5 ms·

Soon: "Software appear less efficient in real life than in tests. Test-driven development considered harmful."

• chadgeidel 11y ago
  It's already happened. Many here probably remember the ATI "Quake/Quack" debacle where the ATI driver was changing the rendering pipeline based on the binary name: http://www.hardocp.com/article/2001/10/23/optimizing_or_cheating_radeon_8500_drivers http://www.hardocp.com/article/2001/10/23/optimizing_or_chea...
  • semi-extrinsic 11y ago
    It's no secret that GPU drivers are tuned differently to different games. That's a feature, not a bug. You can't expect every game to conform to some super-tight code conventions that optimize against the GPU driver, so Nvidia and ATI optimize their drivers for the various games.
    • chadgeidel 11y ago
      I'm not sure if this is what you are referring to, but in the Quake/Quack incident the texture quality and mip levels* were different than the user-specified values when running a different named program. This was producing obvious visual artifacts, and I consider this cheating. If I specify "16x Anisotropic filtering" and the driver changes that to "4x" or some other number behind the scenes, that's an unacceptable optimization. The current "GeForce Experience" program that selects optimal settings for the current card is an acceptable optimization, I can see which settings are configured and change them if I like. [*] I don't remember the specific things they were adjusting.