…
Cloud rental price, compute per dollar and specs, side by side.
Open in the interactive comparer →| Spec / price | B200 SXM | B300 SXM |
|---|---|---|
| Memory | 180 GB HBM3e | 288 GB HBM3ebetter |
| Memory bandwidth | 7,700 GB/s | 8,000 GB/sbetter |
| FP16 / BF16 (dense) | 2.25 PF | 2.25 PF |
| FP8 (dense) | 4.5 PF | 4.5 PF |
| FP4 (dense) | 9 PF | 13.5 PFbetter |
| INT8 (dense) | 4.5 PFbetter | 0.33 PF |
| TF32 (dense) | 1.1 PF | 1.1 PF |
| FP32 (dense) | 0.075 PF | 0.075 PF |
| FP64 (dense) | 0.037 PFbetter | 0.0013 PF |
| Power | 1000 W | 1100 W |
| Architecture | Blackwell (2024) | Blackwell Ultra (2025) |
| Lowest on-demand | $3.74better | $4.50 |
| Median on-demand | $6.80better | $8.37 |
| Lowest spot | $0.99 | $0.99 |
| FP16/BF16 $ per PFLOP·h | $1.66better | $2.00 |
| Providers | 15 | 11 |
The B200 SXM and B300 SXM have the same dense FP16/BF16 compute. The B300 SXM 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 $2.00 per PFLOP-hour at the lowest on-demand prices.
Running one GPU around the clock at the lowest on-demand price costs about $2,730 per month for the B200 SXM and $3,285 for the B300 SXM.
The B300 SXM has more memory: 288 GB vs 180 GB.