Free GPU calculators

Size the model first, then compare the price

Two questions decide which GPU server to rent: how much memory your model needs, and whether a monthly price beats an hourly one at your usage. The two free GPU calculators answer them in that order.

1. Size it with the VRAM calculator. Pick a model or enter its size, choose the precision (BF16, FP8 or 4-bit), the context length and the number of concurrent requests. It adds up weights, KV cache and runtime overhead and lists every configuration that fits, with its monthly price.
2. Budget it with the GPU cost calculator. Enter any hourly price and how many hours a day the GPU will really run. It shows the monthly cost at that rate, the break-even hours against a monthly server and the cost per hour you actually use.

Both run in your browser, need no account and send nothing to us. For the reasoning behind the numbers, read how much VRAM an LLM needs and monthly vs hourly GPU pricing.

How accurate the estimates are

The VRAM figures come from the model's architecture: parameters times bytes per parameter for the weights, plus a KV cache computed from layers, KV heads and head size, plus a margin for activations and the runtime. Real usage varies with the serving engine, batch size and memory settings, so treat the result as a sizing guide and keep 10 to 20% headroom. The cost calculator is exact arithmetic on the prices and hours you enter.