Nvidia GP100 Flagship Pascal GPU Die Shot Shown at Hot Chips
Nvidia GP100 is a top of the line GPU from Nvidia and has been on the market since June this year. The Nvidia GP100 powers the Tesla P100. It is the first chip to use HBM2 and NVLINK interfaces. This Nvidia GP100 chip has not made it to the consumers yet and is only being sold to enterprises.
GP100 consists of Graphics Processing Clusters, Streaming Multiprocessors, and memory controllers. To be exact the chip has six GPCs, 60 SMs, eight 512-bit memory controllers. Pascal architecture is not just better when it comes to raw performance. It is better because of the additional SMs that are present as compared to previous generation GPUs. The SMs found on the chips are also more efficient so this also adds to the power of the chip.
Each of the SM has 64 CUDA cores and 4 texture units, this means that the GP100 has a total of 3840 CUDA cores and 240 texture units. The Nvidia GP100 also houses 4 stacks of HBM2 memory. The GPU features a total VRAM of 16 GB which will be upgraded next year to 32 GB as well.
Nvidia GP100 is a Pascal powerhouse and it can deliver the power needed to deal with high workloads in enterprises. Thr GPU has been designed specifically to handle HPC workloads. The chip has a packed and dense design to it. It houses 15.3 billion transistors. The NVLINK interface can be found on the right side of the chip and it has on the top and bottom, 4, 1024-bit buses.
Nvidia GP100 features 720 GB/s bandwidth. This is, without a doubt, the most powerful chip that we have seen from Nvidia till date. It gets hard to keep up with the tech at times when things surpass each other every other day but it is safe to say that the Nvidia GP100 is going to have the lead for a while now.