Jules Urbach, CEO of OTOY, a company specialized in holographic rendering in the cloud, took advantage of his own OctaneBench benchmark to reveal the performance of the world’s most powerful GPU, and this is none other than the Nvidia A100, the first Nvidia graphics card which employs the Ampere architecture @ 7nm manufacturing process.
A record breaking week🚀
(*standard Linux OB4 benchmark, RTX off, recompiled for CUDA11, ref. 980=102 OB) pic.twitter.com/WlpG7ZUq2R
— Jules Urbach (@JulesUrbach) July 23, 2020
At a die size of 826mm2, the Nvidia A100 offers no less than 54 billion transistors, 6912 CUDA Cores @ 1410 MHz and 432 Tensor Cores. Despite the great show of muscle, this is not full silicon, so full die variant may still come. This translates to 19.5 TFLOPs FP32 performance, 9.7 TFLOPs FP64 performance and 19.5 TFLOPs FP64 Tensor Core performance accompanied by 40GB of HBM2E memory and a PCI-Express 4.0 or SXM4 interface depending on variant.
As revealed by the CEO, the Nvidia A100 is the fastest graphics card passed by the benchmark, which gave a score of 446 OB4 (in Linux with RayTracing disabled), which means that the Ampere architecture is 43 percent more faster than Turing.
It isn’t stated which exact Turing GPU was used for comparison against the NVIDIA Ampere A100 GPU, but looking at the complete average benches of all cards tested, we see a range of interesting results. On average, the Tesla V100, the A100…