No — not on this device
DeepSeek V4 Pro 1.6T-A49B at Q4_K_M needs 928.9 GB against 25.6 GB usable, and the shortfall of 903.3 GB is more than 128 GB of system RAM can cover at a tolerable speed. A smaller sibling or a lower quantisation is the honest answer here.
Does not fit
8K context
Q4_K_M · 927.8 GB
MIT
Released 13 Aug 2026
New this month
Not in the Ollama library
The 0813 refresh of the V4 flagship. Included as the honest ceiling; a terabyte of weights at Q4.
The VRAM budget
weights 927.8 GB
Weights 927.8 GB
KV cache @ 8K 0.54 GB
Runtime overhead 0.6 GB
Over budget 903.3 GB past 25.6 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 1632.7 GB | 1633.9 GB | — | ~0.2 | −0.1% ppl | 1608.3 GB over |
| Q6_K | 1260.1 GB | 1261.2 GB | — | ~0.3 | −0.4% ppl | 1235.6 GB over |
| Q5_K_M | 1089.1 GB | 1090.3 GB | — | ~0.3 | −0.8% ppl | 1064.7 GB over |
| Q4_K_M | 927.8 GB | 928.9 GB | — | ~0.3 | −1.9% ppl | 903.3 GB over |
| Q3_K_M | 751.1 GB | 752.2 GB | — | ~0.4 | −5.4% ppl | 726.6 GB over |
| Q2_K | 643.5 GB | 644.6 GB | — | ~0.5 | −15% ppl | 619.0 GB over |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. This model uses multi-head latent attention, so its cache is a compressed latent rather than full K and V.
How to run it
$ llama-server \
-hf deepseek-ai/DeepSeek-V4-Pro-0813:Q4_K_M \
-c 8192 -ngl 1
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
01Download is 927.8 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02With no GPU, thread count matters more than clock. Start at one thread per physical core.
03If it falls back to CPU silently, drop the context first, then step down a quantisation.