DeepSeek-V4-Flash on RTX 3090 with 128GB RAM: 12.5 tok/s via --n-cpu-moe
Ok_Ninja7526 · reddit · 2026-08-02
Reddit user OkNinja7526 shares how they ran DeepSeek-V4-Flash-0731 UD-IQ3S on an RTX 3090 24GB with 128GB DDR5 overclocked to 5600MHz. By replacing llama.cpp binaries in text-generation-webui and setting the key parameter --n-cpu-moe 39 (keeping some MoE experts in system RAM), they achieved about 12.5 tok/s. The model requires 136GB memory, relying heavily on CPU and RAM bandwidth.
Related event: DeepSeek V4 Flash Sparks a Wave of Local Deployment Tests(26 posts)→
More from Infra
- NVIDIA quietly delists the entry-level RTX 5050 from its marketplace, following the RTX 5090 — chemist_slime · 2026-10-16
- Grok claims 816B MoE model runs locally on dual M5 Ultra Mac Studios via MLX — grok · 2026-10-16
- Same 1,000 Cloudflare Durable Objects: $4,050/month cut to $22 with one trick — threepointone · 2026-10-16
- NVIDIA ships FP4-quantized Qwen3.8-Flash for local use, trending on Hugging Face — nvidia · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16
- DuoDiscover agents auto-build fast eval proxies, scoring 2.68x TPU kernel speedup — berkeley_ai · 2026-10-16