gpu.nz

How much VRAM does this LLM need?

Real GGUF file sizes plus each model's own KV cache, read from its config: 35 open models, any context length, parallel requests, FP16 or Q8 cache.

Parallel requests
KV cache
Total GPU memory
GiB
Weights
KV cache
Runtime

How it is computed

total = weights + kv_cache × requests + runtime

Data: gpunz/llm-vram-requirements (CC BY 4.0). Built by gpu.nz, a GPU cloud billed by the minute.