Slashdot Mirror
ATI Distributing Spurious HL2 Benchmarks
BatonRogue writes "Apparently ATI provided a few Half Life 2 benchmarks to the press and some websites are actually using the benchmarks for their Half Life 2 performance reviews. AnandTech and HardOCP seem to be the only reputable sources of Half Life 2 performance data as they both put together their own benchmarks representative of Half Life 2 gameplay. AnandTech apparently went through every Half Life 2 level and put together a list of the 11 most stressful levels and then created 5 demos, while HardOCP put together two long benchmarks for their review. AnandTech and HardOCP's results appear to agree with each other, while the ATI-backed benchmarks show ATI with a huge performance lead in Half Life 2. Apparently (according to the AnandTech article), ATI was allowed to make their demos while at Valve before Half Life 2 was released, while Valve would not let NVIDIA remove any data from their time at Valve until the game was released. Politics at work as usual."
Gabe Newell seems to have something personal against NVIDIA -- he went absolutely batshit towards NVIDIA, as if it was a personal assault that the GeForce FX 5900 didn't run HL2 well. And yeah, the GeForce FX architecture totally sucks with the source engine. The X800 Platinum is the best at running it, but not by the margin ATI is claiming.
Not to say that given the chance, NVIDIA wouldn't post absurdly inflated numbers. I still personally favor NVIDIA, mostly because thier Linux drivers are of such high quality. And although ATI's Win32 drivers have improved greatly over the past 2 years, in my experience, they aren't quite up to the level of NVIDIA's. Maybe another year and they'll get there. My biggest beef is the lack of support for older products -- the new Catalyst drivers are good, but drivers for the original Radeon and All-in-Wonders suck. NVIDIA's detonator drivers support everything they've ever made, other than the craptastic Riva128 ZX. I'm still using my trusty old TNT2 -- plays a mean game of Quake3 under Linux.
Lex orandi, lex credendi.
You don't see a problem with a company picking and choosing which parts of a game perform better on their card and then releasing benchmarks saying this is how the whole game performs?