NVIDIA GH200 Superchip vs NVIDIA L40 — Comparaison GPU (Apr 2026)
NVIDIA GH200 Superchip (96GB HBM3, 989 TFLOPS FP16, Hopper) vs NVIDIA L40 (48GB GDDR6, 181 TFLOPS FP16, Ada Lovelace)
|
NVIDIA GH200 Superchip
96GB HBM3 · Hopper
|
NVIDIA L40
48GB GDDR6 · Ada Lovelace
|
||
|---|---|---|---|
| Spécifications | |||
| Fabricant | NVIDIA | NVIDIA | |
| Architecture | Hopper | Ada Lovelace | |
| VRAM | 96 GB HBM3 | 48 GB GDDR6 | |
| Bande passante | 4,000 GB/s | 864 GB/s | |
| FP16 (Tensor) | 989.0 TFLOPS | 181.0 TFLOPS | |
| FP32 | 494.5 TFLOPS | 90.5 TFLOPS | |
| TDP | 700W | 300W | |
| Année de sortie | 2023 | 2023 | |
| Segment | Data center | Data center | |
| Idéal pour | Large-scale AI training HPC | Inference video processing rendering | |
| Tarification Cloud | |||
| Le moins cher à la demande | — | — | |
| Le moins cher en Spot | — | — | |
| Fournisseurs | 0 | 0 | |
NVIDIA GH200 Superchip
NVIDIA L40