Drilldown: Recipe

DeepSeek V4.1 Flash on 4x 8 GB (GNmac) (1) · Qwen3.8 Flash Next on 2x 8 GB (DUFRelic PP2) (1)

DeepSeek-V4.1-Flash-1017 (1) · Qwen3.8-Flash-Next (1)

Hybrid: routed experts EXL3 (mostly 3 bpw, 2 bpw on layers 18-22) from LibertAI's 2.9 bpw pack; attention, shared experts, indexer and DSpark drafter official FP8; KV fp8_ds_mla (1) · W4A16 INT4 g128 with the PLE table repacked INT4 g32: zankich/Qwen3.8-Flash-Next-W4A16-Merlin-INT4PLE-FP8KV; KV BF16 (1)

vLLM v0.30.0 + zankich INT4-PLE plugin + six hash-guarded patch layers (1) · vLLM with a pinned sm_80 patch stack (built locally from source) (1)

recipe 1.2.0 (1) · vLLM v0.30.0 (ced6857a); plugin zankich/vllm v0.29.0z (6609bfef) (1)

8 GB · Other · None

2 (1) · 4 (1)

PP2, MTP depth 3 (1) · TP4 (default) or PP4 for x1/x4 links (1)

262,144 (1) · 262,144 (up to 1,048,576) (1)

133.39 (1) · 258 (1)

2,416 (1) · 6,213.66 (1)

64 GiB, --gpu-memory-utilization 0.90 (1) · 64 GiB, GPU_MEMORY_UTILIZATION 0.95 (1)

None · 180 W · 200 W · 140 W · 140 W (200 W observed-draw abort guard) · 150 W per CMP (PRO 400 W) · 170 W (author's 10-card run) · 180 W (author's cards); about 170 W with a modest memory OC in his Discord numbers · 180 W (nvidia-smi -pl 180) · 180 W (one card unstable at 250 W) · Other

CUDA 13 capable; measured on 610.43.02 (1) · NVIDIA 610.43.02 (1)

amoghmunikote/cmpunlocker 6c442eeb (patch 0016 SM reconfig, 74 SMs) plus Bayley P2P patches 0011/0013/0015 (1) · cmpunlocker (64 GiB, 74 SMs) with a P2P-capable driver for TP4 (1)

Proven (1) · Reported (1)

dufrelic (1) · greencandle9753 (1)

October 3, 2026 (1) · October 26, 2026 (1)

https://github.com/DUFRelic/qwen38-flash-next-recipe (1) · https://github.com/GNmac-lab/deepseek-v41-flash-1017-cmp170hx (1)