…
Cloud rental price, compute per dollar and specs, side by side.
Open in the interactive comparer →| Spec / price | GB200 | B200 SXM |
|---|---|---|
| Memory | 186 GB HBM3ebetter | 180 GB HBM3e |
| Memory bandwidth | 8,000 GB/sbetter | 7,700 GB/s |
| FP16 / BF16 (dense) | 2.5 PFbetter | 2.25 PF |
| FP8 (dense) | 5 PFbetter | 4.5 PF |
| FP4 (dense) | 10 PFbetter | 9 PF |
| INT8 (dense) | 5 PFbetter | 4.5 PF |
| TF32 (dense) | 1.25 PFbetter | 1.1 PF |
| FP32 (dense) | 0.08 PFbetter | 0.075 PF |
| FP64 (dense) | 0.045 PFbetter | 0.037 PF |
| Power | 1200 W | 1000 W |
| Architecture | Blackwell (2024) | Blackwell (2024) |
| Lowest on-demand | $16.0 | $3.74better |
| Median on-demand | $21.5 | $6.80better |
| Lowest spot | — | $0.99 |
| FP16/BF16 $ per PFLOP·h | $6.40 | $1.66better |
| Providers | 2 | 15 |
The GB200 has 1.1× more dense FP16/BF16 compute than the B200 SXM. The GB200 has 1.0× more memory bandwidth than the B200 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 B200 SXM is cheaper per unit of compute: $1.66 vs $6.40 per PFLOP-hour at the lowest on-demand prices.
Running one GPU around the clock at the lowest on-demand price costs about $11,680 per month for the GB200 and $2,730 for the B200 SXM.
The GB200 has more memory: 186 GB vs 180 GB.