Will my GPU run it?

Approximate VRAM fit for each accelerator × model size, at a given quant and context. Estimated from weights + KV cache — the one question every local-AI builder asks, in one matrix.

Quant Context

✓ fits≈ tight✕ won't fit

Accelerator	VRAM	1B	3B	8B	14B	32B	70B	120B	180B
Apple M3 Ultra	512 GB	✓	✓	✓	✓	✓	✓	✓	✓
Apple M2 Ultra	192 GB	✓	✓	✓	✓	✓	✓	✓	✓
NVIDIA H200	141 GB	✓	✓	✓	✓	✓	✓	✓	✓
Apple M3 Max	128 GB	✓	✓	✓	✓	✓	✓	✓	✓
Apple M4 Max	128 GB	✓	✓	✓	✓	✓	✓	✓	✓
AMD EPYC 7352 24-Core Processor (znver2)	126 GB	✓	✓	✓	✓	✓	✓	✓	✓
NVIDIA RTX PRO 6000 Blackwell Server Edition	95 GB	✓	✓	✓	✓	✓	✓	✓	✕
NVIDIA RTX PRO 6000 Blackwell Workstation Edition	95 GB	✓	✓	✓	✓	✓	✓	✓	✕
NVIDIA A100 80GB	80 GB	✓	✓	✓	✓	✓	✓	✓	✕
NVIDIA H100	80 GB	✓	✓	✓	✓	✓	✓	✓	✕
NVIDIA RTX 6000 Ada Generation	48 GB	✓	✓	✓	✓	✓	✓	✕	✕
NVIDIA RTX A6000	48 GB	✓	✓	✓	✓	✓	✓	✕	✕
NVIDIA L40S	44 GB	✓	✓	✓	✓	✓	≈	✕	✕
AMD Instinct MI50 32GB	32 GB	✓	✓	✓	✓	✓	✕	✕	✕
NVIDIA RTX 5090	32 GB	✓	✓	✓	✓	✓	✕	✕	✕
NVIDIA Tesla V100	32 GB	✓	✓	✓	✓	✓	✕	✕	✕
NVIDIA RTX 3090	24 GB	✓	✓	✓	✓	✓	✕	✕	✕
NVIDIA RTX 3090 Ti	24 GB	✓	✓	✓	✓	✓	✕	✕	✕
NVIDIA RTX 4090	24 GB	✓	✓	✓	✓	✓	✕	✕	✕
NVIDIA Tesla P40	24 GB	✓	✓	✓	✓	✓	✕	✕	✕
Quadro RTX 6000	23 GB	✓	✓	✓	✓	✓	✕	✕	✕
NVIDIA GeForce RTX 4060 Ti	16 GB	✓	✓	✓	✓	✕	✕	✕	✕
NVIDIA RTX 4080	16 GB	✓	✓	✓	✓	✕	✕	✕	✕
NVIDIA RTX 3080	10 GB	✓	✓	✓	✓	✕	✕	✕	✕
AMD Radeon RX 6650 XT	8 GB	✓	✓	✓	✕	✕	✕	✕	✕
Apple M1	8 GB	✓	✓	✓	✕	✕	✕	✕	✕
NVIDIA GeForce RTX 2070 Super	8 GB	✓	✓	✓	✕	✕	✕	✕	✕
NVIDIA GeForce RTX 3070 Ti	8 GB	✓	✓	✓	✕	✕	✕	✕	✕

Estimates use effective bits-per-weight per quant + a KV-cache approximation; real fit varies with architecture, attention type, and runtime. Apple unified memory is shown as total RAM (usable VRAM is typically ~75%). Treat ≈/✕ near the boundary as “try it.”