Yesterday at its GTC Keynote, NVIDIA CEO Jensen Huang announced and detailed the upcoming V100 accelerator. This Volta-based accelerator will be made for the professional market, but will likely pave the way for the company’s next-generation 2000 series GeForce graphics cards.
This processors packs in 21 billion transistors (up from 15,3 billion found on the P100). It is built on TSMC’s 12 nm FinFET processor (down from the 16 nm that Pascal is based off of), it measures in at 815 mm² (from the P100’s 610 mm²). This is quite a jump up in die area, it will be interesting to see how yields will be for such a large chip.
From a gaming perspective this chip has 5120 CUDA cores out of a total of 5376 in the total chip design which we can guess NVIDIA will leave for the TITAN Xv. These CUDA cores are divided into 84 Volta Streaming Multiprocessor Units with each carrying 64 CUDA cores (84 x 64 = 5,376, and NVIDIA will be cutting 4 Volta Streaming Multiprocessor Units for yields).
Even with the cut-down configuration we are looking at a 42% higher pure CUDA core-count over the P100. The V100 will offer up to 15 FP 32 TFLOPS and will make use of 16 GB of HBM2 delivering up to 900 GB/s bandwidth (up from the P100’s 721 GB/s). Currently there are no details on clock speeds or TDP. As more information surfaces we will let you know.