VRAM & context calculator
Estimate the memory a local LLM needs from its weights, KV cache and overhead, and see which GPUs and Macs can run it.
Estimated memory needed
6.4 GiB
- Weights
- 4.53 GiB
- KV cache
- 1.00 GiB
- Overhead
- 0.82 GiB
KV cache: 128 KiB per token. Overhead is an approximation; real usage varies by runtime and batch settings.
Will it fit on…
- NVIDIA RTX 3060 12GB11 GiB usableFits
- NVIDIA RTX 4060 Ti 16GB15 GiB usableFits
- NVIDIA RTX 4070 12GB11 GiB usableFits
- NVIDIA RTX 3090 24GB23 GiB usableFits
- NVIDIA RTX 4090 24GB23 GiB usableFits
- NVIDIA RTX 5090 32GB30 GiB usableFits
- AMD RX 7900 XTX 24GB22 GiB usableFits
- Apple M2 (16GB)11 GiB usableFits
- Apple M3 Pro (36GB)27 GiB usableFits
- Apple M4 Pro (48GB)36 GiB usableFits
- Apple M3 Max (64GB)48 GiB usableFits
- Apple M4 Max (128GB)96 GiB usableFits
- Apple M3 Ultra (192GB)144 GiB usableFits
- CPU only, dual-channel DDR5 (32GB)26 GiB usableFits
- CPU only, dual-channel DDR4 (16GB)13 GiB usableFits
Learn more