…
Cloud rental price, compute per dollar and specs, side by side.
Open in the interactive comparer →| Spec / price | GH200 | H100 SXM |
|---|---|---|
| Memory | 96 GB HBM3better | 80 GB HBM3 |
| Memory bandwidth | 4,000 GB/sbetter | 3,350 GB/s |
| FP16 / BF16 (dense) | 0.989 PF | 0.989 PF |
| FP8 (dense) | 1.98 PF | 1.98 PF |
| FP4 (dense) | — | — |
| INT8 (dense) | 1.98 PF | 1.98 PF |
| TF32 (dense) | 0.495 PF | 0.495 PF |
| FP32 (dense) | 0.067 PF | 0.067 PF |
| FP64 (dense) | 0.067 PF | 0.067 PF |
| Power | 1000 W | 700 W |
| Architecture | Hopper (2023) | Hopper (2022) |
| Lowest on-demand | $1.99 | $1.66better |
| Median on-demand | $2.29better | $3.19 |
| Lowest spot | — | $0.79 |
| FP16/BF16 $ per PFLOP·h | $2.01 | $1.68better |
| Providers | 3 | 28 |
The GH200 and H100 SXM have the same dense FP16/BF16 compute. The GH200 has 1.2× more memory bandwidth than the H100 SXM. Real-world speedups depend on the workload: training and prefill scale with compute, while LLM decoding scales mostly with memory bandwidth.
Right now the H100 SXM is cheaper per unit of compute: $1.68 vs $2.01 per PFLOP-hour at the lowest on-demand prices.
Running one GPU around the clock at the lowest on-demand price costs about $1,453 per month for the GH200 and $1,212 for the H100 SXM.
The GH200 has more memory: 96 GB vs 80 GB.