Shift. the July 28, 2020 at 10 h 00 min
The Nvidia A100, the primary GPU primarily based on the Ampere structure, has simply undergone its first Benchmark on the OctaneBench platform. It is unsurprisingly the quickest GPU designed thus far. The cardboard examined is 45 factors forward of the Titan V, which held the earlier file. The RTX 2080 Ti in the meantime is behind 144 factors, or about 48%. This comparability is solely indicative, the 2 graphics playing cards do not likely play in the identical class.
Final Might, Nvidia offered the A100, the primary GPU primarily based on the Ampere structure. The agency then introduced efficiency as much as 20 instances larger than that of the V100 GPU meant for the skilled section. Two months have handed for the reason that announcement and right here is lastly a primary GPU Benchmark which provides us an outline of its efficiency. The take a look at was carried out on the OctaneBench platform, a Benchmark designed to check efficiency in OctaneRender, a 3D rendering engine.
The cardboard acquired a rating of 446 factors and snatched the lead of the standings from the Nvidia Titan V which recorded a median of 401 factors. This represents a rise of 11.2%. The GeForce RTX 2080 Ti is fourteenth within the rating with 302 factors. The A100 is sort of 48% quicker than the Turing card. A fairer comparability would put the GPU in confrontation with the Tesla V100 which posted 371 factors. The A100 is subsequently 18% quicker than its actual predecessor.
Ampere A100: its efficiency doesn’t prefigure these of the RTX 3000
For players, this Benchmark shouldn’t be seen as an outline of the efficiency of future RTX 3000 graphics playing cards that aren’t meant for a similar makes use of, even when they’ve the identical Ampere structure. The RTX 3000 playing cards have completely different GPUs and are geared up with extra RT cores in comparison with the A100 GPU meant for the world of synthetic intelligence. The Bechmark was additionally made with the RT cores deactivated.
As a reminder, the A100 GPU is a monster which has 54 billion transistors on an space of 826 mm². Engraved in at 7nm, this chip is supplied with 6912 CUDA cores and 432 Tensor cores supported by 40 GB of HBM2 VRAM. For comparability, the subsequent RTX 3080 Ti graphics card can be geared up with 5,376 CUDA cores and 12 GB of GDDR6x VRAM.