First Nvidia A100 GPU benchmark results appear at OctaneBench

A benchmark has appeared of Nvidia’s A100 GPU based on the Ampère architecture. The Octane benchmark shows that the GPU is about 43 percent faster than the current Turing GPUs, without the use of ray tracing.

The benchmarks were shared by Jules Urbach, Otoy’s CEO. That is a company that specializes in cloud graphics. Urbach reports that the Nvidia A100 in its tests is about 43 percent faster than the current Turing architecture in Otoy’s Octane benchmark. This concerns the sxm4 variant with a tdp of 400W. The pci-e variant of the A100 has a tdp of 250W, but otherwise has similar specifications. This speed improvement is also evident when no ray tracing is used. For the A100, for example, Urbach used the ‘standard OB4 benchmark’ on a Linux 5.3 system, without ray tracing. The video card achieved a score of 446 points, the CEO reports.

To illustrate: at the time of writing, the fastest GPU in the OctaneBench database is the Nvidia Titan V GPU, which achieves 401 points. This makes the A100 about 11 percent faster. The fastest Turing video card in the OctaneBench database is the Titan RTX GPU, with a score of 322. This makes the A100 from Urbach’s tests about 40 percent faster. A GRID version of the RTX 8000 does slightly better in OctaneBench, with a score of 328.

Nvidia announced its Ampere architecture and A100 GPU in May. That GPU has a die-size of 826 mm² and is mainly intended for data centers or other HPC applications. For the time being, the new architecture is only used in the A100 GPU, but in time there will also be GeForce video cards for gamers based on Ampere. It is not yet known when Nvidia will release such GPUs.

A screenshot of the current Octanebench results. Image via Otoy