| 规格 | NVIDIA A40 | NVIDIA H200 NVL |
|---|
| 架构 | Ampere (GA102) | NVIDIA Hopper |
| 显存 | 48GB GDDR6 with ECC | 141GB |
| 显存带宽 | 696GB/s | 4.8TB/s |
| 显存位宽 | 384bit | — |
| 算力 | FP32 37.4 TFLOPS / TF32 74.8(149.6*) / FP16·BF16 149.7(299.4*) / INT8 299.3(598.6*) TOPS / INT4 598.7(1197.4*) TOPS | — |
| CUDA核心 | 10752 (84 SM) / Tensor Core 336 (三代) / RT Core 84 (二代) | — |
| 功耗 | 300W (被动散热, 8-pin CPU供电) | — |
| 互联 | NVLink 112.5GB/s (双向, 双卡可拼96GB) / PCIe Gen4 x16 (64GB/s) | 2- or 4-way NVIDIA NVLink bridge: 900GB/s per GPU | PCIe Gen5: 128GB/s |
| 发布年 | 2020 (2020-10-05 GTC发布) | — |
| 形态 | Dual-slot, 4.4"(H) x 10.5"(L), Passive | PCIe Dual-slot air-cooled |
| 媒体引擎 | 1 Encode / 2 Decode (含AV1解码) | — |
| 显示输出 | 3x DisplayPort 1.4 (出厂默认禁用, 可管理软件启用) | — |
| MIG | 不支持 | — |
| RT Core性能 | 73.1 TFLOPS | — |
| 显存类型 | — | HBM3e |
| FP64算力 | — | 30 TFLOPS |
| FP64 Tensor Core | — | 60 TFLOPS |
| FP32算力 | — | 60 TFLOPS |
| FP8算力 | — | 3,341 TFLOPS |
| FP16算力 | — | 1,671 TFLOPS |
| INT8算力 | — | 3,341 TFLOPS |
| 功耗TDP | — | Up to 600W (configurable) |
| MIG切片 | — | Up to 7 MIGs @16.5GB each |
| 编解码器 | — | 7 NVDEC | 7 JPEG |
| 机密计算 | — | Supported |