The original plan was to cover image quality as well as performance in this review, but with the changing drivers and bugs, it still isn't possible. Besides, if anyone is cheating and drivers get ammended, the chances are performance will not be affected too greatly. Our focus is on performance, as this is what most people will be looking at when they buy a card. It is almost impossible to cover every single combination of AA, AF and resolution, so instead we chose 3 modes, Low, Medium and High.

- Low - 1024x768, no AA or AF.
- Medium - 1280x1024, 2 x AA, 8x AF
- High - 1600x1200, 4x AA, 16x AF

We wanted to stick to these rules throughout all tests, but annoyingly a few things stopped us. First of all, the 5950 Ultra could not handle 16x AF, so in High mode was run with 8x AF.

Far Cry was run using a custom made demo which toured around the island. Details were kept on maximum.

Halo, which was run with Pixel Shader 2.0 support, does not let you manually adjust AA and AF due to off-scene rendering so these were not changed.

Call of Duty, which was used to test OpenGL performance also had problems. The buggy ATi drivers would crash if AA or AF was forced (thankfully now fixed), so to put them on a level playing field, all cards were run with no AA and AF.

3DMark and Aquamark were run in default modes with nothing forced on or off. All games were run with maximum details settings, with only the resolution, AA and AF being modified. Vsync was of course, off at all times.

Average frame rates were consistant across multiple testings, but in Far Cry and Halo, the minimum frame rate fluctuated a lot, probably due to having only 512mb of memory. So these results were not used.

The test bed was an Athlon 64 3200+ on an MSI K8T Neo with 512mb of Corsair CMX512-3200LLPT. The majority of the testing was done using Benchemall, I don't know how we would have survived without it!

Making the most of the benchmarks

Results are shown in their raw frame rates for those who want to trawl through themselves. But for the average reader, we recommend looking at the percentage increases. These are done for each benchmark and also an average over all the benchmarks. These give the simplest indication of performance difference.

The first set of benchmarks compare the next generation cards to their predecessors, while the latter benchmarks put the new cards head to head.

The drivers used were 61.34 on the 6800 and 5950U. The 6800 Ultra was run with the slightly older 61.12 drivers. Both x800 cards were run with 4.4 drivers but we've been told performance would be almost the same, it was more bugs being fixed. It would have been nice to have the same drivers for all the nVidia cards, but circumstances prevented this.

ATI

I think the averages give the best idea of performance of these two cards, at 33 and 65 percent performance increase over their previous flagship product - the 9800XT. In Far Cry's low settings, it is fairly obvious the CPU is the limiting factor, as the results are the same across all 3 cards. This is also case in the OpenGL Call Of Duty, but across all 3 settings. This is more than likely because Call of Duty is now quite an old game and is not particuarly taxing. The most noticeable difference is in Halo, with a massive 113% increase in performance.

In general, it is in the high settings that the next generation cards really shine, having very little performance hit when AA, AF or high resolutions are used. It would be intresting to see what these cards are capable of in a faster machine.

It was interesting to see how CPU dependent Aquamark was found to be, having almost identical results across the X800 cards. 3DMark on the other hand was much less deterred.

<< Back | 1 | | 2 | | 3 | | 4 | | 5 | Next >>