Nvidia has announced a PCIe variant of its monstrous Tesla P100 HPC (High-performance computing) GPU at this year’s Supercomputing Conference held in Frankfurt. Prior to this, the Tesla P100 was only available with the company’s DGX-1 supercomputer – which bears a premium price tag of US$129,000 (about RM521,230).
The introduction of a PCIe variant for Nvidia’s Tesla P100 GPU will definitely benefit consumers that are engaging with weather modelling or small-scale servers who simply do not have the resources to purchase the DGX-1 supercomputer. Prior to PCIe, Nvidia had opted to use the NVLink interconnect protocol with the Tesla P100 GPU, which is said to be five times faster than PCIe.
Knowing this, one shouldn’t be too surprised to know that the PCIe variant of the Tesla P100 HPC GPU will come with a lower performance output compared to its more powerful counterpart. For instance, the PCIe variant will come 4.7 TFLOPs of FP64 (double precision floating point) performance; the NVLink variant has 5.3 TFLOPs instead. In addition the boost clock of the PCIe Tesla P100 is recorded at 1300MHz, which is 180MHz lower compared to the NVLink Tesla P100.
Interestingly, Nvidia has released two storage capacities for the Tesla P100 PCIe GPU: 12GB and 16GB – both of which will be utilising the HBM2 memory standard. The release date for the Tesla P100 PCIe GPU is said to be sometime in Q4 this year. Price was not mentioned, but expect the GPU to cost a fortune.