Running Deepseek v4 Flash Q2 on a Single RTX 4090
jack_smirkingrevenge · reddit · 2026-08-15
A user reports successfully running a quantized Deepseek-v4-flash-0731 (Q2/Q3) on a single RTX 4090 with 64GB RAM. By keeping heavily utilized experts in CPU RAM and using Blaze kernels, the setup achieves a usable token rate of around 8 tps, dropping to 5 tps on disk reads due to expert misses. The author plans to publish details on this stack soon.
More from Infra
- CoreWeave's debt costs falling: DDTL spreads tightened from S+9.62% to S+2.25% — BenBajarin · 2026-10-16
- SoftBank lines up US IPOs for SB Energy and AI robotics spinoff Roze, valuations top $50B — nordicinst · 2026-10-16
- TRL reproduces Modal's mixed-precision async RL: plain GRPO collapses, SC+MIS holds 0.72 — QGallouedec · 2026-10-16
- Hugging Face scales Indic pretraining with 2.8T tokens and an open data pipeline — Hugging Face Blog · 2026-10-16
- NVIDIA Vera Rubin NVL72 hits up to 5.7x GB300 throughput per MW on agent benchmark — ArtificialAnlys · 2026-10-16
- 14 open-source LLM observability tools to know in 2026, from Langfuse to Laminar — TheTuringPost · 2026-10-16