…
Cloud rental price, compute per dollar and specs, side by side.
Open in the interactive comparer →| Spec / price | RTX 5090 | RTX PRO 6000 Blackwell |
|---|---|---|
| Memory | 32 GB GDDR7 | 96 GB GDDR7better |
| Memory bandwidth | 1,792 GB/s | 1,792 GB/s |
| FP16 / BF16 (dense) | 0.209 PF | 0.504 PFbetter |
| FP8 (dense) | 0.419 PF | 1.01 PFbetter |
| FP4 (dense) | 1.68 PF | 2.02 PFbetter |
| INT8 (dense) | 0.838 PF | 1.01 PFbetter |
| TF32 (dense) | 0.105 PF | 0.126 PFbetter |
| FP32 (dense) | 0.105 PF | 0.126 PFbetter |
| FP64 (dense) | 0.0016 PF | 0.002 PFbetter |
| Power | 575 W | 600 W |
| Architecture | Blackwell (2025) | Blackwell (2025) |
| Lowest on-demand | $0.40better | $0.59 |
| Median on-demand | $0.65better | $1.56 |
| Lowest spot | $0.24better | $0.32 |
| FP16/BF16 $ per PFLOP·h | $1.91 | $1.17better |
| Providers | 6 | 17 |
The RTX PRO 6000 Blackwell has 2.4× more dense FP16/BF16 compute than the RTX 5090. The RTX 5090 and RTX PRO 6000 Blackwell have the same memory bandwidth. Real-world speedups depend on the workload: training and prefill scale with compute, while LLM decoding scales mostly with memory bandwidth.
Right now the RTX PRO 6000 Blackwell is cheaper per unit of compute: $1.17 vs $1.91 per PFLOP-hour at the lowest on-demand prices.
Running one GPU around the clock at the lowest on-demand price costs about $293 per month for the RTX 5090 and $431 for the RTX PRO 6000 Blackwell.
The RTX PRO 6000 Blackwell has more memory: 96 GB vs 32 GB.