Drilldown: Recipe

DeepSeek V4.1 Flash on 4x 8 GB (alanoo fidelity build) (1) · DeepSeek V4.1 Flash on 4x 8 GB (CurtisAccelerate TC-MoE) (1) · DeepSeek V4.1 Flash on 4x 8 GB (GNmac) (1) · DeepSeek V4.1 Flash on 4x 8 GB (Morrowmake) (1) · DeepSeek V4.1 Flash on 4x 8 GB plus RTX PRO 6000 (ExLlamaV3 TP5) (1) · DeepSeek V4.1 Flash on 6x 8 GB (Deepinfer-SM80 TP2 x PP3) (1) · DeepSeek V4.1 Flash on 6x 8 GB (TP6) (1) · DeepSeek V4.1 Flash on 8x 8 GB (Deepinfer-SM80 TP4 x PP2) (1) · DeepSeek V4.1 Flash on 8x 8 GB (Zanooda PP8) (1)

DeepSeek-V4.1-Flash-1017 · Qwen3.8-Flash-Next · GLM-5.3-Flash (320B MoE, 18B active) · Qwen3.8-Flash-Next (125B MoE per the recipe) · DeepSeek-V4.1-Flash-1017 (748B in the Maya README) · DeepSeek-V4.1-Flash-1017 heretic (pqhaz abliteration) · GLM-5.3 (full 78-layer DSA MoE, not Flash) · GLM-5.3 (full 78-layer, not Flash) · GLM-5.3-Flash (321B in the Maya README) · MiMo-V2.6-Flash-RL (309B MoE, 15B active; text, image, video and audio input) · Qwen3.8-27B (dense) · Qwen3.8-Flash-Next-Uncensored (refusal-removed, from orcarouter) · Other · None

Official checkpoint (FP8 dense, FP4 experts); software FP8 on sm_80 · AutoRound 3 bpw + MTP: klee100/Qwen3.8-Flash-Next-AutoRound-3bpw-MTP; 95.37 GiB BF16 PLE table on SSD · AutoRound mixed 2/3-bit (klee100, 2.998 effective bpw, 47 GiB on card) plus halt95's FP8 n-gram table in host RAM · AWQ INT4: cyankiwi/Qwen3.8-Flash-Next-AWQ-INT4; 95 GiB BF16 PLE table in host RAM (FP8 table optional) · AWQ W4A16 g32 (leoncca/Qwen3.8-Flash-Next-Uncensored-AWQ-g32); PLE table converted to INT4 and served from SSD · AWQ W4A16: wtdcode/Qwen3.8-Flash-Next-AWQ-W4A16 · EXL3 3.05 bpw: turboderp/GLM-5.3-Flash-exl3 · EXL3 3.05 bpw: turboderp/GLM-5.3-Flash-exl3; DFlash2 drafter converted to EXL3 6 bpw · EXL3 3.5 bpw (LibertAIDAI; routed experts 4 bpw on layers 0-9 and 30-39, 3 bpw on 10-29; attention and shared experts 6 bpw) plus the official Engram shards · EXL3 5.05 bpw; PLE (n-gram) table read from SSD · GLM-5.3-Int4-Int8Mix (INT4 / INT8 mix); KV fp8_ds_mla · GLM-5.3-Int4-Int8Mix-AWQ-g64; drafter incoai/GLM-5.3-DFlash2 · Hybrid: all 40 expert layers at 3 bits (LibertAI 2.9 bpw pack with layers 18-22 grafted from the 3.5 bpw pack), official FP8 dense tensors · Hybrid: LibertAI EXL3 2.9 bpw routed experts (layers 0-39), official FP8/FP4 for attention, shared experts, head, draft and Engram; KV fp8_ds_mla · 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 · Hybrid: routed experts EXL3 2.9 bpw (LibertAIDAI), everything else official FP8/BF16 with the 38 heretic attn.wo_b tensors · Hybrid: routed experts EXL3 2.9 bpw (LibertAIDAI), everything else official FP8/BF16; KV BF16 · Maya GGUF (Maya-S 340 GB or Maya-XS 277 GB; Engram tables stay on SSD) · Maya-M GGUF (116 GB), every expert in VRAM; Maya-S (96.5 GB) and Maya-S24 (94.7 GB) also reported · Mixed INT4 AutoRound (Minachist/Qwen3.8-Flash-Next-INT4-Mixed-AutoRound, about 65 GB) with the full BF16 PLE table on NVMe · NVFP4 (ModelOpt mixed precision): ProCreations/MiMo-V2.6-Flash-RL-NVFP4, run as 4-bit weights with BF16 activations through Marlin · Official checkpoint as downloaded: MXFP4 routed experts as Marlin W4A16, MXFP8 dense as Marlin W8A16; KV fp8 · Official checkpoint: dense MXFP8 through an SM80 Marlin W8A16 path, routed experts MXFP4 through W4A16; KV FP8 · W4A16 (INT4 weights, group size 128, FP16 activations): canada-quant/GLM-5.3-Flash-W4A16-MTP · W4A16 (INT4 weights, Marlin kernels) · W4A16 AutoRound: Intel/Qwen3.8-Flash-Next-W4A16-AutoRound; 95.4 GiB BF16 PLE table streamed from NVMe; KV fp8 e4m3 (software-decoded on SM80) · W4A16 INT4 g128 with the PLE table repacked INT4 g32: zankich/Qwen3.8-Flash-Next-W4A16-Merlin-INT4PLE-FP8KV; KV BF16 · W4A16, about 4.3 bpw, 163 GiB on disk · W4A16: halt95/Qwen3.8-Flash-Next-W4A16-Merlin (AutoRound INT4 g128 experts, INT8 GDN projections, FP8 n-gram table) · Other · None

Custom SM80 vLLM (patched from public commit 79a7108d9) (1) · ExLlamaV3 fork (anoane/exllamav3-anemone) + TabbyAPI fork (anoane/tabbyAPI) (1) · GNmac-lab recipe image 1.2.0 (vLLM with sm_80 patches) plus three patches (1) · Morrowmake/vllm-cmp170hx (bind-mounted sm80 package) plus the author's tensor-core MoE kernels (1) · Morrowmake/vllm-cmp170hx (vLLM fork, container image) (1) · vLLM PR #56201 head (79a7108d) plus a 20-patch sm_80 series; image zanooda/vllm-sm80-ds41f:v41-sm80 (1) · vLLM PR #56201 head (79a7108d) plus the two Deepinfer-SM80 patches and 30 drop-in modules (2) · vLLM with a pinned sm_80 patch stack (built locally from source) (1)

Deepinfer-SM80 2a828fc lineage; vLLM v0.28.1rc0 base (1) · Deepinfer-SM80, vLLM v0.28.1rc0 base (1) · GNmac-lab at 2ad5741ab1; image ds41-cmp170hx:1.2.0-fidelity (1) · recipe 1.0.0, engine a0c57b5237 (1) · recipe 1.2.0 (1) · recipe 31318fd; patches 0001-0020 (1) · reports 0.0.0; patches/public-base-to-runtime.patch (109 files) (1) · tag recipe-09 (309ef6d1); TabbyAPI deepseek-v41 (ce250ff) (1) · vllm-cmp170hx 18305948ac; TC-MoE fb9786ec (1)

8 GB (9)

4 (5) · 6 (2) · 8 (2)

PP8, partition 5,5,5,5,5,5,5,5 (1) · TP2 x PP3, partition 14,14,12 (1) · TP4 (default) or PP4 for x1/x4 links (1) · TP4 with expert parallel (96 whole experts per card) (1) · TP4 x PP2, partition 20,20 (1) · TP4, P2P on (1) · TP4, peer access off (NCCL_P2P_DISABLE=1), host all-reduce (1) · TP5 across four CMP 170HX and one RTX PRO 6000 Blackwell (native backend, reduces through host memory); MTP draft on the PRO (1) · TP6 (virtual-head padding), PP1 (1)

1,048,576 (4) · 1,048,576 FP16 (1) · 262,144 (2) · 262,144 (up to 1,048,576) (1) · 262,144 (weights address 1,048,576) (1)

70 - 120 (4) · 120 - 300 (5)

None (1) · 1,500 - 3,000 (4) · 3,000 - 7,000 (4)

64 GB, --gpu-memory-utilization 0.95 (1) · 64 GB; weights 35.9 GB per rank (ranks 1-6), 37.6 GB rank 0, 46.3 GB rank 7 (1) · 64 GiB (1) · 64 GiB (64,911 MiB usable), GPU_MEMORY_UTILIZATION 0.98 (1) · 64 GiB exposed, --gpu-memory-utilization 0.95 (1) · 64 GiB, --gpu-memory-utilization 0.95 (1) · 64 GiB, GPU_MEMORY_UTILIZATION 0.95 (1) · about 62.5 GB used of 64 GB, --gpu-memory-utilization 0.97 (1) · gpu_split 58 GB per CMP, 70 GB on the PRO (1)

None (3) · 140 W (1) · 150 W per CMP (PRO 400 W) (1) · 180 W (1) · 180 W (nvidia-smi -pl 180) (1) · 180 W (one card unstable at 250 W) (1) · 200 W (1)

None (5) · CUDA 13 capable; measured on 610.43.02 (1) · NVIDIA 580 or newer; P2P needs the Morrowmake/cmpunlocker build (1) · NVIDIA open 610.43.02 from a cmpunlocker fork with a P2P patch (1) · NVIDIA open 610.57.04 patched with Morrowmake/cmpunlocker 1f2c498 (profile 8gb) (1)

None (4) · cmpunlocker (64 GiB, 74 SMs) with a P2P-capable driver for TP4 (1) · cmpunlocker (64 GiB, 74 SMs); Morrowmake/cmpunlocker for P2P (1) · cmpunlocker fork with P2P (without Morrowmake's TRAP31 patch) (1) · cmpunlocker with the BAR1 P2P patches (64 GiB BAR1 per card) (1) · Morrowmake/cmpunlocker 1f2c498; modprobe RmForceEnableGen2=1, RMPcieLinkSpeed=0x1, ForceP2P=0x11 (1)

Proven (1) · Reported (8)

alanoo (1) · anoane (1) · CurtisAccelerate (1) · greencandle9753 (1) · HalfVulpes (2) · voidgiraffe (1) · wgkbfcp (1) · Zanooda (1)

October 19, 2026 (2) · October 20, 2026 (1) · October 22, 2026 (1) · October 24, 2026 (1) · October 26, 2026 (1) · October 27, 2026 (1) · October 28, 2026 (2)

https://github.com/alanoo81/deepseek-v41-flash-4x-cmp170hx (1) · https://github.com/anoane/09-deepseek-v41-flash-1017-exl3-tp5-1m (1) · https://github.com/CurtisAccelerate/DeepSeekv41-Flash-1017-EXL3-CMP170HX (1) · https://github.com/GNmac-lab/deepseek-v41-flash-1017-cmp170hx (1) · https://github.com/HalfVulpes/Deepinfer-SM80 (2) · https://github.com/Morrowmake/dsv41-flash-cmp170hx-recipe (1) · https://github.com/wgkbfcp/deepseek-v4.1-flash-1017-cmp170hx-recipe (1) · https://github.com/Zanooda/deepseek-v4.1-flash-cmp170hx (1)