A GPU rental prices comparison goes wrong when every logo gets one number. OpenSEO's saved keyword, gpu rental prices, is a buyer's query: about 210 monthly searches and difficulty 26. People want a table. A table that mixes a floor, a median, and a contract rate is how renters overpay.
This note keeps three tapes apart. The floor is the cheapest offer on that book. The median is the middle of the book. A reserved cluster is a third product. The dated numbers below are NovaPod's book on November 5, 2026.
GPU rental prices on one book
I am not going to publish a Vast.ai or RunPod hourly figure I did not copy from their market that morning. Their prices move with hosts. So do these.
What I can compare without lying is the unit. NovaPod, Vast.ai, and RunPod bill a running machine down to the second. Lambda and Paperspace are closer to an hour. CoreWeave and Crusoe are closer to capacity you reserve. TensorDock and Spheron are marketplaces again. Genesis Cloud sells instances, with a European footprint.
| Card | Memory | Floor | Median | Offers |
|---|---|---|---|---|
| RTX 4090 | 24 GB | $0.09/hr | $0.35/hr | 419 |
| RTX 5090 | 32 GB | $0.26/hr | $0.46/hr | 648 |
| H100 SXM | 80 GB | $1.21/hr | $1.68/hr | 65 |
| H200 | 141 GB | $2.76/hr | $2.76/hr | 82 |
| B200 | 192 GB | $4.38/hr | Not rented | On the list |
Where a GPU rental prices comparison actually saves money
The 4090 floor was $0.09 and the median was $0.35. Renting the floor and expecting the median experience is how a cheap hour becomes a failed job. Five quiet minutes on the wrong host still cost less than an overnight run, because the clock is per second.
The H100 SXM floor of $1.21 against the 5090 floor of $0.26 is mostly memory: 80 GB against 32 GB. If the model fits in 32 GB, the H100 is not a bargain. If it does not, the 5090 is not a bargain either.
A contract rate at Lambda or CoreWeave can beat a spot floor once the job runs for weeks. I was not buying weeks. I was buying a batch. That is the price advantage I will claim, and the one I will not.
Product differences that change the same hourly ask
NovaPod takes the ask out of a crypto balance: USDT, USDC, SOL, ETH, or BNB. Templates for vLLM, ComfyUI, and PyTorch boot from that same balance. A root KVM machine uses the same GPU at the same ask if you need a disk and SSH.
RunPod's product advantage is serverless next to the pod. Vast.ai's is the width of the host list. Lambda's is a familiar instance if your company already pays by card. Paperspace is a notebook. CoreWeave is a cluster platform. None of those advantages show up if the only column in your sheet is dollars per hour.
Questions about GPU rental prices
Is $0.09 the RTX 4090 price? It was the cheapest NovaPod offer on November 5, 2026. The median was $0.35. Open the marketplace for the offer that is listed now.
Why not rank all ten by price? Because nine of them were not read as a live book that morning. A ranking built from memory is a different article.
Which page has the current ask? The GPU list and the marketplace. This note explains how to read them.
- The November 5 four-card note
- H100 SXM rental price
- RTX 4090 rental price
- Current marketplace asks
- The ten-site overview