DeepSeek V4 Flash 3-bit Quantization Tested on 3x RTX 3090: 119GB VRAM
consultkitapp · reddit · 2026-08-02
A user shared llama-bench benchmark results for the DeepSeek-V4-Flash-0731-UD-Q3KXL quantized model running on 3x RTX 3090 GPUs (Total 72GB VRAM).
Benchmark Data:
- Model Size: 119.40 GiB (approx. 284.33B params).
- Prompt Processing (pp512): 116.04 t/s.
- Text Generation (tg128): 7.71 t/s.
The test utilized 21 GPU offload layers (-ngl 21) and layer split mode (--split-mode layer). The poster noted that the quantization performance could likely be pushed further.
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