…
Cloud rental price, compute per dollar and specs, side by side.
Open in the interactive comparer →| Spec / price | RTX 4090 | RTX 5090 |
|---|---|---|
| Memory | 24 GB GDDR6X | 32 GB GDDR7better |
| Memory bandwidth | 1,008 GB/s | 1,792 GB/sbetter |
| FP16 / BF16 (dense) | 0.165 PF | 0.209 PFbetter |
| FP8 (dense) | 0.33 PF | 0.419 PFbetter |
| FP4 (dense) | — | 1.68 PF |
| INT8 (dense) | 0.661 PF | 0.838 PFbetter |
| TF32 (dense) | 0.0826 PF | 0.105 PFbetter |
| FP32 (dense) | 0.0826 PF | 0.105 PFbetter |
| FP64 (dense) | 0.0013 PF | 0.0016 PFbetter |
| Power | 450 W | 575 W |
| Architecture | Ada Lovelace (2022) | Blackwell (2025) |
| Lowest on-demand | $0.34better | $0.46 |
| Median on-demand | $0.42better | $0.69 |
| Lowest spot | $0.20 | $0.19better |
| FP16/BF16 $ per PFLOP·h | $2.06better | $2.21 |
| Providers | 8 | 9 |
The RTX 5090 has 1.3× more dense FP16/BF16 compute than the RTX 4090. The RTX 5090 has 1.8× more memory bandwidth than the RTX 4090. 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 4090 is cheaper per unit of compute: $2.06 vs $2.21 per PFLOP-hour at the lowest on-demand prices.
Running one GPU around the clock at the lowest on-demand price costs about $248 per month for the RTX 4090 and $338 for the RTX 5090.
The RTX 5090 has more memory: 32 GB vs 24 GB.