~/can-i-run/deepseek-v3
DeepSeek

Can I run DeepSeek V3?

Short answer: yes, on a M3 Ultra 512 (512GB) at Q4_K_M. Long answer below.

deepseek-v3LM StudiovLLMMLXoMLX

The math, in one paragraph.

$ ./vrambudget --explain deepseek-v3

DeepSeek V3 has 671B parameters (MoE: 37B active per forward pass, but all 671B must fit in memory). At FP16 that's 1342 GB of raw weights. Quantization shrinks that, but you also need budget for the KV cache (definition), framework overhead, and safety headroom. The rule of thumb: real usable budget on a card is roughly its nameplate VRAM minus 25%. That's how the table below was computed.

What hardware actually fits.

$ grep "fits" gpus.json
FP16/BF16
1342GB
0 GPUs fit
— none in the catalog —
Q8_0
713GB
0 GPUs fit
— none in the catalog —
Q5_K_M
461GB
0 GPUs fit
— none in the catalog —
Q4_K_M
377GB
1 GPU fits
M3 Ultra 512512GB
Q3_K_M
289GB
1 GPU fits
M3 Ultra 512512GB

Pick your path.

$ ls strategies/
Tightest budget

Smallest GPU that fits DeepSeek V3 at any quant: M3 Ultra 512 at Q4_K_M.

Reference quality (FP16)

Lossless inference needs 1342 GB. Pick from multi-GPU only.

Best quality on a 24GB card

None of the showcase quants fit on a 24GB card. Step up.

Tune the math yourself

Open the calculator pre-tuned for DeepSeek V3: ↗ /calc?model=deepseek-v3

See the full model page.

$ ./open
$ openDeepSeek V3671BDeepSeek · ctx 128K$ rankBest GPU for DeepSeek V3see the ranking →ranked by fit quality

Discussion.

$ gh discussion list

// sign in with github to leave a comment. threads live in the repo's discussions tab.