| 规格 | NVIDIA A10 | NVIDIA T4 |
|---|
| 架构 | Ampere (GA102) | Turing (TU104-895) |
| 显存 | 24GB GDDR6 | 16GB GDDR6 with ECC (默认开启) |
| 显存带宽 | 600GB/s | 320GB/s (早期一页版datasheet口径300GB/s) |
| 显存位宽 | 384bit | 256bit |
| 算力 | FP32 31.2 TFLOPS / TF32 62.5(125*) / FP16·BF16 125(250*) / INT8 250(500*) TOPS / INT4 500(1000*) TOPS | FP32 8.1 TFLOPS / FP16·FP32混合精度 65 TFLOPS / INT8 130 TOPS / INT4 260 TOPS |
| CUDA核心 | 72 SM : 9216 CUDA cores : 288 Tensor Cores | 2,560 (40 SM) / Tensor Core 320 / RT Core 40 (白皮书口径) |
| 功耗 | 150W | 70W (Total board power, 默认) |
| 互联 | PCIe Gen4 x16 (64GB/s) | PCIe Gen3 x16 (支持x8运行, Interconnect BW 32GB/s) |
| 发布年 | 2021 (2021-04-12 GTC发布) | 2018 (Product Brief v01 2018-10-24初版, Dec18版datasheet © 2018 NVIDIA) |
| 形态 | Single-slot, full-height, full-length (FHFL) | 单槽半高半长 Low-Profile 6.6英寸(长169.53mm), 被动散热, 板重301g |
| 媒体引擎 | 1 Encode / 2 Decode / AV1 Decode | 1080p30解码 H.264 32路 / HEVC 44路 / VP9 32路 (白皮书Table 2, 官方页称38路full-HD) |
| 时钟 | — | Base 585MHz / Boost 1590MHz (Base为70W功耗墙下调校) |
| SR-IOV | — | Supported; 16 VF（支持） |
| 工艺 | — | 12nm FFN, 136亿晶体管, 545mm2 (白皮书TU104口径) |
| Compute APIs | — | CUDA, NVIDIA TensorRT, ONNX |