This project you've set up is really great, and useful, and I'm glad you're doing it! I just ran a couple of benchmarks, and my gpu result seemed slower than my previous tests. Using the Blenchmark tile size of 128x64 I got a result of around 90 seconds for my GTX580 card. Changing the tile size to 256x256 dropped the rendertime to 60 seconds, a pretty significant difference. I realize you're trying to set a common standard for all benchmarks, but is there not a way to achieve more accurate results? Thanks Todd McIntosh