ATI admits it ‘optimised’ drivers for 3DMark 03
The 3DMark 03 cheating controversy continues
ATI has admitted that it tweaked its Catalyst drivers to generate better 3DMark 03 scores and said that it will remove the modifications before the next release of the drivers.
The admission followed 3DMark creator Futuremark's audit of graphics card drivers from ATI and arch-rival Nvidia. The audit was prompted by an ExtremeTech investigation of coding within Nvidia's drivers which yields higher 3DMark 03 scores at the expense of image quality.
On Friday, as we reported here, Futuremark said it had found eight instances of cheating on the part of Nvidia. Essentially, renaming the benchmark executable yielded lower scores in certain tests carried out using the same Nvidia drivers. That indicates that the drivers modify the visual output when they detect the benchmark code is running in order to obtain a better score.
ATI came a cropper the same way. Futuremark saw an eight per cent decrease in the score of one benchmark, Game Test 4, when it conducted the test with a renamed executable rather than correctly titled code. ATI's fix, said Futuremark, contributed to an improvement of just under two per cent in the overall 3DMark 03 score.
A clear case of cheating? Not so, claims ATI. In a statement sent to Beyond 3D, ATI spin doctor Chris Evenden claimed the performance gain was the result of optimised code, not a cheat.
"The 1.9 per cent performance gain comes from optimisation of the two DirectX 9 shaders - water and sky - in Game Test 4," said Evenden. "We render the scene exactly as intended by Futuremark, in full-precision floating point. Our shaders are mathematically and functionally identical to Futuremark's and there are no visual artifacts; we simply shuffle instructions to take advantage of our architecture. These are exactly the sort of optimisations that work in games to improve frame rates without reducing image quality and as such, are a realistic approach to a benchmark intended to measure in-game performance."
So everything's fine and dandy - ATI's driver tweaks are above board and 'legal'. Yet Evenden went on to say: "However, we recognise that these [tweaks] can be used by some people to call into question the legitimacy of benchmark results, and so we are removing them from our driver as soon as is physically possible. We expect them to be gone by the next release of Catalyst."
So, according to ATI, it's all very well for, say, John Carmack to code Doom 3 to take advantage of ATI's chip architecture to squeeze a few extra frames per second out of his game engine without sacrificing visual quality, but not for ATI to do so on his behalf.
We can imagine ATI being touchy about accusations of cheating, having been found to have done just that a couple of years ago. It was found to have modified it drivers to adjust Quake III visual quality in order to gain higher framerates. Renaming the Quake III executable blocked the tactic. But this time it's not employing such methods - it's getting better results with the same visual quality. Isn't that what drivers are supposed to do?
So what has ATI to fear? Well, for all its sensitivity to accusations of cheating, it almost certainly wouldn't have discussed the optimisations if Futuremark hadn't uncovered their presence. That will lead gamers to wonder what else it might be hiding, and we're sure many users will be peering closely at screen grabs for the slightest sign of image degradation between optimised and un-optimised benchmark runs.
ATI's stated reason is that it doesn't want to diminish the value of benchmarks, which is fine while it's ahead of the game, but as Nvidia's denunciation of 3DMark 03 earlier this year showed, manufacturers can and will dismiss benchmarks when they don't get a result they like.
Today, Nvidia claims framerates are no longer define the parameters of graphics performance. Instead, it's visuals that matter, and how can you give an objective measure of such a subjective area? We agree, but we note Nvidia's recent press presentation still made much of benchmark scores.
Both ATI and Nvidia have been found at various times to have rigged their drivers to give better benchmarks, particularly when they've felt themselves to have lost the performance lead, so it's hard to trust either of them. In any case, Nvidia fans will suspect ATI scores, and ATI buffs will equally distrust Nvidia benchmarks.
So forget about the numbers and go back to the visuals. Compare the screen grabs of comparable scenes and base your buying decision on the one that looks the best at 30fps. Can't choose between screens? Then buy on price. Or toss a coin - just as valid a basis for judgement as benchmarks these days. ®