NVIDIA CEO Jensen Huang finally unveiled its upcoming Tesla V100 accelerator for the HPC and professional market featuring the company’s next-generation GPU architecture codenamed “Volta.” The Volta-based Tesla V100 packs 21 billion transistors (higher than the 15.3 billion on the Tesla P100) and is built on TSMC’s 12nm FinFET process which is scale-down of the 16nm process used on Pascal. The GPU die measures 815sq.mm. compared to the 610sq.mm. of the P100.
The Volta GPU for the Tesla V100 features 5120 CUDA cores out of 5376. The V100 is rated to offer 15 FP 32 TFLOPS and will be complimented by 16GB of HBM2 memory capable of 900GB/s. Clock speeds or TDP have not been specified.
Analyzing the specs, we’re looking at a beefy GPU with 42% more CUDA cores. The total 5376 can be speculated to be made available once NVIDIA decides to release a Titan X Volta. If NVIDIA decides to proceed with their current naming and strategy, the Tesla V100 GPU configuration could make its way to desktops in the form of the GTX 2080 Ti.