What LLMs can my GPU run?

Pick your GPU or enter your VRAM and Will It Run It instantly lists every local model that fits, with estimated tokens/sec, the recommended quantization, and a memory breakdown.

How much VRAM do I need to run a model locally?

Memory needed = quantized weights + KV cache (sized to your context) + a runtime buffer. The tool computes this per model and quant and shows whether it fits fully on the GPU or needs system-RAM offload.

Are the tokens-per-second numbers accurate?

They are physics-based estimates derived from memory bandwidth and bytes read per token (active parameters for MoE models), validated against real benchmarks. Real-world speed varies with prompt length, backend, and thermals.

~/llm will-it-run-it

Will it run it?

Pick your GPU or type your VRAM, and instantly see which open AI models your machine can actually run — how fast they'll generate, the best quantization, and a one-click Hugging Face download for each.

31Models

41GPUs

VRAM · KV-cache · bandwidthPhysics engine

Your hardware

Select GPU

— OR —

VRAM / usable memory (GB)

Bandwidth drives the tokens/sec estimate. Auto-fills when you pick a GPU.

System RAM for offload (GB, optional)

Lets models spill into system memory — runs bigger models, slower. Unified-memory chips pay no penalty.

Test any model

Hugging Face repo or model name

Paste any repo — we'll read its size (and config when available) and estimate the fit on your hardware.

Filters

Minimum context

Minimum speed (tokens/sec)

Required capabilities

Tuning

Quantization

KV-cache precision

Lower precision = far less memory at long context, with a small quality cost.

Runtime / engine

GPUs

Concurrency

Results

Select a GPU or enter your VRAM to begin.

Sort