H100-SXM-80GB vs RTX-4090-24GB
RTX-4090-24GB is the cheaper of the two: $0.16/hr at Salad, against $2.25/hr for H100-SXM-80GB at TensorDock — 14.1× the price.
Cheapest on each
Aligned facts
Same sheet, same order, both cards — including the gaps.
| Attribute | H100-SXM-80GB | RTX-4090-24GB |
|---|---|---|
| VRAM | 80 GB▲ | 24 GB |
| Bandwidth | 3.35 TB/s▲ | 1.008 TB/s |
| FP16 | 989 TFLOPS▲ | 165 TFLOPS |
| FP8 | 1979 TFLOPS▲ | 330 TFLOPS |
| TDP | 700 W▲ | 450 W |
| NVLink | NVLink 4 | PCIe 4 (no NVLink) |
| Architecture | Hopper | Ada Lovelace |
| Released | 2022 | 2022 |
A blank is not a zero. Where one card states a figure and the other states none, neither wins the row — the second number is unknown, and ranking an unknown as nothing is how a comparison starts to lie.
Every seller, both cards
One row per seller, cheapest tier, on-demand list prices only. Sellers of both cards lead, cheapest combined rate first.
| Seller | H100-SXM-80GB | RTX-4090-24GB |
|---|---|---|
| TensorDock | $2.252026-10-26 | $0.352026-10-26 |
| Vast.ai | $2.402026-10-27 | $0.472026-10-27 |
| Runcrate | $3.252026-10-22 | $0.922026-10-26 |
| RunPod | $3.992026-10-27 | $0.892026-10-27 |
| Oblivus | $4.312026-10-26 | $0.642026-10-26 |
| Thunder Compute | $2.952026-10-26 | ○ no answer |
| Massed Compute | $3.142026-10-26 | ○ no answer |
| Hyperbolic | $3.192026-10-26 | ○ no answer |
| Hyperstack | $3.202026-10-26 | ○ no answer |
| Modal | $3.952026-10-26 | ○ no answer |
| Lambda | $3.992026-10-26 | ○ no answer |
| Verda (DataCrunch) | $4.652026-10-27 | ○ no answer |
| CoreWeave | $6.162026-10-26 | ○ no answer |
| Salad | ○ no answer | $0.162026-10-26 |
| Theta EdgeCloud | ○ no answer | $0.262026-10-26 |
| Packet·ai | ○ no answer | $0.392026-10-27 |
| Shadeform | ○ no answer | $0.402026-10-26 |
| Beam | ○ no answer | $0.442026-10-27 |
| GPUhub | ○ no answer | $0.462026-10-26 |
| Lium | ○ no answer | $0.652026-10-27JSON |
| Spheron | ○ no answer | $0.722026-10-26 |
What this page will not tell you
On-demand list rates only · spot and committed terms priced separately and not mixed in here · no performance-per-dollar figure appears on this page, because it needs a throughput benchmark we do not hold and dividing a price by a vendor’s own TFLOPS claim produces a number that looks authoritative and means nothing