The Nvidia Volta Tesla V100 released a year after the Pascal-based Tesla P100 and to say that the Nvidia Volta Tesla V100 is an upgrade would be an understatement as the GPU beats its predecessor in every single way that you can think of. Nvidia Volta GV100 comes in SXM2 form factor and features four stacks of HBM2 memory, adding up to 16 GB.
The Nvidia Volta Tesla V100 is meant for artificial intelligence and machine learning and at GTC 2017 we got to see some numbers from Nvidia. Here we have some Nvidia Volta Tesla V100 GPU benchmarks from other sources and the numbers are very interesting. An Nvidia DGX-1 was used for these benchmarks and the system has 960 TFLOPs of FP16 compute.
To give you an idea of what kind of a machine the Nvidia DGX-1 is, the system has 8 Tesla V100 GPUs with 5120 cores each. This adds up to a total of 40,960 CUDA Cores and 5120 Tensor Cores. On the CPU side of things the system has dual Intel Xeon E5-2698 V4 processors, this means 20 cores, 40 threads at 2.2 GHz.
The Nvidia Volta Tesla V100 benchmarks are as follows:
The Nvidia Volta Tesla V100 has the highest scores in these test so far and there is nothing that comes close to this beast. HP Z8 G4 Workstation managed to score 278706 points in the OpenCL API. Furthermore, a DGX-1 system scored 418504 in OpenCL API and 743537 points in the CUDA API.
These numbers show the lead that the Nvidia Volta Tesla V100 has over the Pascal P100 and it is safe to say that the Nvidia Volta Tesla V100 is a major improvement over the previous generation GPUs. With such improvements, you can expect much better performance in the future from Nvidia GPUs. We are still waiting for Nvidia Volta gaming GPUs though.
Let us know what you think about the Volta Tesla V100.