At the moment, NVIDIA has introduced only one next-generation Ampere graphics processor - the flagship GA100, which formed the basis of the NVIDIA A100 computing accelerator. And now the head of the company OTOY, specializing in cloud rendering, shared the first test results of this accelerator.
The Ampere GA100 graphics processor used in the NVIDIA A100 includes 6912 CUDA cores and 40 GB of HBM2 RAM at once. The GPU itself is made using the 7nm process technology at the facilities of TSMC. The computing accelerator is presented in versions with PCIe 4.0 and SXM4 interfaces. At first, NVIDIA A100 accelerators are available as part of proprietary NVIDIA DGX A100 computing systems, which include up to eight GPUs.
The NVIDIA A100 compute accelerator has been tested in the not-so-popular OctaneBench benchmark, which tests GPU rendering performance with the Octane Render graphics engine. It relies on NVIDIA CUDA technologies, meaning it can only render when using NVIDIA GPUs. And the mentioned company OTOY is developing this engine.
It is reported that NVIDIA A100 accelerator showed a record result in OctaneBench, which amounted to 446 points. By comparison, the Volta-based NVIDIA Titan V scores 401 points (down 11 percent), while the fastest Turing-gen graphics card, the Quadro RTX 8000, scores just 328 points (43 percent behind).
Thus, the high theoretical performance of the Ampere processor really translates into faster rendering speed. Recall that the peak performance of the NVIDIA A100 is 19,5 and 9,7 Tflops with single and double precision, respectively. At the same time, the Turing generation Quadro RTX 8000 mentioned above can offer performance only at the level of 16,0 and 0,5 Tflops.
Source:
Source: 3dnews.ru