from the CPU-Z program.

This decision by AMD has a logical explanation. The processors of the Athlon 64 family have a fairly large die area, which makes them quite expensive to manufacture. the number of marriages is quite large. A large percentage of marriage falls on cache memory, due to the fact that the latter occupies 50% of the crystal area. Thus, this is not the first time that AMD has been trying to kill two birds with one stone, let's remember the situation with the Barton and Thorton cores. Thus, the company succeeded in:

Release the necessary budget solution, like AMD Athlon 64 3000+, the market demand for which is very high

Good to get rid of more defective crystals

Theoretically, based on the situation with Barton and Thorton, as well as with the Pentium 4 and Pentium 4 Extreme Edition, we can assume that the performance of the Athlon 64 3000+ will not drop much compared to the Athlon 64 3200+. By itself, reducing or increasing the amount of cache memory does not lead to a significant decrease or increase in performance, striking examples of this are AMD Athlon processors based on Barton and Thorton cores, as well as Pentium processors 4 and Pentium 4 Extreme Edition. But theory is theory, and practice is practice, so we will postpone our conclusions until objective testing.

AMD Athlon 64 3000+, like older models, supports Cool'n'Quiet technology - an intelligent technology for reducing heat dissipation. In essence, Cool'n'Quiet is an advanced PowerNow! a large number of time is used in "mobile" processors from AMD. The principle of operation of the technology is quite simple: by means of a driver that resets or increases the processor clock frequency, the load degree of the central processor is determined, and in accordance with the data obtained, the operating frequency of the processor and, accordingly, the voltage on the processor are optimized. Here you can imagine a completely logical situation: the user is working with Word, so the degree of processor load is insignificant, the driver reduces the operating frequency and voltage on the processor core. The situation changes dramatically if you launch a modern computer game or other resource intensive application. Again, the driver determines the degree of loading, which has become maximum, after which the operating frequency of the processor increases and, accordingly, the voltage on the processor core.

Let's move from theory to practice. When starting normal office applications, clock frequency the processor dropped to 800 MHz, and the core voltage to 1.3V. It is also worth noting that the decrease in the operating frequency of the processor occurs due to a change in the multiplier, so in our case, at a frequency of 800 MHz, the multiplier was 4x.

After the load on the processor increases, the operating point changes and the processor operates at a frequency of 1800 MHz, and the voltage and multiplier are 1.4V and 9x, respectively.

In addition, AMD Athlon 64 processors can switch to the so-called “standby mode” (Halt/Stop Grant).

For the convenience of presenting the operating points and the amount of heat release on a particular operating frequency Here is a small table.

Box, cooler, sticker...

The AMD Athlon 64 3000+ processor was tested in a boxed version. In addition to the Athlon 64 3000+, the small box also contained: a cooler, a device for attaching the CPU to the motherboard, and a sticker with the processor logo.

It is worth saying a few things about the cooler that comes with AMD processors Athlon 64. The cooler has an alloy base with a large percentage copper content, to which a large number of thin ribs are attached. The fan, on two ball bearings with a built-in temperature sensor, has a rotation speed of 3050 rpm - 6000 rpm (varies depending on the temperature of the processor, the threshold is 42 degrees Celsius (below 42 - 3050 rpm, above 42 -<=6000 об/мин). Уровень шума не высок: субъективно он значительно ниже нежели у кулеров, поставляемых с процессорами Intel Pentium 4.

Testing

Test tests were carried out on a test bench with the following configuration:

Motherboards: MicroStar K8T Neo (VIA K8T800) and ASUS P4C800 Deluxe (Intel 875P)

Processor: AMD Athlon 64 3000+ and Intel Pentium 4 3000 MHz (800 MHz FSB, Northwood)

Memory: 2x256 MB PC3200 Hynix DDR SDRAM CL 2.0

Video card: ASUS V9560 Ultra (NVIDIA GeForce FX 5600 Ultra)

Hard drive: Seagate Barracuda 7, 80 GB

Of course, it would be nice to compare the performance of AMD Athlon 64 3000+ with other processors from the Athlon 64 line, but this is not possible due to the lack of such processors. Therefore, we had to limit ourselves to comparing AMD Athlon 64 3000+ with its main competitor Intel Pentium 4 3000 MHz.

On the test stand, the Microsoft Windows XP Service Pack 1 operating system was installed, as well as test programs and real gaming applications:

The memory timings on both boards were set as 2.0/5/3/3.

Synthetic tests 3DMark 2001 SE and 3DMark 2003, as well as the gaming benchmark GunMetal BenchMark, used maximum detail, 640x480 resolution and 32-bit color.

When archiving data, the WinRAR 3.20 archiver and the data folder (PCBench) from the ZD Winstone 2004 test package were used. This folder was chosen because it is large and contains almost all types of files.

Video compression tests were carried out using VirtualDub 1.5.1 and DivX codec 5.05a Pro. The compressed video file was 74.5 megabytes in size.

Mp3 encoding tests were carried out using the RazorLame 1.1.5.1342 encoder and the Lame codec 3.93.1 codec. Wave format files, namely Metallica's robbed album "Master Of Puppets", were compressed into Mp3 files with a bit rate of 128 kb / s and a sampling rate of 41 kHz.

Real gaming applications used 32-bit color and 800x600 resolution. VSync was disabled. Texture compression was disabled directly in game applications. All gaming applications were tuned for maximum detail.

With each new tested board, operating systems and all test applications were reinstalled.

Test results

Performance in office and multimedia applications

In the ZD Winstone 2004 test package, the system performance on the Athlon 64 is higher than the performance of the system based on the Intel Pentium 4 microprocessor.

Performance in synthetic benchmarks

The microprocessor benchmark from the Sandra 2003 test package indicates that the AMD microprocessor lags behind the Intel product. However, you should not take the results of this test to heart: it is a well-known fact that the SiSoftware Sandra test suite is already very loyal to Intel products.

Memory tests show us a similar picture.

The results of PCMark 2004 speak of the complete defeat of the Athlon 64 3000+. The results were really devastating, and the idea of ​​optimizing this test for the Pentium 4 microprocessor architecture crept into my brain. Let's try to understand the current situation, for this, let's look at the detailed test results from the PCMark 2004 package. here Hyper-Threading technology does its job.

The rest of the tests are mainly related to coding tasks, and the Intel Pentium 4 does a better job with this. Athlon 64 wins only in the grammar check and physical modeling tests.

The semi-synthetic package 3DMark 2001SE shows the superiority of AMD Athlon 64 3000+. The results of 3DMark 2003 show almost identical performance of both processors, which means that this test package is very dependent on the performance of the video subsystem.

Performance in tests for data compression, encoding / decoding audio and video

Data archiving is critical to the performance of the memory subsystem, and as we can see, the memory subsystem of the Athlon 64 platform is better organized.

In video encoding, the performance of the memory controller also plays an important role, and we see an identical slight lag behind the Athlon 64.

But Mp3 encoding is the opposite: it is very loyal to the memory subsystem, but categorically - to the performance of the central processor, and as we see here, the Intel Pentium 4 3000 MHz processor wins by a fairly large margin.

Performance in professional applications

As we can see, AMD Athlon 64 handles professional tasks better.

Performance in real gaming applications

In computer games, AMD Athlon 64 3000+ shows its best side: in almost all applications, Athlon 64 showed better performance than Pentium 4. The exception is gaming applications based on the engine or a modified version of the Quake 3 Arena engine, as the latter is known much better processed by systems with Intel Pentium 4 processors.

Performance in semi-synthetic pseudo DirectX 9 benchmarks

Again, we see the superiority of the platform based on the AMD Athlon 64 3000+ microprocessor.

conclusions

AMD has made a great product. The AMD Athlon 64 3000+ processor delivers excellent performance, often better than its main competitor, the Intel Pentium 4 3000 MHz. There is no need to talk about the absolute superiority of the Athlon 64 3000+, due to the fact that the processor is not good in all tests, for example, in coding tasks and tests with two computational threads, Intel Pentium 4 wins. At the moment there is no 64-bit Windows XP OS and 64-bit applications, AMD has presented a wonderful product that will undoubtedly be a success, given its official (in the AMD price list) price of 212 USD. At this price, a competitor represented by Intel can only offer an Intel Pentium 2800 MHz microprocessor. However, let's look at our Minsk prices: AMD Athlon 64 3000+ costs about 275-280 USD, but what can you buy from Intel for this price? Only Intel Pentium 4 2800 MHz. And if we take into account the prices for motherboards based on VIA K8T800 and for motherboards based on Intel 875P... In my opinion, further words are unnecessary...

The author is grateful to the companygreen line - official product distributorMSIon the territory of the Republic of Belarus for the central processor provided for testingAMDAthlon64 3000+ andsystem boardmicrostarK8T Neo