Nvidia announces PCI-e version of Tesla P100

Nvidia has announced PCI-e versions of the Tesla P100 accelerator. Nvidia already unveiled the HPC card with Pascal GPU in April, but that version had a Mezzanine connector developed by the manufacturer itself.

Nvidia will make two versions of the PCI-e variant of the P100 available at the end of this year: one with 16GB hbm2 and one with 12GB. The cards do not have the nvlink interconnect: Nvidia has developed its own Mezzanine connector for that interface with which P100 cards can communicate with each other.

The cards have slightly lower boost clock speeds than the Mezzanine version and the TDP is also lower: 250W instead of 300W. The 12GB version of the pci-e-P100 also has a 3072-bit wide memory bus, compared to the 4096-bit interface of the other two cards. Like the P100 card that Nvidia announced in April, it concerns accelerators for high performance computing such as supercomputers with a Pascal GPU and high bandwidth memory of the second generation.

Nvidia announced the P100 cards at the International Supercomputing Conference in Frankfurt.

Table sourced from Anandtech