Nvidia finally released their latest generation GPU series codenamed Fermi, which was based on the GF100 architecture, on 26 March 2010. The arrival of the GeForce GTX 480 and GTX 470 graphics cards marked the return of the empire, and while this release saw them take back the performance crown, there were a few casualties.

First and foremost the GeForce GTX 480 is expensive, around $100 more than the Radeon HD 5870 which was more than half a year old by the time the Fermi graphics cards arrived. The GeForce GTX 480 also turned out to be a hot and hungry graphics card, sucking down up to 100 watts more power at idle and under full load, while it can operate at around 100 degrees.

Therefore the GeForce GTX 480 really is a power house, consuming and delivering the highest numbers on all fronts. Given the wait, the anticipation and the eventual price of the GeForce GTX 480, it is fair to say everyone was expecting more than what they got. However recent driver updates have improved performance, cementing its victory over the Radeon HD 5870.

Even the GeForce GTX 470 is starting to pick up the slack, though it still suffers from many of the same issues that its bigger brother the GTX 480 faces. With the GeForce GTX 480 priced at $500 and the GTX 470 around $350, there was a massive pricing gap in the Nvidia DirectX 11 lineup that desperately needed to be filled. Therefore two months after the initial launch, Nvidia released the GeForce GTX 465 with a price tag of $280, under cutting the Radeon HD 5850 by around $20.

The problem with the GeForce GTX 465 when compared to the Radeon HD 5850 is that it’s much louder and much more power hungry, not to mention slower. Therefore, while gamers stand to save $20, they lose out in pretty much every other category. So then although the GTX 465 helps to fill out the lower end of the pricing spectrum, it does do a pretty poor job of it.

What Nvidia desperately needs is a new GeForce GTX 400 series graphics card that can provide gamers with a serious bang for their buck, something they simply cannot refuse. Such a product might now exist, as we unveil the GeForce GTX 460, which is based on the new GF104 architecture. Unlike the GeForce GTX 465 which was just a cut down version of the existing GF100 products and as a result inherited all their weaknesses, the GeForce GTX 460 is a completely new animal.

This means the GeForce GTX 460 might not be the power hungry, hot graphics card that we have come to expect of GeForce GTX 400 series. In fact, with a TDP (Thermal Design Power) rating of just 160 watts for the 1GB version, we are willing to bet that it is not, as this makes the GeForce GTX 460 more power conscious than the Radeon HD 5830 should it be true. Before we jump into the benchmark results, here are two new GeForce GTX 460 graphics cards from Inno3D and Palit.

What’s more interesting is the fact that the GeForce GTX 460 has the same number of TMUs (Texture Mapping Units) as the GeForce GTX 470 with 56 in total, which is 27% more than the GeForce GTX 465. The Inno3D GeForce GTX 460 (768MB) version features 24 ROPs (Raster Operations Pipelines), while the 1024MB versions will carry 32 like the GeForce GTX 465.

The graphics clock speed for fixed function units is 675MHz, while the CUDA Cores operate at a more aggressive 1350MHz. This means that the fixed function units and CUDA Cores are clocked 11% higher on the GeForce GTX 460 than they are on the GTX 465 and GTX 470 graphics cards.

There are two versions of the GeForce GTX 460, one teamed with 1024MB of GDDR5 memory and a smaller cheaper version with just 768MB of GDDR5 memory. Both are clocked at 900MHz (3600MHz DDR). While they both share the same operating frequency, the 768MB cards only feature a 192-bit wide memory bus, while the larger 1024MB cards get a bigger 256-bit bus.

This means that while the 1024MB cards will have a peak memory bandwidth of 115.2GB/s, the smaller 768MB cards are limited to 86.4GB/s, a 25% bandwidth reduction. It is interesting to note that the larger 1024MB version of the GeForce GTX 460 will actually have 12% more bandwidth at its disposal when compared to the more expensive GeForce GTX 465.

Unlike the GeForce GTX 465, the Thermal Design Power (TDP) rating of the GeForce GTX 460 is very reasonable, with the 1024MB version rated at 160 watts and the 768MB version 150 watts. This makes the 1024MB cards 20% more fuel efficient when compared to the GeForce GTX 465 and 9% more efficient when compared to the Radeon HD 5830.

Other than the PCI Express slot, the GeForce GTX 460 draws in power through a pair of external PCI Express power connectors. The GTX 460 requires dual 6-pin connectors, which is the same configuration used by the Radeon HD 5830. The GeForce GTX 460 also features a single SLI bridge connector allowing it to be paired with another graphics card.

Hardware - Intel Core i7 965 Extreme Edition (Overclocked @ 3.70GHz) - x3 2GB G.Skill DDR3 PC3-12800 (CAS 9-9-9-24) - Seagate 500GB 7200-RPM (Serial ATA300) - Palit GeForce GTX 470 (1280MB) - Palit GeForce GTX 460 (1GB) - Inno3D GeForce GTX 460 (768MB) - HIS Radeon HD 5850 (1GB) - Asus Radeon HD 5830 (1GB) - HIS Radeon HD 5770 (1GB) Crossfire - HIS Radeon HD 5770 (1GB) - Asus P6T Deluxe (Intel X58) Software - Microsoft Windows 7 Ultimate (64-bit) - Nvidia Forceware 257.21 - Nvidia Forceware 258.56 (For GeForce GTX 460 Cards) - ATI Catalyst 10.6

At 2560x1600 the GeForce GTX (768MB) is 23% faster than the Radeon HD 5770, though it is also 22% slower than the Radeon HD 5830, which is a little troubling to see. The GeForce GTX 460 (1GB) was 8% faster than the 768MB version, though it was still 15% slower than the Radeon HD 5830, which was again disappointing.

For testing with Aliens vs. Predator we have used the custom DX11 benchmark as it is highly demanding and plays though an test scene in the same sequence every time for highly accurate results. The maximum possible quality settings were enabled along with 4xAA/16xAF.

When testing with Aliens vs. Predator at 1920x1200 the GeForce GTX 460 (768MB) was 4% faster than the Radeon HD 5830, which equated to just 1fps. The GeForce GTX 460 (1GB) on the other hand was 8% faster than the 768MB version, delivering two extra frames per second. This meant that the 1GB card was 10% slower than the GeForce GTX 470 and 13% slower than the Radeon HD 5850.

For testing with Mass Effect 2 we once again relied on Fraps which was used to test 60 seconds to game-play from the level “Lazarus Research Station”. Again we used maximum in-game quality settings with 4xAA/16xAF enabled.

When testing with Mass Effect 2 we find that the performance margins are much greater than those seen when testing with Aliens vs. Predator. However this time the GeForce GTX 460 (768MB) is a whopping 26% slower than the Radeon HD 5830. Meanwhile the GeForce GTX 460 (1GB) delivered just 6fps more performance, making it 15% faster than the 768MB version. Still despite this the GeForce GTX 460 (1GB) was slower than the Radeon HD 5830, while it trailed the Radeon HD 5850 by a 33% margin.

Once again in order to test Call of Duty Modern Warfare 2 we were forced to test with Fraps. This time we have changed the scene that we test with to the Hummer scene in “ACT I - Team Player” where your unit comes under fire traveling though the city. Again maximum in-game quality settings were used which includes 4xAA while 60 seconds to game-play was used for testing.

The Call of Duty Modern Warfare 2 results are interesting as we find that the GeForce GTX 460 (768MB) is able to outperform the Radeon HD 5830 by a 13% margin as it delivered 5fps more at 1920x1200. The GeForce GTX 460 (1GB) was still 14% faster than the 1GB version, while it was able to match the Radeon HD 5850 performance despite trailing the GeForce GTX 470 by a 14% margin.

Dirt 2 has a fantastic in-built benchmark that measures actually game play very accurately. Therefore we put down Fraps for this title and just use the built in benchmark. The game was run in the DirectX 11 mode with 4xAA enabled in game with maximum quality visuals.

The GeForce GTX 400 series fares very well in Dirt 2 and the new GeForce GTX 460 is no exception. At 1920x1200 the GeForce GTX 460 (768MB) card was 14% faster than the Radeon HD 5830, which was a 7fps gain. The GeForce GTX 460 (1GB) was still 9% faster than the 768MB version, while it also defeated the Radeon HD 5850 by a relatively small 3% margin.

Next up we have Battlefield Bad Company 2 which was tested using Fraps to record 60 seconds to gameplay from the level “Cold War”. Again the game was tested using maximum quality settings with 4xAA/16xAF enabled.

When testing with Battlefield Bad Company 2 at 1920x1200 we found that the GeForce GTX 460 (768MB) graphics card was 10% slower than the Radeon HD 5830. Moreover, while the 1GB version of the GeForce GTX 460 did deliver 7% more performance, it was still 16% slower than the GeForce GTX 470 and 21% slower than the Radeon HD 5850.

For testing Far Cry 2 we used the “Action Scene” from the custom benchmark as this is an excellent representative of real in-game performance. For this DirectX 10 title maximum in-game quality settings were used along with 8xAA.

Far Cry 2, which is a game that works much better on Nvidia hardware, saw the GeForce GTX 460 (768MB) deliver an incredible 61% more performance than the Radeon HD 5830. The GeForce GTX 460 (1GB) was 14% faster than the 768MB version, which allowed it to defeat the Radeon HD 5850 by a 27% margin, while it was again 14% slower than the GeForce GTX 470.

When testing with Just Cause 2 the in-built benchmark was used, there are three tests to choose from and we went with the most demanding which is “Concrete Jungle”. Fraps was still used to record the minimum frame rate as the in-built benchmark does not record that result. Again maximum in-game quality settings were used along with 8xAA/16xAF.

The Just Cause 2 results at 1920x1200 place the GeForce GTX 460 (768MB) at a 14% performance advantage when compared to the Radeon HD 5830. The GeForce GTX 460 (1GB) was then another 12% faster than the 768MB version, making it 18% slower than the GeForce GTX 470 while it matched the performance of the Radeon HD 5850.

Rather than use the Crysis Warhead benchmark we elected to use Fraps instead. In doing so we have recorded 60 seconds of actual gameplay from the level “All The Fury” using the enthusiast quality setting with AA and AF disabled.

Crysis Warhead found the GeForce GTX 460 (768MB) to be 6% slower than the Radeon HD 5830 at 1920x1200 which was surprising. Meanwhile the GeForce GTX 460 (1GB) was just 7% faster than the 768MB card. This did allow the GeForce GTX 460 (1GB) to match the Radeon HD 5830, while it was 16% slower than the Radeon HD 5850 and GeForce GTX 470 graphics cards.

Metro 2033 is another game that has no kind of in-built benchmark so we have relied on Fraps once again to gauge the performance of each graphics card. The “Chase” level was used where we recorded the first 60 seconds of the cart ride. The game was tested using DirectX 11 though the Tessellation feature was disabled as it hammers performance too much in this game. Furthermore anti-aliasing was disabled and instead we used the default analytical anti-aliasing with 4xAF.

The Metro 2033 frame rates were very low at 1920x1200 on all tested graphics cards. With just 14fps at this resolution the GeForce GTX 460 (768MB) was only able to match the performance of the Radeon HD 5830. The GeForce GTX 460 (1GB) graphics card, which was 14% faster with 2fps extra, was still 11% slower than the Radeon HD 5850 and 24% slower than the GeForce GTX 470.

Splinter Cell Conviction requires Fraps to measure in-game performance. For this game we used the first level know as “Merchant's Street Market” where we recorded 60 seconds of game-play. The highest possible in-game quality settings were used along with 4xAA/16xAF.

Splinter Cell Conviction is a game where the ATI cards appear to fare better than the Nvidia competition. At 1920x1200 the GeForce GTX 460 (768MB) was 7% slower than the Radeon HD 5830. The GeForce GTX 460 (1GB) provided just 7% more performance with 44fps, making it 17% slower than the Radeon HD 5850 and GeForce GTX 470 graphics cards.

S.T.A.L.K.E.R.: Call of Pripyat may not be the best looking game but it is a DirectX 11 title and it is a game that we have been testing with for some time now. The game does have an official DirectX 11 benchmark but we have decided to dump that and for the first time will test actual in-game performance using Fraps. The level of choice is the first which is called Zaton and once again we have recorded 60 seconds to game-play.

The S.T.A.L.K.E.R.: Call of Pripyat results at 1920x1200 show the GeForce GTX 460 (768MB) leading the Radeon HD 5830 by a 3% margin. The 1GB version of the GeForce GTX 460 was 17% faster than the 768MB card, which was impressive. Despite this huge performance gain the GeForce GTX 460 (1GB) was still 3% slower than the Radeon HD 5850 and 24% slower than the GeForce GTX 470.

World in Conflict has a built in benchmark that works rather well so we decided to stick with that for testing this game. This time we used the “very high” quality preset with 4xAA/16xAF enabled.

Finally we have World in Conflict, and at 1920x1200 we found that the GeForce GTX 460 (768MB) graphics card was 8% faster than the Radeon HD 5850. The GeForce GTX 460 (1GB) on the other hand was 13% faster than the 768MB card, which allowed it to match the performance of the Radeon HD 5850 while it was still 15% slower than the GeForce GTX 470.

Overclocking the GeForce GTX 460 cards makes things very interesting. The GeForce GTX 460 (768MB) goes from being slower than the Radeon HD 5850 to matching it with 49fps. However the GeForce GTX 460 (1GB), which also managed 49fps before any overclocking took place, shot up to 60fps once overclocked. This made the GeForce GTX 460 (1GB) faster than even the GeForce GTX 470, which is incredible.

This time when testing with Crysis Warhead we see that the GeForce GTX 460 (768MB) went from 30fps to an impressive 37fps once overclocked, which made it faster than the Radeon HD 5830 and just a single frame slower than the Radeon HD 5850. Meanwhile the GeForce GTX 460 (1GB) went from 37fps to 40fps, which again made it faster than both the Radeon HD 5850 and GeForce GTX 470 graphics cards.

The last game we tested the overclocked configurations with was Battlefield Bad Company 2 and here we see that the GeForce GTX 460 (768MB) went from 43fps to an impressive 53fps once overclocked. This placed it within 2fps of the GeForce GTX 470 which is impressive. The GeForce GTX 460 (1GB) on the other hand was able to overtake the GeForce GTX 470 once overclocked, making it just a single frame slower than the Radeon HD 5850.

If the benchmark numbers were not impressive enough the GeForce GTX 460 will certainly blow you away when it comes to the power consumption figures. The GeForce GTX (768MB) consumed just 285 watts under full load, making it 9% more efficient when compared to our Radeon HD 5830 graphics card. Although that is 24% more power than the Radeon HD 5770, when you consider just how much faster the GeForce GTX (768MB) was most of the time, these are pretty impressive numbers.

The GeForce GTX 460 (1GB), which featured an improved PCB design when compared to the 768MB version, actually used less power when at idle. Under full load the GeForce GTX 460 (1GB) used 12% more power, though it still used 22% less power than the GeForce GTX 470.

Those impressive power consumption figures have allowed these GeForce GTX 460 graphics cards to operate at very reasonable temperatures. While a full load temperature of 64 degrees for the GeForce GTX (768MB) is impressive, it is made far more impressive by the fact that when torching these cards with FurMark, the GeForce GTX (768MB) and even the GeForce GTX 460 (1GB) remained extremely quiet.

That said, if we remove the Far Cry 2 result which saw the GeForce GTX 460 768MB lead the Radeon HD 5830 by a 61% performance margin, the GeForce GTX 460 was just 1% faster overall. The other extreme result was seen when testing with Mass Effect 2, where the GeForce GTX 460 was 26% slower and if we also remove that result the GeForce GTX 460 was 3% faster.

Whichever way you slice it, the GeForce GTX 460 768MB was overall faster than the Radeon HD 5830 by a small margin in the dozen games that we tested with. Then when we measured the power consumption of each graphics card the GeForce GTX 460 again came out on top, using slightly less power. Furthermore the reference cooler also appears to do a much better job, as the GeForce GTX 460 was found to be 18 degrees cooler than the Radeon HD 5830 when under load.

The GeForce GTX 460 768MB is clearly the better choice of the two graphics cards in our opinion. However having said that the Radeon HD 5830 is almost 5 months old now, making the GeForce GTX 460 a considerably newer product. Furthermore the Radeon HD 5000 series has been available for almost 10 months now, which hardly makes the technology cutting edge anymore.

The next battle takes place in the $220 - $240 US price range, and with the cheapest Radeon HD 5850 graphics cards priced at around $300 US, the GeForce GTX 460 1GB appears to have found a hole in the ATI pricing structure. There are a few factory overclocked Radeon HD 5830 graphics cards that are priced around $240 US, but with mild 20–40MHz overclocks they represent fairly poor value.

Performance wise the GeForce GTX 460 1GB graphics card was on average 11% faster than the 768MB version. This is a nice performance boost, but also take into account the price premium Nvidia are asking for this steroid enhanced version. Nvidia has suggested that the 1GB version will cost between 10 and 20% more. With an average of 11% more performance, we are hoping most manufacturers price their cards at around $220 US.

Something else to consider is that at $220 US the GeForce GTX 460 1GB graphics card will be 27% cheaper than the Radeon HD 5850, despite being just 7% slower on average. Nvidia clearly wanted to make a splash with the GeForce GTX 460 graphics cards and we feel they have done so, and perhaps even at the expense of the GeForce GTX 470.

Another interesting comparison can be made between the GeForce GTX 460 1GB and GeForce GTX 470 graphics cards. On average the GeForce GTX 460 1GB was 17% slower than the GeForce GTX 470, which is a fairly decent margin. However at $220 US the GeForce GTX 460 is 31% cheaper and even at $240 US it is still 25% cheaper.

Then when you consider the overclocking potential of this graphics card and the fact that for around $260 US Palit will be releasing a version that performs much like the GeForce GTX 470, you have to wonder what this will do to sales of Nvidia’s own high-end graphics cards.

-

-

Another interesting thought is the fact that GeForce GTX 460 1GB SLI graphics cards could cost as little as a whisker over $440. That is interesting because the GeForce GTX 480 happens to retail for around $500 and we are willing to bet that a pair of GeForce GTX 460 graphics cards would offer a similar level of performance. Sadly we were unable to deliver SLI results in this review as our second 1GB card arrived with defects. However this is something we will be looking at in the near future.

Overall, by redesigning the GeForce GTX 460 to remove the unnecessary bits, making it a more compact and efficient GPU, it appears that Nvidia has created the perfect mid-range graphic cards. While we were not completely sold on the GeForce GTX 480, 470 or 465 graphics cards, we genuinely like the GeForce GTX 460 in either its 768MB or 1GB flavors, and prefer it to the ATI competition.