NVIDIA unveiled the Tesla P100, the first product based on the company’s “Pascal” GPU architecture. Tesla P100 powered by new multi-chip module, similar type layout to the AMD “Fiji.” Tesla P100 has 15 billion-transistor with a 4096-bit wide HBM2 memory interface wires it to four 3D HBM2 stacks; sitting on top a fiberglass substrate that’s rooted into the PCB. The GPU die is built on the 16 nm FinFET process, and is 600 mm² in area.
The P100 offers a double-precision (FP64) compute performance of 5.3 TFLOP/s, FP32 performance of 10.6 TFLOP/s, and FP16 performance of a whopping 21.2 TFLOP/s. The chip has registers as big as 14.2 MB, and an L2 cache of 4 MB. In addition to PCI-Express, each P100 chip will be equipped with NVLink, and in-house developed high-bandwidth interconnect by NVIDIA, with bandwidths as high as 80 GB/s per direction, 160 GB/s both directions. This allows extremely high-bandwidth paths between GPUs, so they could share memory and work more like single-GPUs. The P100 is already in volume production, with its target customers already having bought it all the way up to its OEM channel availability some time in Q1-2017.