Crysis can be particularly memory-dependent, which in this case highlights the difference between the 768MB and 1GB GTX 460. 2560x1600 is not a viable option on the 768MB card here (or in most other games) due to the lack of VRAM, while at lower resolutions we can still clearly see the impact of having less RAM, less L2 cache, and less ROP power. At both 1680 and 1920, the 1GB GTX 460 is roughly 10% faster than the 768MB card. This much less than the 33% lead the 1GB GTX 460 has in terms of RAM/L2/ROP, but it’s still clear that there is a price to pay on the 768MB GTX 460.
Meanwhile the Radeon 5830 - already a hobbled card due to having half the ROPs of a full Radeon HD 5870 – takes a hard loss here. The 768MB GTX 460 comes ahead at 1920 by roughly 14% even with its RAM disadvantage. AMD definitely has their work cut out for them. As for the Radeon 5850, the 1GB GTX 460 trails right behind it until we hit 2560, where AMD’s continued advantage at high resolutions helps the card pull away some.
Looking at the minimum framerates, the difference in RAM/L2/ROPs becomes more pronounced. Here the 1GB GTX 460 has a 15% advantage at 1920, and strangely enough even beats a 5850 here. With the greater overhead of SLI this becomes even more of an issue, with our ragtag SLI set of 1GB GTX 460s beats our pair of factory overclocked EVGA 768MB GTX 460s by 33%. Finally the 5830 fares even worse, losing to the 768MB GTX 460 by 35%. In Crysis there is no substitute for more RAM and more ROPs.
BattleForge switches things up some. No longer are we RAM/L2/ROP dependent as much as we are on raw shader power. This gives the 1GB GTX 460 only the slightest advantage over the 768MB card, while the factory overclocked cards clearly dominate here.
This also gives us a situation where we can gauge the GTX 460 and GTX 465 in a shader-bound test and see just how well the GTX 460 can do. The result in the favor of the GTX 465 by quite a bit, where it beats the 1GB GTX 460 by nearly 15%. With the need to extract ILP from the GF104 GPU to fully utilize the GPU’s CUDA cores, this looks to be approaching a worst-case scenario for the GTX 460, as it even falls behind last year’s GTX 275 and Radeon HD 4890.
As for the Radeon 5000 series, AMD’s situation improves here. The 5830 still loses, but only by 10%. Meanwhile the 5850 pulls ahead of the 1GB GTX 460 by 18%.
By moving to DX11 on BattleForge we can not only test a card’s shaders for graphics, but also computing through DirectCompute. The addition of a ComputeShader doesn’t seem to be benefitting anyone here, with the GTX 465 leading the GTX 460 by 15%, a tiny gap between the two GTX 460s, and finally the 5830 bringing up the rear by 11%. This will be one of the only games that favors the GTX 465 over its heir apparent.
