目录
服务器GPU
| GPU | Architecture | TF32 Tensor Core (TFLOPS) | INT8 Tensor Core (TOPS) | GPU Memory (GB) | Memory Bandwidth (GB/s) | Max Power Consumption (W) |
|---|---|---|---|---|---|---|
| NVIDIA H200 | NVIDIA Hopper | 989 | 3958 | 141 | 4.8 | 700 |
| NVIDIA H100 | NVIDIA Hopper | 989 | 3958 | 80 | 3.35 | 700 |
| NVIDIA A100 | NVIDIA Ampere | 156 | 624 | 80 | 1,935 | 400 |
| NVIDIA A16 | NVIDIA Ampere | 4x 18 | 4x 71.8 | 4x 16 | 4x 200 | 250 |
| NVIDIA A40 | NVIDIA Ampere | 74.8 | 299.3 | 48 | 696 | 300 |
| NVIDIA A30 | NVIDIA Ampere | 165 | 661 | 24 | 933 | 165 |
| NVIDIA A10 | NVIDIA Ampere | 62.5 | 500 | 24 | 600 | 150 |
| NVIDIA A800 | NVIDIA Ampere | 156 | 1248 | 80 | 1935 | 300 |
| NVIDIA A2 | NVIDIA Ampere | 18 | 72 | 16 | 200 | 40-60 |
| NVIDIA L40S | NVIDIA Ada Lovelace | 183 | 733 | 48 | 864 | 350 |
| NVIDIA L40 | NVIDIA Ada Lovelace | 90.5 | 362 | 48 | 864 | 300 |
| NVIDIA L4 | NVIDIA Ada Lovelace | 120 | 485 | 24 | 300 | 72 |
| NVIDIA T4 | NVIDIA Turing | 65 | 130 | 16 | 300 | 70 |
参考材料：https://www.nvidia.cn/data-center/tesla-product-literature/