Yes — with 44.1 GB to spare

DeepSeek V4 Flash 284B-A13B at Q4_K_M fits your CPU only · DDR5 8-channel server entirely on the GPU at 8K context, at an estimated 7.8 tokens per second. Past 942K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.

Fully on GPU 8K context Q4_K_M · 159.7 GB MIT Released 31 Jul 2026 New this month Not in the Ollama library

The V4 that 128 GB machines can actually run at Q3. Cache is modelled as a 576-wide latent; V4 compresses it further at long context, so this is conservative.

What hardware do I need for DeepSeek V4 Flash 284B-A13B? →

The VRAM budget

weights 159.7 GB
Weights 159.7 GB KV cache @ 8K 0.38 GB Runtime overhead 0.6 GB Free 44.1 GB of 204.8 GB

Quantisation ladder

QuantWeightsTotal @ 8KMax contextTok/sQualityFit
Q8_0 281.0 GB 282.0 GB ~4.4 −0.1% ppl 77.2 GB over
Q6_K 216.9 GB 217.9 GB ~5.8 −0.4% ppl 13.1 GB over
Q5_K_M 187.5 GB 188.4 GB 354K 6.7 −0.8% ppl Long context
Q4_K_M 159.7 GB 160.7 GB 942K 7.8 −1.9% ppl Recommended
Q3_K_M 129.3 GB 130.3 GB 1024K 9.7 −5.4% ppl Long context
Q2_K 110.8 GB 111.7 GB 1024K 11 −15% ppl Long context

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

terminal
$ llama-server \
    -hf deepseek-ai/DeepSeek-V4-Flash-0731:Q4_K_M \
    -c 8192 -ngl 99

The engine underneath most of the others. Every knob is exposed. More on llama.cpp.

01Download is 159.7 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.
03There is room to go to 942K context on this card.
See all models for this rig Compare with another model