NVIDIA Tesla V100 absolutely smokes Pascal (and Vega) in Geekbench

NVIDIA's Volta-based Tesla V100 supercomputing cards have been shipping for some time but until now there were no benchmarks of this card. This changes today as someone with access to a NVIDIA DGX-1 system, which costs $149,000 and features eight Tesla V100 cards, decided to upload some benchmark scores to the Geekbench database.

The single-core Geekbench 4 Compute test measures just one card so you can easily compare scores with other configurations. The results are very impressive and show this huge 815mm² chip packs a massive punch. You can view the full ranking over here.

The Tesla V100 scores 743537 in the CUDA test and 481504 in the OpenCL test. For comparison, the fastest Pascal-based Tesla P100 system on the list has a CUDA score of "just" 320031. This means the Volta chip is 2.32x faster than Pascal and that doesn't even take Volta's new Tensor cores into account!



The chart below provides a breakdown of the tests and shows the performance difference between Pascal and Volta is extreme across all tests. Some of these tests are showing performance increases of over 350 percent!



So how does AMD score in Geekbench 4 Compute? The fastest AMD part in the database (a Radeon RX Vega) has a score of 205242. This means Volta is 2.35x faster than Vega with OpenCL and 3.62x faster with CUDA.

The one thing we don't know yet is how this will impact the future Volta-based consumer video cards. Of course, the high-end Volta-based GeForce GTX cards will not show double or even triple the performance like we see here in these compute workloads. But it does seem likely that the future GeForce GTX 1180 (or whatever it's called) will offer a very meaningful performance increase versus the current GTX 1080 lineup.