Nvidia unveils monstrous new PCIe Tesla GPU

Nvidia has announced a powerful new PCIe version of its Tesla P100 GPU.

Based on the company’s Pascal and 16nm GP100 GPU architectures , the P100 is almost a complete upgrade over the Tesla Kepler and Maxwell-based cards with larger caches, instruction level preemptive context switching, and double speed (packed) FP16 compute.

This means the Tesla P100 Nvidia’s is highest performing GPU, demanding a whopping 300W to power a base core clock of 1,328MHz and 16GB of VRAM.

NVIDIA will be shipping two versions of the PCIe Tesla P100 with the higher-end PCIe configuration essentially acting as a down-clocked version of the original P100 on a PCIe card, reports Anandtech.

This puts theoretical throughput of the cards at 9.3 TFLOPs for FP32 and 4.7 TFLOPs for FP64, versus 10.6 TFLOPs and 5.3 TFLOPs respectively for the original P100.

The cards are expected to be available in Q4 of this year, with pricing still to be announced.