| 规格 | NVIDIA RTX 5000 Ada Generation | NVIDIA A30 |
|---|
| 架构 | Ada Lovelace (AD102, AD102-850-KAB-A1) | NVIDIA Ampere |
| 显存 | 32GB GDDR6 with ECC | 24 GB HBM2 |
| 显存带宽 | 576GB/s | 933 GB/s |
| 显存位宽 | 256bit | — |
| 算力 | FP32 65.3 TFLOPS (Boost) / RT Core 151.0 TFLOPS / Tensor 1044.4 TFLOPS (注4: Effective FP8 TFLOPS using sparsity) | — |
| CUDA核心 | 12800 (100 SM×128) / Tensor Cores 400 (第四代) / RT Cores 100 (第三代) | — |
| 功耗 | 总板功耗 250W (主动散热, 双槽) | — |
| 互联 | PCIe 4.0 x16 / NVLink: No (官方 Datasheet 规格表明确无) | PCIe Gen4: 64GB/s / Third-gen NVLINK: 200GB/s |
| 发布年 | 2023 (家族 GTC 2023-03 官宣, 2023-08-08 首发开售——RTX 5000/4500/4000 Ada 三卡中第一个 GA) | — |
| 形态 | Dual-slot 双槽全高, 4.4"H x 10.5"L, Active 主动散热 | Dual-slot, full-height, full-length (FHFL) |
| 显示输出 | 4x DisplayPort 1.4a (vGPU 模式下 DP 不激活) | — |
| 编解码 | 2x Encode / 2x Decode (+AV1 编解码) | — |
| VR | Yes (VR Ready, GPUDirect for Video / GPUDirect RDMA / Quadro Sync II 兼容) | — |
| API | DirectX 12 (Shader Model 6.7) / OpenGL 4.6 / Vulkan 1.3 / CUDA 12.2 / OpenCL 3.0 / DirectCompute | — |
| FP32算力 | — | 10.3 teraFLOPS |
| FP64算力 | — | 5.2 teraFLOPS |
| TF32算力 | — | 82 teraFLOPS | 165 teraFLOPS |
| FP16算力 | — | 165 teraFLOPS | 330 teraFLOPS |
| INT8算力 | — | 330 TOPS | 661 TOPS |
| 功耗TDP | — | 165 W |
| MIG | — | 4 GPU instances @ 6GB each |