It just so happened that on the domestic video accelerator market after $ 150 and up to $ 250+ the deepest failure ensues. In other words, there is simply NOTHING to buy in the range of 1300-2000 UAH, apart from of course old stuff like 260GTX, which, although good, is absolutely morally outdated and is prohibitively expensive. And the fault is not only the glorious state of Ukraine ... But the situation is changing rapidly!
The components for testing were as follows: 
The site already has a number of articles on how desperately fussing Nvidia filling the gaps in mid-range and cheap deals and store shelves as AMD Radeon they also do not sleep and will be released from day to day Radeon 6870 and its trimming.
So - a somewhat stripped-down version entered the market. GTX460, which has a 192-bit memory bus and its size is accordingly reduced to 768MB. But the point is that the architecture (the number of shader units, CUDA cores in the first place) practically did not suffer even in comparison with the older members of the family. And some bus cut can be forgiven due to the fact that DDR5 memory is installed and the final throughput practically did not suffer. And the competitor to this video card is positioned Radeon 5770 ... Let's compare these two cards:
Specifications table
| Video cards | GeForce 8800 GTX |
GeForce 9800 GTX |
GeForce GTX 260 |
GTX 460 768MB |
Radeon HD 4850 |
Radeon HD 5770 |
| GPU | G80 | G92b | GT200b | GF104 | RV770 | RV840 Juniper |
| Process technology, nm | 90 | 55 | 55 | 40 | 55 | 40 |
| Number of transistors, mln. Pcs. | 681 | 754 | 1400 | 1950 | 956 | 1040 |
| Universal processors, pcs | 128 | 128 | 216 | 336 | 800 | 800 |
| Texture blocks, pcs | 32 | 64 | 72 | 56 | 40 | 40 |
| ROP, pcs | 24 | 16 | 28 | 24 | 16 | 16 |
| GPU clock frequency, MHz | 576 | 675 | 575 | 675 | 625 | 850 |
| Shader domain clock frequency, MHz | 1350 | 1688 | 1240 | 1350 | 625 | 850 |
| Memory bus, bit | 384 | 256 | 448 | 192 | 256 | 128 |
| Memory type | GDDR3 | GDDR3 | GDDR3 | GDDR 5 | GDDR3 | GDDR 5 |
| Memory size, MB | 768 | 512 | 896 | 768 | 512 | 1024 |
| Memory clock frequency, MHz | 1800 | 2200 | 2000 | 3600 | 2000 | 4800 |
| release date | 2006 | 2008 | 2008 | 2010 | 2008 | 2009 |
| Cost for September 2010, UAH | *~600 | *~700 | *~1100 | 1500 | *~650 | 1350 |
* such video cards are discontinued, but they can be purchased at flea markets for approximately the indicated price.
I would like to draw your attention to the GTX260 right away. Although this video card is powerful enough, even though it has a 448-bit bus, I personally would not take it for a number of reasons:
- huge dimensions
- high power consumption
- lack of support for DX11
- insufficient shader power (only 216 shaders)
- overpriced (they manage to offer new copies for UAH 1800 each!)
At the same time, the hero of today's review has as many as 336 shaders, which is more than 1.5 times more. Only the 192-bit bus pumped up, but the presence of high-frequency DDR5 well compensates for this disadvantage, and the competitor Radeon 5770 it is actually 128 bits.
But the most main characteristic apart from performance, it is the price. And we see that "For the money" c Radeon 5770 only competes GTX460 768Mb... So let's compare these two video cards head-on and find out how it behaves Radeon 5770 1024MB versus GTX460 768MB in games ..
Let's look at the test results in three modern games.
As we can see GTX460 successfully defeats both the old guard from its own camp and competitors from AMD. In particular Radeon 5770 ... The game actively uses DirectX 11 tessellation, in which the cards Nvidia very strong. Therefore, this game does not need a large memory bus, but, first of all, a high computing power (including geometric) of the chip. The 336 shaders on the GTX 460 do the trick.
The situation is repeated in MAFIA 2. New map from Nvidia still compares favorably with Radeon 5770 and all other competitors.
METRO 2033 is perhaps the most challenging video card game of our time. There is no point in even commenting on something - the situation repeated itself. Moreover, over time, the lag of other video cards will be even more aggravated in new games. Therefore, the video card Nvidia GTX 460 768MB is by far the best in its price range, and the fact that its memory is 256MB less than gig competitors is not a hindrance to it. Moreover, a lot of variants of GTX 460 appeared (which testifies to the popularity of the card), among which I personally allocate the GeForce GTX 460 SE 1024MB for the required money - this is a real hit, cheap and devoid of problems with a narrow bus and lack of video memory.
What did NOT please in the GTX 460 .. The cooling system of the version with 768MB of video memory did not please. Despite the fan and a hole for blowing out of the case, due to the absence of a normal casing, the air was blown out of nowhere. Under heavy loads, the card happened to freeze after 20-30 minutes. But the 1024MB version turned out to be all right. Have Nvidia in general, now there are problems and scandals with the cooling of their video cards. Nevertheless, you can and should find good versions, since there are many offers from manufacturers of all stripes on the market and you can buy a video card to your taste.
The incredibly hot summer of 2010 brought on its hot wings not only drought, smoke from burning peat bogs and forty-degree heat. In July of this year, Nvidia announced a powerful mid-price player, which the former was sorely lacking in order to close the big gap between the GeForce GTX 470 and the GeForce GTS 250.
The recently released GTX 465, although it turned out to be a rather powerful and productive solution, did not differ at all at an attractive price and an acceptable level of heat dissipation. As a heavily stripped-down version of the GTX 480, the card retains all the positive and negative features of its older brother, including good DirectX 11 performance, and with it a hot temper and huge appetites in terms of power consumption. The stocks of defective GF100 are not unlimited, and something had to be done with such thermal characteristics.
And now the Californians are announcing a complete replacement for the GeForce GTX 465 video card, based on a completely different chip - GF104, with a cooler and more economical disposition and even slightly higher performance. An inexpensive PCB and a less sophisticated power subsystem, as well as a relatively inexpensive GPU in production, did their job - video cards based on the GeForce GTX 460 were born not only powerful and cold, but also quite inexpensive. Compared to a potential competitor, the Radeon HD 5830, the GeForce GTX 460 looked more than decent, delivering on average more performance at a comparable price.
However, an incredibly large number of former middle-class leaders, both GT200b and G92b-based, are prevalent today. And if the advantage of the GeForce GTX 460 over the GeForce 9800 GTX or 8800 GTS is obvious, then it is not so easy to predict the absolute advantage of the former over the GeForce GTX 260 or GeForce 9800 GX2 video cards. Considering that the GeForce GTX 2x0 family has successfully competed even with the Radeon HD 5800 line for a long time, it can be assumed that these video cards still have a margin of safety, and they may well provide a level of comfortable play similar to the GTX 460 in a number of modern games.
In addition, there are many owners of video cards of the Radeon HD 4800 line, which can rightfully be called the “Renaissance” series of ATi's former glory, which the division lost after the release of the Radeon HD 2x00 video adapters. At one time, the Radeon HD 4870 went with the GTX 260 on an equal footing, overtaking the version with 192 shader processors and slightly yielding to the forced version with their increased number, and the Radeon HD 4850 successfully competed with almost all implementations of Californian boards based on the G92b.
In other words, users still have a very large "running" fleet of video cards in their hands. different generations and architectures that, nevertheless, give a light to, and their speed is enough for almost any setting and the highest resolutions. Based on the results of this testing, the owners of “last year's” motherboards will be able to make a conclusion for themselves about the advisability of buying a new video card and decide whether the purchase of a GTX 460 will be so justified, and how much the “old guard” will lag behind it.
Testing Heroes
A little preface. All test participants will be tested in two modes - working at the standard factory frequencies and in overclocking under a water cooling system (CBO). Thanks to it, you can forget about the temperature limits set by air coolers and overclock video cards to the maximum. At the same time, due to some circumstances (mainly due to the fact that most video cards do not belong to the author), voltmod will not be carried out. The use of CBO is only necessary here to provide “ideal cooling conditions” to achieve a higher overclocking potential than under air CO.
GeForce 8800 GTX
Without a doubt, the hero of his time. By releasing at the end of 2006 a truly powerful solution based on the revolutionary G80 chip, Nvidia felt like a leader in 3D graphics like never before. A sophisticated graphics core with a shader architecture scrapped the once mighty Radeon X1950 XTX and GeForce 7950 GX2 at the moment, delivering twice the performance for the same price.
However, the performance of the GeForce 8800 GTX was directly proportional to its power consumption and heat dissipation - the video card turned out to be very hot and quite often failed due to the BGA splicing of the GPU balls or the NVIO chip.
Nevertheless, we must pay tribute to the GeForce 8800 GTX: for almost two years it remained one of the most powerful single video cards and, until the HD 4870 and GTX 280 approached, it was inferior only to the dual-processor GeForce 9800 GX2. Now the 8800 GTX is a frequent visitor to flea markets, and is sold at very low prices. Many users bought it for a second home PC or used it as the main video subsystem: the performance of the 8800 GTX is still sufficient in many games and applications.
Gainward GeForce 8800 GTX 768 MB
Recently tested hero and former king of 3D graphics. The native cooling system outlived its obsolete, so it was replaced by a powerful CO Zalman GV1000.
GPU frequency - 576 MHz, shader domain - 1350 MHz, memory - 1800 MHz.
Overclocking
Not the most obeyed the video card high frequencies, despite active cooling cold water. Since the memory chips were left without heatsinks and were blown only air flow, it overclocked worse than other components of the video card - only “plus” 90 MHz. The GPU ran stably at 675 MHz and the shader domain at 1566 MHz. Getting old, what can I say ...
Geforce 9800 GTX
Video cards based on the GeForce 9800 GTX have not received a very good reputation. Being essentially renamed and slightly overclocked versions of the GeForce 8800 GTS 512 MB, they were sold at too high a price and were nothing new globally compared to the previous flagship, turning out to be even a little slower in some tests. Of course, there was no talk of any replacement for the GeForce 8800 GTX, and AMD, who was advancing on the heels with its aggressive pricing policy, made it clear that its product based on the Radeon HD 4850 was faster and cheaper.
The GeForce 9800 GTX itself had a certain demand for six months before the HD 4800 line appeared and had a fairly high performance. Its subsequent reincarnation as the GeForce GTS 250 gave G92b-based video cards a third wind, and gradually falling prices made it one of the most popular and inexpensive graphics cards on the market.
Why is it the 9800 GTX taking part in the tests, and not the 8800 GTS 512 or GTS 250? The first graphics accelerator can be considered the “arithmetic mean” of these three video cards, so it will demonstrate average overall results on behalf of G92b-based motherboards.
Asus GeForce 9800 GTX 512 MB
The Asus GeForce 9800 GTX graphics card runs at overpriced clock frequencies and the factory overclocking is to blame.
For today's testing, the frequencies were artificially lowered to the recommended ones:
GPU frequency - 675 MHz, shader domain - 1688 MHz, memory - 2200 MHz.
Overclocking
The 9800 GTX overclocked very well - 849/2052/2484 MHz. Decent result.
GeForce 9800 GX2
“Leviathan” 9800 GX2 gained records even before its appearance. Its heavy cooling system made the total weight of the video card so high that it was not recommended to transport it, being installed in ready-made cases - it could rip the PCI-E slot from the motherboard. As a dual version of the GeForce 8800 GTS 512 MB, the video card demonstrated incredible speed results, noticeably outperforming the once powerful Radeon HD 3870x2 and raising the FPS bar to a level unattainable for other video cards.
However, the Californians set themselves up for themselves, because the single-chip GeForce GTX 280 that appeared six months later did not always turn out to be faster and walked on an equal footing with the "old man". Naturally, everyone expected much more from the powerful monolithic flagship, which is a really new and truly updated successor to the GeForce 8800 GTX, but the GeForce 9800 GX2 turned out to be too powerful a solution, which even its older brother found it difficult to defeat.
Today, the GeForce 9800 GX2 is not the most frequent guest in users' computers. But if at one time he plugged the GeForce GTX 280 into the belt, then we can assume that he will be able to compete even with such a powerful newcomer as the GeForce GTX 460.
Asus GeForce 9800 GX2 1024 MB
The video card is made on the reference design, but received a slight overclocking, which was removed to the recommended clock frequencies. Total values:
GPU frequency - 600 MHz, shader domain - 1500 MHz, memory - 2000 MHz
Overclocking
Unfortunately, in view of design features It was impossible to install a regular water block in the GeForce 9800 GX2. As a result, this is the only participant in the test whose CO has not been replaced. Despite this, under the reference air cooling, the video card overclocked excellently, reaching frequencies of 774/1944/2376 MHz.
GeForce GTX 260 216 sp
First version GeForce video cards GTX 260 turned out to be really fast: 192 stream processors, 448-bit memory bus, 896 MB DDR3 memory - incredibly productive filling by the standards of 2008. But stiff competition from the Radeon HD 4870 forced Nvidia to lower prices for the GTX 260, and soon release an improved version with more execution units.
The forced GeForce GTX 260 with 216 stream processors, which appeared literally six months later, turned out to be slightly faster than the HD 4870 and also became widespread due to its low price, availability, quiet CO and excellent performance.
As the name suggests, the GTX 460 claims exactly to replace the GeForce GTX 260, thus, in this testing we will clearly see how much a newcomer is better than his predecessor, and whether the difference is so significant to change a well-proven “veteran” for a fresher and more muscular brother.
Palit Sonic GeForce GTX 260 216 sp 896 MB
This sample, which came to us after being overclocked by the manufacturer, was returned to the reference frequencies and artificial overclocking was removed.
Final clock frequencies:
GPU frequency - 575 MHz, shader domain - 1240 MHz, memory - 2000 MHz
Overclocking
The overclocking potential of the GTX 260 is simply magnificent - the impressive 790/1704/2404 MHz have resigned to the video card. The memory took on such high frequencies due to the fact that the factory heatsink remained on it, which was simply impossible to remove: it was attached to the memory and the NVIO chip with hot melt glue and, despite all efforts, did not budge.
INTRODUCTION Anyone with any interest in the state of the consumer 3D graphics industry is aware that Advanced Micro Devices' leadership in the single-processor single-GPU gaming class has been successfully challenged by Nvidia with the GeForce GTX 480, however, in the category of cards equipped with more than one GPU. the ATI Radeon HD 5970 still reigns supreme, carrying a pair of RV870 "Cypress" chips. Before it, the GeForce GTX 480 falls behind, which, taking into account the monstrous lag in the texture sampling rate, is not surprising. Chances of winning Nvidia solutions in this situation, simply no, and in terms of other technical characteristics, the brainchild of the former ATI Techologies is ahead of its rival. But Nvidia will certainly want to challenge AMD in this sector, so there is no question of creating and releasing a dual-processor card with Fermi architecture - it's just a matter of time and technical feasibility.
Until recently, it was with the last point that Nvidia had serious problems, since even in the GeForce GTX 470 SLI version, the hypothetical dual-processor GeForce GTX 490/495 turned out to be a cumbersome, hot and extremely uneconomical solution, and even about combining a pair of GeForce GTX 480 on one board could be out of the question. Yes, a pair of GeForce GTX 470s operating in SLI mode provided the required level of performance, however, for the above reasons, this did not yet open a direct path to the GeForce GTX 490/495. But now that Nvidia has a new, less sophisticated and more economical core, the GF104, the situation has changed radically - this relatively inexpensive GPU is suitable for creating a dual-GPU gaming card. top class much better than the monstrous GF100. As our tests have shown, at sufficiently high clock speeds, in some cases it is able to compete with the GeForce GTX 470, but even at the factory frequencies, the GF104 pair combined in an SLI tandem looks very impressive, at least in theory:
If we assume that the hypothetical GeForce GTX 490/495 will be based on the GF104 cores, then, of course, this option will be inferior to the previously reviewed version, similar in characteristics to the GeForce GTX 470 SLI tandem. Nevertheless, it will surpass a single GeForce GTX 480 by a head at a comparable or even lower cost and will look very good even in comparison with the Radeon HD 5970, seriously inferior to the latter only in texture processing speed. The latter is the trouble with all solutions using the Fermi architecture, due to the fact that when designing the corresponding graphics processors, Nvidia decided to save on "pure" power of the TMU subsystem in favor of advanced geometry processing capabilities. As practice shows, this does not prevent solutions based on the GF100 and GF104 from feeling good in modern games, and even more so, the presence of powerful tessellation units will not prevent them from playing the next generation games that actively use this feature.
True, if the cost of a card equipped with two GF104 chips is comparable to the cost of the GeForce GTX 480 (16 thousand rubles), this will make the latter unpopular among potential buyers, but so far the GeForce GTX 490/495 does not exist in the form a separate card, it is too early to make any predictions on this score, especially since solutions based on the GF100 have some reserve in the form of the ability to activate all 512 stream processors. One thing is clear - this option will be significantly cheaper than the version with two GF100s, as a result of which AMD will have to actively reduce prices, since even now they often ask for the Radeon HD 5970 more than 20 thousand rubles... Of course, the GeForce GTX 490/495, even at a cost of about 16-17 thousand rubles, by definition, will not be able to make a solid sales volume - such is the fate of all expensive discrete cards - but its appearance will certainly help Nvidia gain popularity among demanding players who do not want to mess with discrete tandems SLI or CrossFire. In the future, we can assume the appearance of a card with two GF104 cores on board, working in a full configuration with 384 stream processors and 64 TMUs, but this is only speculation so far; however, we can assume that such a monster will receive the name GeForce GTX 495, while the GeForce GTX 490 can be built on the basis of a pair of truncated GF104s.
Exactly last option We're going to test it in real life today using a pair of GeForce GTX 460 1GB running in SLI mode. One of the components of this tandem will be the Zotac GeForce GTX 460 1GB (model ZT-40402-10P), which we are going to acquaint our readers with.
Zotac GeForce GTX 460 1GB: packaging and contents
The packaging in which Zotac delivers its products to retail chains does not differ in a variety of shapes and originality of design. Their advantage lies elsewhere - on the boxes there is always a window through which you can see the video adapter without taking it out, and, therefore, without opening the seals. The buyer immediately sees what exactly he will be dealing with, and this, in our opinion, is a noticeable advantage of such packages:
This time, in addition to the basic information about the product and the mention of the extended Zotac brand warranty, there is also a sticker on the box informing that the full version of Prince of Persia: The Forgotten Sands is included in the package.
As usual, the inner space of the box is occupied by a polyurethane foam pallet, in which grooves are cut for the video adapter itself and the accompanying set of accessories. The card is securely fixed in its box and reliably protected from minor twists and turns that might befall it during transportation and storage; from serious trouble, of course, no reasonable protection will save. The Zotac GeForce GTX 460 1GB video adapter comes with the following set of accessories and accessories:
In our opinion, the package bundle of the Zotac GeForce GTX 460 1GB can be assessed as good: it contains everything you need to fully operate the described card. In addition, the buyer will receive as a gift full version sequel to the famous Prince of Persia series. Recently, a rare manufacturer graphics cards spoils customers with free games, so Zotac can only be praised for a pleasant surprise.
In general, the packaging and equipment of the Zotac GeForce GTX 460 1GB do not give rise to any complaints. Let's take the card out of the box and look at it in more detail, especially since in our practice this is the first copy of the GeForce GTX 460 1GB using a reference PCB design developed by Nvidia itself.
Zotac GeForce GTX 460 1GB: design and specifications
The reference version of the GeForce GTX 460 1GB is very similar to all previous Nvidia designs of a fairly high-end and uses a cooling system with a radial turbine and the release of heated air outside the system case.
The card looks more massive than the previously described Gainward GeForce GTX 460 GS GLH, and surpasses it in length; however, the difference is not fatal and in most cases Zotac GeForce GTX 460 1GB will be installed without problems. Unless in models with a shorter base, there may be problems with the connection of power connectors - here they are located on the end side of the printed circuit board, and not on the top, as in Palit / Gainward products.
The PCB layout in general terms resembles the layout of the aforementioned Gainward card - one of the memory chips is located in the same proud isolation from the others, next to the DVI connectors. The power subsystem is built according to the "3 + 1" scheme, the three-phase GPU power regulator is controlled by the ON Semiconductor NCP5388 NCP5388 chip, which is also used in the design of the GeForce GTX 470 and GeForce GTX 465.
MSI's Afterburner utility supports this chip, however, the upper voltage limit for the graphics core is fixed at 1.087 Volts. This limitation can only be removed by firmware modified BIOS version, but, unfortunately, the attempt we made was unsuccessful, and in order to return the card to operability, we had to flash the original BIOS back. The Anpec Electronics APW7165 microcircuit is responsible for the power supply of the memory; it has also been encountered earlier in our practice - in particular, in the design of the GeForce GTX 465. Like the Palit design, the reference design of the Nvidia GeForce GTX 460 provides for connection external power supply via two six-pin PCIe 1.0 connectors.
Memory marking does not open anything new - we have before us all the same K4G10325FE-HC05 microcircuits manufactured by Samsung Semiconductor, having a capacity of 1 Gbit (32Mx32) and designed for a frequency of 1000 (4000) MHz. But, unlike the Gainward GeForce GTX 460 1GB GS GLH, the memory frequency corresponds to the reference value and is 900 (3600) MHz. Again, by the standards of AMD / ATI solutions, this figure does not look serious, at least in combination with a 256-bit access bus. The total amount of local video memory is naturally 1024 MB, which is enough for any modern games and applications. In power-saving modes, the memory frequency can be reduced to 324 (1296) MHz and to 135 (540) MHz, depending on the task - if the card works with typical office applications, the more economical second mode is used, and in the case of high-resolution video decoding, the first one. In this sense, the Zotac GeForce GTX 460 1GB is no different from the previously reviewed Gainward GeForce GTX 460 1GB GS GLH.
The Zotac GeForce GTX 460 1GB has an Instance of GF104 revision A1, produced on the 23rd week of 2010. It operates in the maximum currently available configuration: 336 ALUs, 56 TMUs and 32 RBEs; the clock speed of the graphics core meets the official specifications and is 675 MHz for the main part of the GPU and 1350 MHz for the compute domain. In power saving modes, the frequencies are reduced to 405/810 MHz or to 51/101 MHz. The core supply voltage in 3D mode is 0.987 Volts, in the first energy-saving mode it, accordingly, decreases to 0.912 V, and in the most economical mode - to 0.875 V. more extreme values require BIOS modification. Note also that the current version of GPU-Z still incorrectly displays the parameters of cards based on GF104, in particular, the number of active shader processors. Having seen the number "224", you shouldn't be afraid that you got a defective card - in fact, 336 ALUs out of the 384 physically available on the die are working.
The interface configuration of the Nvidia reference card is better suited to modern realities than the one that was implemented in its Gainward product - all the demanded digital interfaces are present here, including DisplayPort, and support analog connection D-Sub can be easily realized using a standard adapter. There is also one MIO interface connector, which serves to combine a pair of GeForce GTX 460 in tandem SLI. The hero of today's review worked without any problems in this mode and passed the full cycle of tests without a single failure.
The design of the Nvidia GeForce GTX 460 reference cooling system is simple and nothing new, tracing its lineage to the GeForce 8800 GTX cooling system. True, it uses the "direct contact" technology, which implies direct contact of heat pipes with the GPU heat spreader cover, but we have already seen this in the design of the GeForce GTX 480/470/465. Oddly enough, any spacers providing thermal contact of the base with memory chips are not provided for by this design; the power transistors of the power stabilizers are also devoid of such gaskets.
The fan markings can also be seen here. The Protechnic MGT8012YB-W20 model is well known to everyone who prefers Nvidia solutions, since almost all reference cooling systems developed by this company are equipped with it. Depending on the rotation speed, this fan can be either very quiet or compete in volume with a household vacuum cleaner.
The radiator itself has rather modest dimensions and, accordingly, not too large a heat discharge area. In any case, it looks rather modest against the background of the Gainward / Palit design. Yes, such a radiator will be enough to remove heat from the GF104, but you can't count on a particularly comfortable temperature regime; however, this assumption will be verified in the next chapter of the review. We believe that a more solid heatsink should have been used in the design of the Nvidia GeForce GTX 460, despite the fact that the GF104, unlike its older brother, is not at all a fire-breathing monster. Also, in our opinion, the developers should not have used the "two-story" layout of the DVI connectors - due to this, the number of slots in the mounting plate was reduced, and it is through them that the cooling system removes heated air outside the system case. How this will affect the cooling efficiency of the reference version of the GeForce GTX 460 1GB, read below.
Power consumption, thermal conditions, noise and overclocking
Since the previously obtained data on the level of power consumption of cards based on GF104 was obtained using a non-standard model GeForce GTX 460 1GB, we double-checked them using a Zotac card and a test platform already familiar to our readers with the following configuration:
Intel Core 2 Quad Q6600 Processor (3 GHz, 1333 MHz FSB x 9, LGA775)
DFI LANParty UT ICFX3200-T2R / G (ATI CrossFire Xpress 3200) Motherboard
PC2-1066 memory (2x2 GB, 1066 MHz)
Power supply unit Enermax Liberty ELT620AWT (power 620 W)
Microsoft Windows 7 Ultimate 64-bit
CyberLink PowerDVD 9 Ultra / "Serenity" BD (1080p VC-1, 20 Mbps)
Crysis warhead
OCCT Perestroika 3.1.0
We remind you that this stand is equipped with a special measuring module described in the review “ Power consumption of computers: how many watts do you need?". Its use allows you to get the most complete data on electrical characteristics ah modern graphics cards in different modes... As usual, the following tests were used to load the video adapter in various modes:
CyberLink PowerDVD 9: FullScreen, Hardware Acceleration Enabled
Crysis Warhead: 1600x1200, FSAA 4x, DirectX 10 / Enthusiast, frost card
OCCT Perestroika GPU: 1600x1200, FullScreen, Shader Complexity 8
For each mode, with the exception of the OCCT limit load simulation, measurements were taken for 60 seconds; In order to avoid card damage due to overloading of power circuits, for the OCCT: GPU test, the test time was limited to 10 seconds. Using this technique, the following data were obtained:
If in the desktop mode the Nvidia and Palit / Gainward designs behave in a similar way, demonstrating a peak power consumption level of about 20 watts, then quite significant differences begin later. In particular, when performing tasks that are not too difficult for a modern GPU, such as decoding high-definition video, Nvidia's reference design demonstrates significantly worse efficiency, at least at its peak. After switching to the power-saving mode with lower clock frequencies, the average power consumption level is about 23 W, which is comparable to the indicators of the Gainward card, however, the mode switching itself is not so fast. In modern games, the Nvidia reference card also consumes a little more, but the difference is less than 10 watts, and of course it does not do the weather. Note that in this mode, the first power connector is loaded significantly more than the second - the current flowing through it can reach 7.4 A, while the second connector has a maximum of only 3 A. Theoretically, one eight-pin connector could be enough. power supply PCIe 2.0, for which the standard allows power up to 150 W, but the GeForce GTX 460 will be installed, including in not the most expensive systems whose power supplies may or may not have these graphics card power connectors.
Thus, it is obvious that re-examination of the electrical characteristics of the GeForce GTX 460 1GB has not revealed anything extraordinary to us and our readers. In general, the figures are quite acceptable for the level of performance demonstrated by solutions based on the GF104, although they are inferior to those of Advanced Micro Devices solutions.
A brief test of the ability of the reference Nvidia cooling system to cope with its work showed that it does it worse than the solution proposed by Gainward / Palit, which is not surprising, given the modest heat dissipation area of the radiator and the difficult release of heated air outside the system case by the "two-deck" DVI connector ... However, it cannot be said that this is a failure - even in hot weather, when the tests were carried out, the card worked absolutely stable. If you are not fond of overclocking, you do not need to worry, but for those who like overclocking, we can only recommend changing the reference Nvidia cooler to something more efficient.
Alas, the modest capabilities of the radiator, coupled with the difficult air exchange, led to a natural result - the reference cooling system Nvidia GeForce GTX 460 1GB under no circumstances can be called silent. In 3D mode, it successfully competes with the flagship of the line, the GeForce GTX 480, but we did not manage to achieve silence in the desktop mode either. The excuse can be said about hot weather during the tests, however, Gainward was able to release a much quieter version of the GeForce GTX 460 1GB with much higher clock speeds of the graphics core, so the excuse is not very convincing. The advice for lovers of silence will be the same as for overclockers - get rid of the reference cooling system or immediately purchase the version of the GeForce GTX 460 equipped with a non-standard cooler.
With overclocking, however, everything turned out to be not as bad as we thought - the Zotac GeForce GTX 460 1GB graphics processor worked steadily at a frequency of 820/1640 MHz, and the memory reached 1065 (2130) MHz. This is a little more than what the previously described Gainward GeForce GTX 460 GS GLH was capable of, however, 20 MHz in the frequency of the graphics core is too insignificant an increase to carry out a separate test cycle for the sake of it, and the performance level of the GF104 operating at 800/1600 MHz is already known to us and to our readers. In addition, the main purpose of this review is a study of the performance of GeForce GTX 460 1GB when working in tandem SLI. We proceed to describe the results of this study.
Test platform configuration and performance testing methodology
Testing the gaming potential of the GeForce GTX 460 SLI tandem was carried out on a universal test platform with the following configuration:
Intel Core i7-975 Extreme Edition Processor (3.33 GHz, 6.4 GT / s QPI)
Cooler Scythe SCKTN-3000 "Katana 3"
Gigabyte GA-EX58-Extreme motherboard (Intel X58)
Memory Corsair XMS3-12800C9 (3x2 GB, 1333 MHz, 9-9-9-24, 2T)
Hard Samsung drive Spinpoint F1 (1 TB / 32 MB SATA II)
Ultra X4 850W Modular Power Supply (850W)
Dell 3007WFP Monitor (30 ”Max Resolution [email protected] Hz)
Microsoft Windows 7 Ultimate 64-bit
The following versions were used ATI drivers Catalyst and Nvidia GeForce:
ATI Catalyst 10.7 for ATI Radeon HD
Nvidia GeForce 258.96 WHQL to Nvidia GeForce
The drivers themselves were configured as follows:
ATI Catalyst:
Smoothvision HD: Anti-Aliasing: Use application settings / Box Filter
Catalyst A.I .: Standard
Mipmap Detail Level: High Quality
Wait for vertical refresh: Always Off
AAMode: Quality
Nvidia GeForce:
Texture filtering - Quality: High quality
Vertical sync: Force off
Antialiasing - Gamma correction: On
Antialiasing - Transparency: Multisampling
CUDA - GPUs: All
Set PhysX configuration: Auto-select
Ambient Occlusion: Off
Other settings: default
The test package includes the following games and applications:
3D first-person shooters:
Aliens vs. Predator (1.0.0.0, Benchmark)
Battlefield: Bad Company 2 (1.0.1.0, Fraps)
Call of Duty: Modern Warfare 2 (1.0.182, Fraps)
Crysis Warhead (1.1.1.711, Benchmark)
Far cry 2 (1.03, Benchmark)
Metro 2033 (1.0.0.1, Fraps)
S.T.A.L.K.E.R .: Call of Pripyat (1.6.02, Fraps)
3D shooters with a third person view:
Just Cause 2 (1.0.0.1, Benchmark / Fraps)
Resident Evil 5 (1.0.0.129, Benchmark)
RPG:
Mass effect 2 (1.01, Fraps)
Simulators:
Colin McRae: Dirt 2 (1.1, Benchmark)
Tom Clancy "s H.A.W.X. (1.03, Benchmark)
Strategy games:
BattleForge (1.2, Benchmark)
StarCraft II: Wings of Liberty (1.0.2, Fraps)
World in Conflict: Soviet Assault (1.0.1.0, Benchmark)
Semi-synthetic and synthetic tests:
Futuremark 3DMark Vantage (1.0.2.1)
Final Fantasy XIV Official Benchmark (1.0.0.0, Fraps)
Unigine Heaven Benchmark (2.0)
Each of the test set software games has been tuned to provide the highest possible level of detail. A fundamental refusal to manually modify any configuration files means that only the means available in the game itself to any uninitiated user were used for customization. The tests were carried out at resolutions 1600x900, 1920x1080 and 2560x1600. Except where otherwise noted, the standard 16x anisotropic filtering was complemented by 4x MSAA anti-aliasing. Anti-aliasing was activated either by means of the game itself, or, in their absence, it was forced using the appropriate settings of the ATI Catalyst and Nvidia GeForce drivers.
In addition to the GeForce GTX 460 SLI tandem, the following solutions took part in the testing:
Nvidia GeForce GTX 470 SLI
Nvidia GeForce GTX 480
Nvidia GeForce GTX 460 1GB
ATI Radeon HD 5970
ATI Radeon HD 5870
To obtain performance data, we used the testing tools built into the game with the obligatory use of the original test videos, and, if possible, recording the data on the minimum performance. In the absence of the aforementioned tools, the Fraps 3.1.2 utility was used in manual mode with a three-fold test pass, fixing the minimum values and then averaging the final result.
Game Tests: Aliens vs. Predator
Testing in this game is carried out in DirectX 11 mode with maximum settings quality graphics.
The new tandem started very briskly, outperforming the GeForce GTX 470 SLI in all resolutions - most likely due to the higher clock speeds of the GPU. The advantage is small, but it suggests that Nvidia's new dual-processor card will be based on a pair of GF104 chips. By itself, the performance level of such a solution is high enough to use a resolution of 1920 × 1080, but, alas, not 2560 × 1600, but it is precisely for the sake of extreme resolutions that AMD and Nvidia's flagship solutions are often purchased.
Game tests: Battlefield: Bad Company 2
Of course, it's hard for a pair of GF104s to compete with a pair of GF100s, albeit working in a truncated configuration, especially where clock frequencies are not everything. But the lag is as small as the superiority in the previous test, which means that dual-processor cards based on the GF104 are likely to be! Moreover, the performance of such a solution will allow using a resolution of 2560x1600 or the forthcoming 2560x1440 without regard to detail settings.
Game tests: Call of Duty: Modern Warfare 2
At 2560x1600, the described tandem behaves in the same way as a regular GeForce GTX 480, and a pair of GeForce GTX 470s operating in SLI mode does not look much faster. In other words, there is no point in using such powerful dual-processor solutions in this game - a more or less powerful single-processor graphics card, for example, the Radeon HD 5870, is enough. Moreover, a single GeForce GTX 460 1GB can also handle it, although there is some reserve there is practically no durability at 2560x1600.
Game tests: Crysis Warhead
The GeForce GTX 460 SLI system has blundered - and exactly where more was expected of it, that is, at ultra-high resolutions. It demonstrated average performance at the level of a single GeForce GTX 480 and twice as low as the minimum, which, of course, makes the use of such a tandem meaningless. But in resolutions up to 1920 × 1080 inclusive, everything is fine, especially with the minimum performance, with which the hypothetical GeForce GTX 490 is much better than the Radeon HD 5970.
Game tests: Far Cry 2
Unlike Crysis, in Far Cry 2, the GeForce GTX 460 SLI looks very promising against the background of the Radeon HD 5970, even at 2560x1600. We did not find any problems with image quality or control lags, and a comfortable level of performance was maintained even in difficult scenes; however, Far Cry 2 is no longer the leader in system requirements. It is also worth noting an extremely insignificant lag behind the GeForce GTX 470 SLI. Apparently, we will not see cards based on a pair of GF100, unless Asus decides to release another unique monster in a limited edition, but dual models based on GF104, especially with 384 unlocked shader processors, will represent a very serious force.
Game tests: Metro 2033
This game is tested without anti-aliasing, since enabling MSAA only degrades the quality of textures, while at the same time severely degrading performance.
But in Metro 2033, the Radeon HD 5970 managed to stay in the lead, and this is especially noticeable in the 2560x1600 resolution. At least, if we are not talking about the inclusion of tessellation, which is not very noticeable in dark post-apocalyptic dungeons. Here Nvidia will need something more powerful in order to compete on an equal footing with the AMD / ATI flagship. As already noted, most likely it will be a tandem of the GF104, working in the highest possible configuration.
Game Tests: Game Tests: S.T.A.L.K.E.R .: Call of Pripyat
In this test, DX10.1 and DX11 modes are used for cards with the corresponding capabilities.
In the first two resolutions, the tested tandem cannot outperform the GeForce GTX 470 SLI, but it is not inferior to it, but at 2560x1600 it unexpectedly loses its positions. The minimum performance in this mode is less than 20 frames per second and it becomes impossible to play. This is unlikely to be due to the characteristics of the texture subsystem, since it is, at least, not worse than the similar subsystem of the GeForce GTX 470 SLI tandem, and in theory even better due to the higher frequency of the GPUs. Perhaps this is due to the lower performance of the computing part, but then the situation can be corrected only by unlocking all 384 ALUs in the GF104. But the Radeon HD 5970 does not suffer from a lack of computing power, so it feels pretty confident even at 2560x1600; in any case, you can play in this mode quite successfully.
Game tests: Just Cause 2
The integrated test tool does not display minimum performance information.
The tandem GeForce GTX 460 SLI behaved strangely at 1920x1080, yielding to the single GeForce GTX 480. At 2560x1600, it showed a good average result, but the minimum was too low; however, here only the Radeon HD 5970 coped with the task of providing more or less acceptable performance.
Game tests: Resident Evil 5
The results obtained in Resident Evil 5 are also quite ambiguous - here the solution under study was able to outperform the GeForce GTX 470 SLI in two out of three resolutions, but it managed to achieve performance comparable to that of the Radeon HD 5970 only in 2560x1600, but in more low resolutions a pair of GF104s were able to compete on an equal footing only with the lone RV870 on board the Radeon HD 5870.
Game tests: Mass Effect 2
In this test, full-screen anti-aliasing is forced using the technique outlined in the Contemporary Graphics Accelerators in Mass Effect 2 review.
But fans of Mass Effect 2, the numbers we received can please - here only a pair of GeForce GTX 470 can compete with a pair of GeForce GTX 460s, which will cost fans high performance much more expensive. Taking into account the high minimum performance of the solution under study, even at 2560x1600, this is an excellent result.
Game Tests: Colin McRae: Dirt 2
The tested tandem performed excellently, yielding only to the more powerful version based on two GeForce GTX 470. True, at 2560x1600 its performance turned out to be the same as that of the Radeon HD 5970, but taking into account the lower cost of such a solution, even in the form of a pair of separate GeForce GTX 460 1GB is an excellent result.
Game tests: Tom Clancy's H.A.W.X.
For testing, the tools built into the game are used, which do not provide for fixing the minimum indicators. Uses DirectX 10 / 10.1 modes.
Another victory for Nvidia, albeit in a long-established territory - in H.A.W.X. the solutions of this company are almost always leaders in their weight categories. True, the owners of the Radeon HD 5870, not to mention the owners of the Radeon HD 5970, do not have to worry - they do not need to run to the store for a pair of GeForce GTX 460 1GB. Even single modern high-end video adapters are capable of providing a comfortable level of performance in this game even at 2560x1600.
Game tests: BattleForge
For cards supporting DirectX 11, the corresponding mode is used.
Alas, the achievements of the GeForce GTX 460 SLI in BattleForge are quite modest - this is just parity with the Radeon HD 5970 in 1600x900 resolution, though with a good superiority in the minimum performance. In other resolutions, we unexpectedly saw a complete failure with a drop in the minimum performance of the solution under study below all acceptable limits. When the speed can suddenly drop to 5-12 frames per second, there is no need to talk about any gaming comfort. To a lesser extent, this is characteristic of the GeForce GTX 470 SLI tandem, but a single GeForce GTX 480 feels great even in 2560x1600, demonstrating by its example the main advantage of classic single-processor cards - independence from multi-GPU software optimization, which is necessary for modern multi-processor graphics solutions like air ...
Game Tests: StarCraft II: Wings of Liberty
As noted in a review entirely dedicated to the sequel to the legendary StarCraft, the second part of the game cannot boast of an ultra-modern engine packed with advanced special effects; this is clearly not Crysis Warhead. Therefore, the game does not differ in the outrageous requirements for the graphics subsystem. However, when using anti-aliasing, StarCraft II is not as fast as without it, so in extreme resolutions the performance of single-processor solutions based on the RV870 and GF100 may not be enough. Here, the two-processor solutions of the Radeon HD 5970 class and the hypothetical GeForce GTX 490 fall into place, and they work equally quickly. In less resource-intensive modes, the AMD solution turns out to be somewhat more productive, but there are enough single-core graphics cards.
Game tests: World in Conflict: Soviet Assault
It looks like Nvidia's dual-processor solutions have reached their ceiling - the performance of the two tandems based on the GeForce GTX 470 and GeForce GTX 460 1GB are the same at 1600x900 and 1920x1080. In the resolution of 2560x1600 that is relevant for such systems, the hero of today's review successfully competes on an equal footing with the Radeon HD 5970, which is an excellent indicator for a pair of cards with a total cost of less than 15 thousand rubles, because an AMD solution at such a price is simply impossible to find on sale.
Semi-synthetic and synthetic benchmarks: Futuremark 3DMark Vantage
To minimize the impact of the CPU, the "Extreme" profile is applied when testing in 3DMark Vantage, which uses a resolution of 1920x1200, FSAA 4x and anisotropic filtering. For the sake of completeness, the results of individual tests are taken over the entire resolution range.
The result of the GeForce GTX 460 SLI tandem cannot be called a failure, but at nominal clock frequencies it did not manage to overcome the 13000 points bar, unlike the Radeon HD 5970.
In both graphics tests, the SLI tandem under study was able to take the lead only in 1600x900 resolution, in the first test it surrendered to the mercy of the Radeon HD 5970 already at 1920x1080, and in the second, although it was able to maintain parity in this mode, it lagged behind at 2560x1600.
Semi-synthetic and synthetic benchmarks: Final Fantasy XIV Official Benchmark
Since initially the FF XIV Official Benchmark produces a meaningless result in glasses, Fraps is used to obtain data on the performance of graphics cards. The test only supports resolutions of 1280x720 and 1920x1080.
As noted in previous reviews, the FF XIV prefers the ATI Radeon HD 5000 architecture. As a result, the results shown by the GeForce GTX 460 SLI tandem did not surprise us at all. Moreover, in a resolution of 1920x1080, it unexpectedly lost even to a single GeForce GTX 480, although the performance level was quite comfortable.
Semi-synthetic and synthetic benchmarks: Unigine Heaven benchmark
There are not enough stars from the sky, a new tandem in Unigine Heaven. Of course, here he plays in his field, however, all that he manages to achieve is parity with the Radeon HD 5970. This is a good result, especially, we repeat, taking into account the price, however, frankly, in this test from a pair of GeForce GTX 460 1GB operating in SLI mode we expected more.
Zotac GeForce GTX 460 1GB: pros and cons
Advantages:
High performance in its price range
High minimum performance
High performance when performing tessellation
Significantly outperforms Radeon HD 5830
Due to overclocking, it can compete with the Radeon HD 5850
Can compete with Radeon HD 5970 in SLI mode
Wide range of FSAA modes
Improved CSAA / TMAA quality
Minimal FSAA performance impact
DirectX 11, Shader Model 5.0, DirectCompute 11 support
CUDA and PhysX support
Full hardware support for HD video decoding
High-quality post-processing and HD video scaling
HDMI 1.3a support
Dolby TrueHD and DTS-HD Master Audio support
Acceptable level of power consumption and heat dissipation
Full version of Prince of Persia: The Forgotten Sands included
Flaws:
Low efficiency of the reference cooling system
Increased noise level
Conclusion
So, to reiterate the main thesis made in today's review: with the release of GF104, Nvidia has a graphics core that is almost ideal for creating a powerful dual-processor graphics adapter... The previous solution of the new generation, GF100, was extremely poorly suited for this role due to low efficiency, high heat dissipation and wide external memory bus, which significantly complicates the PCB layout for such a two-processor monster. The GF104 lacks these disadvantages. Since a dual-processor card based on a middle-class core does not yet exist in nature, we tested its analogue using a pair of GeForce GTX 460 1GB operating in SLI mode, because, in fact, any two-GPU Nvidia GeForce model is such a tandem. What did we find out as a result of this study?Let's look at the results obtained in detail using pivot diagrams:
With the scalability of the GeForce GTX 460 1GB, the order is complete - on average, the performance gain from enabling the SLI mode is 75%, and at times it turns out to be more than 100%. With the confrontation between the Radeon HD 5970, the situation is not so clear: although we can speak of an average advantage of about 10%, in some tests the tested tandem lost to AMD's solution, though not too much, with the exception of Resident evil 5... Taking into account the cost advantage, the start is very good, but the resolution 1600x900 is not the arena where the battles of such monsters as the Radeon HD 5970 and the hypothetical GeForce GTX 490 take place.
And in the resolution of 1920 × 1080 the GeForce GTX 460 SLI tandem does not look so confident anymore - the average superiority over the Radeon HD 5970 has decreased to an insignificant 3%; besides, in ten tests out of nineteen he lost. The largest lag was recorded in Metro 2033, Just Cause 2, Resident Evil 5 and test Final Fantasy XIV Official Benchmark... Such a result cannot be called a failure, but it was not possible to achieve an unambiguous victory over the AMD flagship. As for the comparison with the GeForce GTX 470 SLI tandem, the hero of this review is slower by an average of 10%, which, again, can be called a good achievement, especially considering the prices and the difference in power consumption levels.
Alas, in the most critical for graphic solutions the top-class round of the tandem GeForce GTX 460 SLI failed to perform really well, despite the impressive potential. The final score remained the same, 10: 9 in AMD's favor, but the average lag behind the Radeon HD 5970 increased to more than five percent; in addition, in individual tests, in particular, in Metro 2033, the loss reached very impressive values. The significant achievements of the GeForce GTX 460 SLI in 2560x1600 resolution can be attributed only to the result in Mass effect 2, where the tested tandem seriously bypassed the Radeon HD 5970 in the minimum performance.
In general, the results shown by a pair of GeForce GTX 460 1GB in SLI mode disappoint us a little. Frankly speaking, we expected more from this tandem. No, its performance cannot be called a failure - in a number of tests it is able to compete on equal terms or even outperform such a recognized champion as ATI Radeon HD 5970. the dual-GPU GeForce GTX 490 will be based on a stripped-down version of the GF104 with 336 active stream processors. Yes, a lower price is an important trump card, and buying a pair of GeForce GTX 460 1GB right now can be a very profitable decision for those who crave high performance in their favorite games, but nothing new potential buyers this option will not open. We can only assume that a dual-processor card based on the full-featured version of GF104 with all ALUs activated will be able to reach new performance horizons. However, Advanced Micro Devices' graphics division should be quicker to think about lowering the cost of the Radeon HD 5970 - with current level prices tandem GeForce GTX 460 SLI looks much more attractive to the end customer, albeit from a purely economic point of view.
As for the Zotac GeForce GTX 460 1GB, which, as already mentioned, is an exact copy of the Nvidia reference card, we liked this version of the GeForce GTX 460 1GB less than the one developed in the bowels of Palit / Gainward. First of all, the insufficiently well-thought-out design of the cooling system is to blame for this, in which the release of heated air outside the system case is difficult due to the unsuccessful layout of the connectors on the mounting plate. Together with a not very efficient radiator, this has led to the fact that in order to maintain an acceptable cooling efficiency, the described design sacrifices comfortable acoustic characteristics... Nevertheless, the card overclocks well, works stably, and we can safely recommend it for purchase, but only if the noise level is not a critical parameter for you, or if you plan to replace the reference Nvidia cooling system with something right after purchase the more efficient. As a gift, you will receive the full version of Prince of Persia: The Forgotten Sands.
Other materials on this topic
Professional overview NVIDIA graphics cards Quadro 5000
Youngest in the family: GeForce GTX 460 1 GB
Cooking the stove in the summer: 4 Radeon HD 5870, 3 GeForce GTX 480 and motherboard Gigabyte GA-X58A-UD9
Surely many wondered whether it is possible to save money when buying a hi-end gaming computer, or are they equipped with extremely expensive parts? It is certainly possible to save money, it does not depend on the class of the system. The whole difference between low-end and hi-end computers is that in the case of hi-end solutions, having a decent amount of money allows you to more creatively approach the solution of the problem. At the same time, if you try, you will be able to save good money on hardware. Today, using the example of the graphics component of a PC, we will try to demonstrate a creative approach to choosing a video card and at the same time try to save a little.
First, let's define the terms of the problem. First and foremost, we want to build a powerful gaming computer capable of showing an acceptable frame rate in all modern games, at least in Full HD resolution and at maximum picture quality settings. As our previous studies have shown, to achieve this goal, it is sufficient to purchase one modern hi-end accelerator.
So, suppose you have already set aside a certain amount of money and are ready to go to the store. However, before you make your final choice, we recommend that you explore an alternative approach to solving the problem. To do this, let's first take a look at how things are with the prices of modern top-end video accelerators:
- AMD Radeon HD 5870 1 GB ~ 12,300 rubles.
- NVIDIA GeForce GTX 480 1.5 GB ~ 15 300 rubles.
- AMD Radeon HD 5970 2 GB ~ 19,000 rubles.
The current king of 3D graphics, the Radeon HD 5970, is the most expensive solution. This is understandable, maximum performance is provided by the forces of two Cypress GPUs - an excellent solution for the uncompromising gamer. Next comes the GeForce GTX 480 based on the NVIDIA GF100 chip and, finally, with a "lag" of 3000 rubles, the top three is closed by the Radeon HD 5870 accelerator, which is slightly inferior in performance to the GeForce GTX 480 video card.
Any of the solutions presented will allow you to achieve the above goal, but can you solve this problem differently? In our previous article, we considered a very promising solution - the GeForce GTX 460 accelerator, which showed decent performance at a relatively low price. I wonder if it will be possible for a price comparable, say, to the price of a GeForce GTX 480, to assemble a tandem from a GeForce GTX 460 and get the same performance? Let's try. Let's estimate how much an SLI bundle of two GeForce GTX 460 video cards will cost:
- 2 x GeForce GTX 460 768 MB ~ 12,500 rubles
- 2 x GeForce GTX 460 1 GB ~ 14000 rubles
So, a pair of GeForce GTX 460 768 MB video cards will cost an amount comparable to the cost of a Radeon HD 5870, and two GeForce GTX 460 video cards with 1 GB of video memory will be about 1,300 rubles cheaper than one GeForce GTX 480 accelerator.
We have at our disposal two video cards GeForce GTX 460 1 GB, and we will try to find out whether the performance of such an SLI bundle can surpass the speed of a single GeForce GTX 480.
| GeForce GTX 460 768MB | GeForce GTX 460 1GB | GeForce GTX 470 | GeForce GTX 480 | |
| GPU technical process, nm | 40 | 40 | 40 | 40 |
| Graphics processing clusters, pcs. | 2 | 2 | 4 | 4 |
| Number of streaming multiprocessors | 7 | 7 | 14 | 15 |
| # Of CUDA cores | 336 | 336 | 448 | 480 |
| Number of texture units | 56 | 56 | 56 | 60 |
| Number of ROPs | 24 | 32 | 40 | 48 |
| GPU frequency, MHz | 675 | 675 | 607 | 700 |
| CUDA core frequency, MHz | 1350 | 1350 | 1215 | 1401 |
| Effective video memory frequency, MHz | 3600 | 3600 | 3348 | 3696 |
| Video memory size, MB | 768 | 1024 | 1280 | 1536 |
| Memory bus width, bit | 192 | 256 | 320 | 384 |
| Video memory bandwidth, GB / s | 86,4 | 115,2 | 133.9 | 177.4 |
| DirectX support | 11 | 11 | 11 | 11 |
| Maximum TDP, W | 150 | 160 | 215 | 250 |
| Recommended power supply unit, W | 450 | 450 | 550 | 600 |
| GPU temperature limit, ° C | 104 | 104 | 105 | 105 |
| Estimated cost according to the data, rubles | 6200 | 7500 | 10250 | 15000 |
If you take a close look at the characteristics of the GeForce GTX 460 1 GB and present them in an SLI tandem, then the “biceps” of the GeForce GTX 480 no longer seem so intimidating. With a high degree of probability, we can say that a pair of GeForce GTX 460 1 GB video cards in SLI mode will be faster than one accelerator GeForce GTX 480. Another question is how much, and whether in all games. We will check this a little later, but for now let's talk about what else a potential owner of an SLI system might need.
Firstly, for the SLI link to work, a special motherboard with support for this technology is required. This is unlikely to be a problem, since most hi-end computers are based on system logic and processors. Intel... Moreover, even not the most expensive modern motherboards based on Intel X58 / P55 chipsets fully support NVIDIA SLI technology.
Secondly, for the SLI tandem to work, you need a powerful power supply unit. However, single hi-end accelerators also consume a lot of energy, so a computer's power supply that can pull out a GeForce GTX 480 will certainly be able to cope with a pair of GeForce GTX 460s.
Third, game compatibility. Today, perhaps, there is not a single game left that would work worse on an SLI or CrossFireX system than on a single video card. And “couples” have no more problems with compatibility than “singles”.
To test our arguments in practice, we used two graphics accelerator based on GPU NVIDIA GF104, one of which is manufactured by Gainward, and the other is manufactured at ZOTAC facilities. For a more detailed look at these accelerators, you can refer to our previous review. Here we will provide only a few photographs of the used accelerators.
“… If you want to get DirectX 11 relatively cheap, then the only choice is GTX 460 and its overclocked versions. They have no competitors in their price category. It's not even worth looking towards the dubious GTX 465, the card works at the level of the younger version
Gambling addiction https: //www.site/ https: //www.site/
Iron workshop
And now - a new attempt: NVIDIA presents two video cards in the middle price range. One, GTX 465, built on the familiar GF100, and the second, GTX 460, got it fresh GF104... They will compete with AMD Radeon HD 5830 and HD 5850.
Break down
|
ZOTAC GeForce GTX 465. |
Let's start with the GTX 465. The video card is based on our old friend, the GF100. The same one: expensive, hot and with three billion transistors. It was not accidental that it appeared on a budget board. It has long been known that the more complex a processor is, the more rejects in its production. It is expensive to throw out partially non-working crystals, therefore junior versions are made on their basis. Everyone does this - and Intel, and AMD, and NVIDIA... So it happened with the GTX 465.
On the new board out of 16 SM blocks (Streaming Multiprocessors), only 11. As a result, 352 of 512 CUDA cores are left. This is almost a third less than that of the GTX 480. The number of texture units has also dropped sharply - from 64 to 44. Tessellation engines - NVIDIA's main pride - also left only 11 instead of 16. The pipeline at the exit was also cut. The number of rasterization blocks was reduced from 48 to 32 pieces, and the memory bus was narrowed down to 256 bits. As a result, the maximum amount of GDDR5 has decreased from 1.5 to 1 GB.
The GTX 465 clocks were tricky. The GPU and stream cores run at GTX 470 speeds of 607 and 1215 MHz, respectively. But GDDR5 was launched at 3200 MHz, which is very little.
Piping hot
The reference GTX 465 was sent to us by the company ZOTAC... Externally, the video card does not differ from the GTX 470. The printed circuit board is completely covered with a plastic cover. The processor has a large aluminum heatsink with five heat-conducting pipes. The board is cooled according to the principle of a turbine: a 70-mm fan takes air from the case, drives it through all the elements and throws it out through the I / O panel. The GTX 465 requires 200 watts and two 6-pin plugs to operate. The number of video outputs has not changed - two DVI and one mini-HDMI.
Next
Interesting that high hopes the GTX 465 was not pinned on from the very beginning - even manufacturers admitted that they expected something cheaper and faster from NVIDIA. As a result, NVIDIA issued a new crystal GF104 and the first video card on it - GTX 460.
The differences between the GF100 and GF104 lie in the details. The processor is also built around large clusters called Graphics Processing Clusters. In the first Fermi chip, there were four such clusters, the lightweight version received two. Each of them has four SM-blocks - there are only eight SM-blocks and the same number of tessellation engines. But if earlier there were 32 CUDA cores per SM-block, then in GF104 their number increased to 48 pieces. The number of texture units has doubled - 8 units per SM. This means that in the maximum configuration, the GF104 has 384 CUDA cores, 64 texture units and 8 tessellation engines. The rear pipeline of the GF104 is equipped with 32 ROPs, a 256-bit memory bus and a maximum of 1 GB GDDR5.
Unfortunately, NVIDIA is in no hurry to release the complete GF104. In the version for GTX 460, one SM block is disabled, and the new board has 336 CUDA cores left, the number of tessellation engines has decreased and 8 texture blocks have disappeared (as you can see, there are 56 of them instead of 64).
But the board was lucky with the frequencies. Due to the fact that the number of transistors in the GF104 was reduced to 1.95 billion, NVIDIA was able to reduce heat generation and increase the operating speed. GTX processor The 460 runs at 675 MHz (just 25 MHz less than the GTX 480), the CUDA cores are running at 1350 MHz, and the GDDR5 memory gets a normal speed of 3600 MHz.
External signs
NVIDIA has released two variants of the GTX 460. The processors are the same, the differences are in the rear pipeline and the amount of video memory. There is a version with 1 GB, and there is a version with 768 MB. And if the lack of 256 MB GDDR5 is a trifle, then the loss of 64 bits on the memory bus and 8 ROPs will affect the game in high resolutions.
Outwardly, the two versions are indistinguishable. The cards are collected on a short, 21 cm, printed circuit board. The cooling system is original. For the first time in for a long time NVIDIA abandoned the turbine: as standard, an aluminum cooler is installed on the crystal, the radiator of which rests on the processor with a copper base and two heat-conducting pipes. The design is crowned with a 70 mm fan. The power requirements of the GTX 460 are modest, with only 160 watts from two 6-pin plugs. Video outputs remain the same as on the GTX 465: two DVI and one mini-HDMI.
Not in a dream, but in reality
|
ZOTAC GeForce GTX 460 1GB. |
We tested two GeForce GTX 460s, both with 1GB GDDR5. The first one was provided by ZOTAC. This board meets the technical parameters of NVIDIA, but differs in design: ZOTAC engineers put a time-tested turbine on the GTX 460 and changed the set of video outputs - in addition to two DVIs, full-size HDMI and DisplayPort appeared.
The second version of the GeForce GTX 460 was sent by the company Palit... It was an overclocked card model: the processor speed was increased to 800 MHz, the stream processors were forced to work at 1600 MHz, and the memory was completely caught up to 4000 MHz.
Despite the increase in frequencies, the board has decreased in size. The Palit version is 2 cm shorter. The cooling system was left with the reference: a radiator, two pipes and a bright orange fan. The set of outputs is original. There are two DVIs, full HDMI and VGA for older monitors.
Confrontation
Finding competitors for NVIDIA cards was a piece of cake. AMD has two required fees- Radeon HD 5830 and Radeon HD 5850, the former is at the level of the GTX 460, and the latter is slightly more expensive than the GTX 465.
AMD Radeon HD 5830 for tests presented by the company Sapphire... Onboard there is a Cypress chip with 1120 stream processors, 800 MHz and 1 GB of memory at 4000 MHz.
Radeon HD 5850 came from Gigabyte... Unfortunately, we came across the board with factory overclocking: the crystal worked 5% faster than the 725 MHz it should have, and before the tests we had to reduce the speed to normal values. Otherwise, everything is standard: 1440 stream processors and 1 GB of memory at 4000 MHz. Gigabyte's cooling system is original: a huge aluminum radiator, four heat pipes and a pair of fans. Note that both AMD graphics cards longer than the GTX 460 by 3-6 cm - you can't put such in an HTPC.
| Table 1 | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Who will win
|
Palit GeForce GTX 460 Sonic Platinum (800 MHz / 4000 MHz). |
For the new NVIDIA video cards, we have assembled a system on motherboard Gigabyte GA-X58A-UD9, put a processor on it Core i7-920, overclocked it from 2.66 to 4 GHz and added three memory sticks Kingston HyperX DDR3-1666 2 GB each. The system was running Windows 7 Ultimate 64-bit.
On tests, we had two synthetic benchmarks ( 3DMark Vantage and Unigine Heaven Benchmark 2.0), three games on DirectX 10 ( Crysis, Devil May cry 4 , Resident evil 5) and two games on DirectX 11 ( Colin McRae DiRT 2 and Aliens vs. Predator).
First, take a look at the standard versions of the boards. The first place in 3DMark Vantage was taken by the Radeon HD 5850. But the second place was unexpectedly taken by the GTX 460, which surpassed not only the HD 5830, but also the GTX 465. In Unigine Heaven Benchmark 2.0, the situation repeated itself, only now both boards also lagged behind the GTX 460 AMD.
|
Sapphire HD 5830. |
The results leveled off in DX10 applications, where the Radeon HD 5850 confidently held the lead. The HD 5830, as expected, was in the last position, outperforming NVIDIA cards only in Crysis. But the GTX 460 and GTX 465 fought to the death. Due to the higher frequencies on the processor and memory, the GTX 460 not only kept up with its older sister, but also overtook it in Crysis.
With tests in DX11, the situation has not changed. Additional GTX 465 tessellation engines didn't help. In DiRT 2, she managed to lag behind the younger version by 12%. A confident victory over AMD did not work out either: in AvP, the eldest of the "Radeon" is 16% ahead of the GTX 465.
Well, for sweetness - the results of the overclocked version of the GTX 460. The crazy 800 MHz on the processor simply defeated the competitors. Even the HD 5850 fell 5.2% behind.
Outcomes
Let's draw the line. GeForce GTX 460 - beyond praise. Low power consumption, high operating frequencies and huge overclocking potential. The new crystal, even without additional cooling, easily overclocks to 850 MHz, which is what manufacturers use.
|
Gigabyte Radeon HD 5850 GV-R585OC-1GD. |
But the GTX 465 turned out to be uninteresting. We don’t understand NVIDIA’s logic at all: why release a cut-down GF100 if we had an almost finished GF104 on hand? The only logical option is to earn extra money.
This is evidenced by the pricing policy of NVIDIA. When we just started preparing this material, they asked about 9500 rubles for the GTX 465. After the GTX 460 for 8,500 rubles appeared on sale, the older version quickly lost in price - at the time of the room delivery they wanted 7,500 rubles for it. The very same GTX 460 for a month of sales managed to fall in price to 8,000 rubles.
But the prices for AMD motherboards have remained practically unchanged. They still ask for 10,000 rubles for the HD 5850, and 7,000 rubles for the HD 5830. In principle, the policy is clear, the cards have been on sale for almost a year, and until they come out new line Radeon - no price collapse can be expected.
* * *
Our conclusion is simple: if you want to get DirectX 11 relatively cheap, then the only choice is GTX 460 and its overclocked versions. They have no competitors in their price category. It's not even worth looking towards the dubious GTX 465, the card works at the level of the junior version, but at the same time it heats up more and eats more energy. AMD has nothing to offer yet: the Radeon HD 5850 is too expensive, and we do not recommend saving 1000 rubles and buying the HD 5830, it is very weak in DX11.
| table 2 | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Table 3 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
