NVIDIA GB200 Superchip 对比 NVIDIA L40 — GPU 比较 (Apr 2026)
NVIDIA GB200 Superchip (384GB HBM3e, 4,500 TFLOPS FP16, Blackwell) vs NVIDIA L40 (48GB GDDR6, 181 TFLOPS FP16, Ada Lovelace)
|
NVIDIA GB200 Superchip
384GB HBM3e · Blackwell
|
NVIDIA L40
48GB GDDR6 · Ada Lovelace
|
||
|---|---|---|---|
| 规格 | |||
| 制造商 | NVIDIA | NVIDIA | |
| 架构 | Blackwell | Ada Lovelace | |
| VRAM | 384 GB HBM3e | 48 GB GDDR6 | |
| 带宽 | 16,000 GB/s | 864 GB/s | |
| FP16 (Tensor) | 4500.0 TFLOPS | 181.0 TFLOPS | |
| FP32 | 150.0 TFLOPS | 90.5 TFLOPS | |
| TDP | 2700W | 300W | |
| 发布年份 | 2024 | 2023 | |
| 细分市场 | Data center | Data center | |
| 适用场景 | Largest-scale AI training multi-trillion parameter models | Inference video processing rendering | |
| 云端价格 | |||
| 最便宜的按需 | — | — | |
| 最便宜的竞价 | — | — | |
| 供应商 | 0 | 0 | |
NVIDIA GB200 Superchip
NVIDIA L40