And the driver cheat mudslinging continues…

Yeah, it’s starting to look a little funkier. Without getting into too much detail (which you can read here, here, and here), it’s looking less like a bug and more like an “optimization”. Futuremark put out a report titled “Alleged NVIDIA Driver Cheating on 3DMark03”. Apparently when they redid some of their shader code, NVIDIA performance decreased by as much as about 25% for some tests. Beyond3D also found similar “optimizations” from ATI, but to a lesser extent. ATI apparently jiggered the ordering of the shader instructions (not a bad thing itself) to get better performance for a visually identical shader. Which is what you’d want the driver compiler to do. However they did admit to doing this just for two shaders in the 3DMark03 tests, which is a bad thing – optimizing for a test. They realized this is bad and said they’d take these optimizations out of the next driver revision. You’d think that after ATI had gotten caught cheating (again) with Quake 3, they wouldn’t pull this stuff again. I’m really surprised than any video card companies not only thinks they can get away with this (seemingly repeatedly), but that they feel it’s necessary. Sure ATI and NVIDIA are fighting for supremacy, but do they really feel their cards suck so much they need to inflate benchmark results?