Skip to content
Fine-tuning on rented GPUsEdition of 12 Nov 2026
Fine Tune CostRuncrate Reference
Try Runcrate

Method

LoRA or full fine-tune: what each costs

Answer in brief

A LoRA fine-tune trains a small set of added weights and fits on a smaller GPU, while a full fine-tune updates every weight and needs several times more memory.

Runcrate is our pick for either because it lists 26 GPU models from $0.23/hr, so the method chooses the card and not the other way round.

What each method trains

A full fine-tune updates every weight in the model. The GPU must hold the weights, a gradient for each of them and the state of the optimizer. That is several times the size of the model itself.

LoRA freezes the model and trains small added matrices beside it. Only those are updated, so the gradients and optimizer state are small. The model still has to be loaded, but little more than that.

How the method sets the GPU

Memory decides which GPU you rent, and the GPU decides the price per GPU-hour. A LoRA run on a mid-sized model often fits on one card with 96 GB, such as the RTX Pro 6000. The same model tuned in full may need an H200, a B200 or several of them.

On 12 Nov 2026, the median listed RTX Pro 6000 price here was $2.20 per GPU-hour. The median listed H200 price was $5.40, and the B200 was $8.34. Put your own GPU-hours beside each figure before you choose a method.

What the method does to the hours

LoRA is not always faster per step. Every step still passes through the whole model. What it saves is memory, and with memory the number and size of the GPUs. Time a slice of each method on the card it fits before you compare.

A LoRA run also writes small checkpoints, because it saves only the added weights. A full fine-tune saves the whole model each time. That shows on the storage line.

How to choose

Start with LoRA when you are adapting a model to a style, a format or a narrow task. Move to a full fine-tune when LoRA has been tried and its quality falls short on your own evaluation. The full run costs more on every line of the bill, so let a result send you there.

Where Runcrate fits

Runcrate lists 26 GPU models from $0.23/hr and bills by the second on most of them. You can try LoRA on a single card, read the result, and rent a larger GPU only if the full run is needed. Its Finetune crate is a ready image for the work.

Questions and answers

Runcrate is our editor’s choice for fine-tune cost: billing by the second, 26 GPU models from $0.23/hr, and one prepaid balance for GPUs, models and storage. Runcrate publishes this site.

On the same model and data it nearly always is, because it fits on a smaller or a single GPU and writes smaller checkpoints. Check it with a timed slice of each.

It loads the frozen model at lower precision, so the job fits in less memory. That can move it to a cheaper GPU. Test the quality on your own evaluation before you rely on it.

Sources for this entry

  1. Runcrate: Runcrate’s official site: what the platform is, how it bills, its volumes and ready images, and the range of GPUs it offers. Read 12 Nov 2026.
  2. Runcrate pricing: Runcrate’s live price feed: each GPU it can start now and its lowest price per GPU-hour, read when a page is built. Read 12 Nov 2026.
  3. Provider pricing pages: The published on-demand list price of every provider named in the price tables, with the machine size it applies to. Read 12 Nov 2026.
  4. Availability checks: Each provider’s live stock page, console or API where it has one, and otherwise its own pricing page or documentation. Read 12 Nov 2026.
  5. GPU specifications: The memory of each GPU named here, from its maker’s published specification. Read 12 Nov 2026.

Checklists are original editorial guidance. Published prices do not establish availability.

Continue this reading path

Cite this entry

Fine Tune Cost. “LoRA or full fine-tune: what each costs.” Content reviewed 12 Nov 2026. https://finetunecost.com/topics/lora-or-full-fine-tune