| 规格 | NVIDIA GeForce RTX 3090 | NVIDIA A30 |
|---|
| 架构 | Ampere (GA102-300, 82 SM) | NVIDIA Ampere |
| 显存 | 24GB GDDR6X | 24 GB HBM2 |
| 显存带宽 | 936GB/s (19.5Gbps x 384bit) | 933 GB/s |
| 显存位宽 | 384bit | — |
| 算力 | FP32 35.6 TFLOPS (官方发布口径, Wikipedia按Boost核算35.58) / Tensor 142 TFLOPS [284稀疏] | — |
| CUDA核心 | 10496 (82 SM) / Tensor Core 328 (三代) / RT Core 82 (二代) | — |
| 功耗 | 350W TGP (建议系统电源750W) | — |
| 互联 | PCIe Gen4 x16 + NVLink 3.0 双向112.5GB/s (30系首发唯一保留NVLink的消费卡, 与后期3090 Ti; 40系起移除) | PCIe Gen4: 64GB/s / Third-gen NVLINK: 200GB/s |
| 发布年 | 2020 (2020-09-01发布会, 2020-09-24开售, MSRP $1,499) | — |
| 频率 | Base 1.40GHz / Boost 1.70GHz | — |
| 工艺 | Samsung 8N, 283亿晶体管, 628.4mm² (GA102满血84SM:10752CUDA, 3090启用82) | — |
| 媒体引擎 | 第7代NVENC / 第5代NVDEC (官方页标注NVDEC为Decoder) | — |
| 显示输出 | HDMI 2.1 + DP 1.4a x3, 4屏, HDCP 2.3, 7680x4320 | — |
| 形态 | FE三槽 313x138mm, 供电 2x8pin(附1x12-pin转接), 需750W电源 | Dual-slot, full-height, full-length (FHFL) |
| CUDA算力等级 | Compute Capability 8.6 | — |
| 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 |