News
News Categories

NVIDIA announces Tesla P100 and touts it as the most advanced GPU for PCIe-based servers

By Wong Chung Wee - on 20 Jun 2016, 2:00pm

NVIDIA announces Tesla P100 and touts it as the most advanced GPU for PCIe-based servers

(Image source: NVIDIA)

NVIDIA has announced Tesla P100 and claims it’s the most advanced GPU for PCIe-based servers. Based on the Pascal GPU architecture, the new GPU is targeted at deep learning applications, as well as for mixed applications hosted at data centers.

(Image source: NVIDIA)

The Tesla P100 is not to be confused with the NVLink-based Tesla P100 that was seen in GPC this year. Though they share the same moniker, the newer Tesla P100 is targeted at more generic PCIe-based servers; while the earlier Tesla P100 is only meant for NVLink-enabled systems. Also, the newer Tesla P100 differs in performance, according to its specifications, it delivers up to 4.7 TFLOPS for double precision floating-point operations, 9.3T FLOPS for single precision floating-point operations.

There are two configurations for the Tesla P100 GPU accelerator; the first SKU supports up to 16GB of stacked VRAM, with a memory bandwidth of 720GB/s, the second configuration entails support for up to 12GB of stacked video memory that is accessible with a memory bandwidth of 540GB/s. The Tesla P100 features a new memory design, Chip on Wafer on Substrate (CoWoS) with HBM2, resulting in the increased memory bandwidths highlighted above.

(Image source: NVIDIA)

The Tesla P100 GPU accelerator aims to provide a highly scalable parallel processing system, vis-à-vis CPU-based HPC systems. In a single node, up to 8 Tesla P100 GPUs can be supported; its performance, vis-à-vis single CPU nodes, can be scaled up more efficiently for various applications like machine learning and other HPC information systems. This also translates to more savings as it’s more economical to build a single powerful compute node than having to link up several weaker compute nodes to aggregate their compute capabilities. The pricing of the Tesla P100 hasn’t been revealed; however, the GPU accelerators will be ready to ship by Q4 2016 to all major OEMs globally.

With a PCIe version of the promising NVLink-based Tesla P100 officially announced, in our opinion, this would probably be the basis of a future GeForce GTX 1080 Ti or Titan equivalent for the DIY enthusiasts and hardcore gamers in the later part of this year.

(Source: NVIDIA)

Join HWZ's Telegram channel here and catch all the latest tech news!
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.