英伟达显卡算力一览:https://developer.nvidia.com/zh-cn/cuda-gpus
下表数据以PCIE接口为基准
GPU | 算力(FP64/FP32/FP16/INT8/INT4) | 显存 | 显存带宽 | 功率 |
Tesla P4 | -/5.5T/-/22T/- | 8G | 192GBps | 50w/70w |
Tesla P40 | -/12T/-/47T/- | 24G | 346GBps | 250w |
Tesla T4 | -/8.1T/65T/130T/260T | 16G | 320GBps | 70w |
Tesla V100 | -/14T/112T/-/- | 16G/32G | 900GBps | 250w |
Tesla P100 | 4.7T/9.3T/18.7T | 12G/16G | 549GBps/732GBps | |
Tesla A100 | 9.7T(19.5T)/19.5T/312T/624T/- | 40G/80G | 1555GBps/1935GBps | 250w/400w |