Cheapest GPU to fine-tune a 7B model.
Prices checked
RTX A5000 · Community Cloud at RunPod does this job for $0.96 per run, the cheapest of 148 machines that can, checked .
Cheapest 10
all 162 rentals- 1
RTX A5000 · Community Cloud
Marketplace
RunPod · 1x RTX A5000 · 24 GB$0.96per run - 2
RTX 3090 · Community Cloud
MarketplaceRunPod · 1x RTX 3090 · 24 GB$1.32per run - 3
RTX A5000 · Secure Cloud
Own machinesRunPod · 1x RTX A5000 · 24 GB$1.62per run - 4
RTX A6000 · Community Cloud
MarketplaceRunPod · 1x RTX A6000 · 48 GB$1.98per run - 5
RTX 4090 · Community Cloud
MarketplaceRunPod · 1x RTX 4090 · 24 GB$2.04per run - 6
A40 · Community Cloud
MarketplaceRunPod · 1x A40 · 48 GB$2.10per run - 7
L4 · Community Cloud
MarketplaceRunPod · 1x L4 · 24 GB$2.64per run - 8
RTX 3090 · Secure Cloud
Own machinesRunPod · 1x RTX 3090 · 24 GB$3per run - 9
A40 · Secure Cloud
Own machinesRunPod · 1x A40 · 48 GB$3.54per run - 10
RTX A6000 · Secure Cloud
Own machinesRunPod · 1x RTX A6000 · 48 GB$3.54per run
Other ways to pick
What matters
- 24 GB of memory
- A 7B model in 16-bit weights is about 14 GB. The adapter, the optimiser state and a batch of training data go on top, which is what the rest of the 24 GB is for.
- One GPU is enough
- Nothing about this job needs two cards talking to each other, so the interconnect does not matter and a single-GPU instance is the cheapest way to buy it.
- Billing by the second or the minute
- A run this short is where fine-grained billing pays: an hour's minimum on a six-hour job is a rounding error, on a twenty-minute one it is not.
- A disk you can write to
- The weights have to be downloaded and the result written back. Several providers include no disk at all in the hourly price, which the row says.
How we count
A LoRA fine-tune of a 7-billion-parameter model fits on one card with 24 GB of memory, which puts most of this market in reach. We count only machines with 24 GB of GPU memory or more on a single board, and 148 of the 162 rentals we read qualify. The cheapest of them is $0.96 per run at RTX A5000 · Community Cloud, which rents for $0.16 per GPU-hour. They are ranked by what the work costs rather than by the hourly rate, because a faster card at twice the price can still be the cheaper way to buy the job.
A LoRA fine-tune of a 7-billion-parameter model is counted as 6 hours on the machine you rent. That is a working figure for planning, not something measured here, and a full fine-tune rather than a LoRA is several times longer. Each price is that work at the machine's on-demand rate: RTX A5000 · Community Cloud rents for $0.16 an hour, and $0.96 per run is what the work costs on it.
- Marketplace
- The provider matches you with someone else's machine.
- Own machines
- The provider owns or leases the machines.
Frequently asked
- What is the cheapest GPU to fine-tune a 7b model?
- RTX A5000 · Community Cloud at RunPod, $0.96 per run. That is 1x RTX A5000 with 24 GB of memory each, renting for $0.16 per GPU-hour. Read on 1 Nov 2026.
- How is the cost worked out?
- A LoRA fine-tune of a 7-billion-parameter model is counted as 6 hours on the machine you rent. That is a working figure for planning, not something measured here, and a full fine-tune rather than a LoRA is several times longer.
- Why do only 148 machines qualify?
- The job needs 24 GB of GPU memory or more on a single board. The other 14 rentals we read are still in the index; they just cannot do this job as described. Serverless rates are left out entirely, because they buy a container that runs when it is called rather than a machine you hold.
- Can I pay less on spot?
- 64 of the 148 qualifying machines publish an interruptible rate, and each of those rows shows what the same work would cost on it if the run were never preempted. It can be, so treat the spot figure as a floor rather than a quote.
- How current are these prices?
- Prices are read every 3 days from 10 providers' own public pages and keyless APIs, and this ranking recomputes with them. Last refresh: 1 Nov 2026.