LLM VRAM and GPU cost calculator

Pick a model and a mode. See the memory it needs and the cheapest public on-demand GPU rental that fits, from our sourced price table. Prices checked Nov 9, 2026. Comparing a commit? Try the reserved vs on-demand calculator. Renting GPUs vs paying an inference API? Try the self-host vs API calculator.

Calculator

Estimate memory and rental cost

Estimated VRAM: Pick a model; Weights plus 20%
Cheapest option: Pick a model; Per hour, public on-demand

GPU options for the chosen model and mode
GPU	Memory each	GPUs needed	Provider and price	Per hour	For your hours	Source

Choose a model to see options.

How it works. Memory = parameters x bytes per parameter x 1.2. Bytes: FP16 2, FP8 1, INT4 0.5, LoRA 2, QLoRA 0.5, full fine-tune 16 (2 weights, 2 gradients, 12 Adam bytes). It is a rule of thumb: long contexts, big batches and the framework change real use. Listings sold as a whole node must be rented whole, and the cost counts that. Only on-demand prices rated high or medium confidence are used.

Model guides

Static guides for each model

Llama 3.1 8B Llama 3.1 70B Llama 3.3 70B Llama 3.1 405B Llama 4 Scout Mistral 7B Mixtral 8x7B Qwen2.5 7B Qwen2.5 32B Qwen2.5 72B Qwen3 32B Qwen3 30B-A3B Gemma 2 27B DeepSeek V3 gpt-oss-120b gpt-oss-20b All models

Need this capacity?

Tell us the GPU, count, term, region and budget. We reply with the public prices that match and the checks to run before you pay. Free, and we do not sell compute.

Send a GPU request