Drilldown: Recipe

GLM 5.3 Flash on 2x 8 GB (ExLlamaV3 TP2) (1) · Qwen3.8 Flash Next on 2x 8 GB (DUFRelic PP2) (1) · Qwen3.8 Flash Next on 2x 8 GB (ElectroHobo W4A16 TP2) (1) · Qwen3.8 Flash Next on 2x 8 GB (mannydelrio PP2) (1)

GLM-5.3-Flash (320B MoE, 18B active) (1) · Qwen3.8-Flash-Next (1) · Qwen3.8-Flash-Next (125B MoE per the recipe) (1) · Qwen3.8-Flash-Next-Uncensored (refusal-removed, from orcarouter) (1)

AWQ W4A16 g32 (leoncca/Qwen3.8-Flash-Next-Uncensored-AWQ-g32); PLE table converted to INT4 and served from SSD (1) · EXL3 3.05 bpw: turboderp/GLM-5.3-Flash-exl3; DFlash2 drafter converted to EXL3 6 bpw (1) · W4A16 INT4 g128 with the PLE table repacked INT4 g32: zankich/Qwen3.8-Flash-Next-W4A16-Merlin-INT4PLE-FP8KV; KV BF16 (1) · W4A16: halt95/Qwen3.8-Flash-Next-W4A16-Merlin (AutoRound INT4 g128 experts, INT8 GDN projections, FP8 n-gram table) (1)

ExLlamaV3 (patched, pinned) (1) · Morrowmake/vllm-cmp170hx image (Qwen EXL3 release) with patch files mounted read-only (1) · vLLM v0.30.0 + zankich INT4-PLE plugin + six hash-guarded patch layers (1) · wtdcode/vllm-backport 3142ded1 plus a two-file patch and two tinybeast PLE plugins (venv, no container) (1)

recipe 0.3.1; engine 18305948ac (1) · upstream dev 0f685aa9fd plus 9 patches; repo db01c256 (1) · vLLM v0.30.0 (ced6857a); plugin zankich/vllm v0.29.0z (6609bfef) (1) · vllm-backport 3142ded1 (1)

8 GB (4)

2 (4)

PP2 (layer partition 25,23), no P2P (1) · PP2, MTP depth 3 (1) · TP2 over NCCL (GPU P2P), DFlash2 K7 plus hybrid drafting (1) · TP2 with whole-expert expert parallel (1)

262,144 (3) · 392,960 per request; 655,360-token Q8 KV pool shared by two requests (1)

107.67 (1) · 133.39 (1) · 177.8 (1) · 349.7 (1)

1,533 (1) · 4,696 (1) · 6,213.66 (1) · 7,006 (1)

64 GB, --gpu-memory-utilization 0.94, TB_TORCH_MEM_CAP_GIB=61.5 (1) · 64 GiB, --gpu-memory-utilization 0.90 (1) · 64 GiB, GPU_UTIL 0.95 (fixed) (1) · about 62.9 / 62.3 GiB idle, peaks about 64.4 / 63.0 GiB with vision (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

580 or newer (measured on 610.57.04); open modules with HMM for the table-on-disk mode (1) · 610.57.04 (1) · NVIDIA 610.43.02 (1) · nvidia-open 610.43.02 + cmpunlocker v0.5 or newer (1)

amoghmunikote/cmpunlocker 6c442eeb (patch 0016 SM reconfig, 74 SMs) plus Bayley P2P patches 0011/0013/0015 (1) · cmpunlocker (64 GiB); P2P-capable driver for the NCCL backend (1) · cmpunlocker (64 GiB, 74 SMs) with a P2P-capable build (1) · cmpunlocker v0.5+ (sudo ./install.sh --profile=8gb --no-iommu) (1)

Proven (1) · Reported (3)

dufrelic (1) · mrmannyatx (1) · nathang6866 (1) · nestaa51 (1)

October 3, 2026 (1) · October 13, 2026 (1) · October 15, 2026 (1) · October 23, 2026 (1)

https://github.com/alexvnesta/glm53-flash-cmp170hx-tp (1) · https://github.com/DUFRelic/qwen38-flash-next-recipe (1) · https://github.com/ElectroHobo/qwen38-flash-next-w4a16-2x-cmp170hx-recipe (1) · https://github.com/mannydelrio/qwen38-flash-next-2x-cmp170hx (1)