FitsGPU is a GPU memory-planning tool that helps users check whether a machine-learning model or workload can fit on a graphics card before renting computing time. Its browser-based calculators estimate memory requirements by accounting for model weights, precision, context length and KV cache, plus a configurable reserve for runtime overhead. Users can compare estimates across 12 reference GPUs, including NVIDIA T4, RTX, L-series, A-series and H-series cards, and adjust model settings to see how the verdict changes.

The service offers dedicated planning flows for model inference, 3D rendering, fine-tuning and research notebooks, with workload-specific hardware starting points. FitsGPU also includes saved plans, export options, a manual and a memory lab explaining its calculations. Estimates are presented as approximations rather than guarantees, helping users make a more informed hardware choice.

Plans remain in the user’s browser and can be taken to any provider. The site is built on Robinhood Chain and offers optional EVM wallet connection to read a wallet address and native balance; it states that this requires no signature, transfer or email