AMD Instinct MI350X 对比 NVIDIA GB200 Superchip — GPU 比较 (Apr 2026)
AMD Instinct MI350X (288GB HBM3e, 1,800 TFLOPS FP16, CDNA 4) vs NVIDIA GB200 Superchip (384GB HBM3e, 4,500 TFLOPS FP16, Blackwell)
|
AMD Instinct MI350X
288GB HBM3e · CDNA 4
|
NVIDIA GB200 Superchip
384GB HBM3e · Blackwell
|
||
|---|---|---|---|
| 规格 | |||
| 制造商 | AMD | NVIDIA | |
| 架构 | CDNA 4 | Blackwell | |
| VRAM | 288 GB HBM3e | 384 GB HBM3e | |
| 带宽 | 8,000 GB/s | 16,000 GB/s | |
| FP16 (Tensor) | 1800.0 TFLOPS | 4500.0 TFLOPS | |
| FP32 | 72.0 TFLOPS | 150.0 TFLOPS | |
| TDP | 1000W | 2700W | |
| 发布年份 | 2025 | 2024 | |
| 细分市场 | Data center | Data center | |
| 适用场景 | Next-gen AI training inference at scale | Largest-scale AI training multi-trillion parameter models | |
| 云端价格 | |||
| 最便宜的按需 | — | — | |
| 最便宜的竞价 | — | — | |
| 供应商 | 1 | 0 | |
AMD Instinct MI350X
NVIDIA GB200 Superchip