The NVIDIA Ampere ampere architecture has been out for more than two months and there are more and more server products involved, but we’ve never seen its actual performance data before. Now, Jules Urbach, CEO of oTOY, a holographic rendering company, has finally satisfied our curiosity. The first product of the NVIDIA ampere architecture is the A100 Computing Accelerator, 7nm process manufacturing, integrated 6912 CUDA cores, 40GB HBM2 memory, and PCIe 4.0 support.
The tool used for this test is OctaneBench, which measures Octane Render rendering performance and is the first commercially available ray-traced renderer that can be fully accelerated with a GPU, but only NVIDIA graphics cards because of its reliance on CUDA technology.
The NVIDIA A100 scored 446 points in this test, the highest ever, and Jules Urbach claims it’s 43% faster than Turing!
We don’t know which Turing product he’s talking about, but the Turing family’s best result in the OctaneBench test came from the GRID RTX 8000, with a run-scorer of 328, a 36 percent lead over a 100.
The Volta Volt architecture performs better than Turing because of its expertise in accelerated computing, with the best results of titan V, Tesla V100 and Quadro GV100 being 401, 371, 354, and the A100 leading by 11-33%.
As for the AMX 30 series of game cards, there are no surprises coming in September, with two possible premieres, including RTX 3080 and RTX 3080 Ti.