can i run · seo
No — DeepSeek-V4 Flash won't fit 2x RTX 3090 (48GB) at Q4
DeepSeek-V4 Flash needs 76GB+ and exceeds 2x RTX 3090 (48GB). Try a smaller quant or model.
284B MoE with 13B active and 1M context. Even aggressive UD-IQ2 needs ~75 GB combined — 128 GB unified or multi-GPU + fat RAM.
Alternatives that do fit 2x RTX 3090 (48GB):
faq · straight answers
Can DeepSeek-V4 Flash run on 2x RTX 3090 (48GB)?
DeepSeek-V4 Flash needs 76GB+ and exceeds 2x RTX 3090 (48GB). Try a smaller quant or model. Model context 1024k. Compare with all models or re-check with your exact specs on the bench.
What if it OOMs?
Unload with ollama stop, cap layers --num-gpu 28, cut context to 2k. See the OOM fixer on the homepage.
Q4 or Q8 on this rig?
Start Q4. Bigger model at Q4 beats smaller at Q8. Only go Q8 if 2GB+ headroom remains.
How fast will it be?
Slower than GPU — this combo needs offload. Pick a smaller model for interactive chat.
Is MIT commercial-safe?
Yes — MIT allows commercial use.
also fits 2x RTX 3090 (48GB)
DeepSeek-V4 Flash on other rigs