Drilldown: Recipe

DeepSeek V4 Flash 0731 on 3x 8 GB (PP3 DSpark) (1) · DeepSeek V4.1 Flash on 4x 8 GB (alanoo fidelity build) (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 8x 8 GB (Zanooda PP8) (1) · GLM 5.3 Flash on 2x 8 GB (ExLlamaV3 TP2) (1) · GLM 5.3 Flash on 4x 8 GB (kaka86mm EXL3 + Marlin sidecar + DFlash2) (1) · GLM 5.3 Flash on 8x 8 GB (promisezackr PP8 NVFP4 DFlash2) (1) · MiMo V2.6 Flash on 4x 8 GB (1) · Qwen3.8 Flash Next on 1x 8 GB (ElectroHobo 3 bpw) (1) · Qwen3.8 Flash Next on 2x 8 GB (DUFRelic PP2) (1) · Qwen3.8 Flash Next on 2x 8 GB (mannydelrio PP2) (1) · Qwen3.8 Flash Next on 2x 8 GB (Qwen-Flash-SM80, GDS TEP2) (1) · Qwen3.8 Flash Next on 3x or 4x 8 GB (zebgop-ops FP8 PP) (1)

DeepSeek-V4-Flash-0731 (1) · DeepSeek-V4.1-Flash-1017 (4) · GLM-5.3-Flash (1) · GLM-5.3-Flash (320B MoE, 18B active) (1) · GLM-5.3-Flash (320B MoE, 18B active, multimodal) (1) · MiMo-V2.6-Flash-RL (309B MoE, 15B active; text, image, video and audio input) (1) · Qwen/Qwen3.8-Flash-Next-FP8 (revision bcd9f01d) (1) · Qwen3.8-Flash-Next (1) · Qwen3.8-Flash-Next (125B MoE per the recipe) (1) · Qwen3.8-Flash-Next (nvidia/Qwen3.8-Flash-Next-NVFP4, revision fc694b54, 11 shards, 132.68 GB) (1) · Qwen3.8-Flash-Next-Uncensored (refusal-removed, from orcarouter) (1)

AutoRound mixed 2/3-bit (klee100, 2.998 effective bpw, 47 GiB on card) plus halt95's FP8 n-gram table in host RAM (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) · 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 (1) · EXL3 4 bpw (Mia-AiLab/GLM-5.3-Flash-EXL3-TR3-4bpw, 164 GB) plus an offline-generated Marlin INT4 decode sidecar (151 GB) (1) · 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 (1) · Hybrid: routed experts EXL3 2.9 bpw (LibertAIDAI), everything else official FP8/BF16; KV BF16 (1) · LibertAIDAI/GLM-5.3-Flash-NVFP4 (ModelOpt NVFP4, routed experts only, 181 GiB), run through Marlin W4A16 kernels; GLM-5.3-Flash DFlash2 drafter (5 sliding-window layers) (1) · Native MXFP4 experts + FP8 attention (official deepseek-ai checkpoint, about 155.4 GiB); KV cache fp8 (1) · NVFP4 (ModelOpt mixed precision): ProCreations/MiMo-V2.6-Flash-RL-NVFP4, run as 4-bit weights with BF16 activations through Marlin (1) · NVFP4 main weights through Marlin (no native FP4 on sm_80), mixed FP8 and BF16 MTP weights, FP8 PLE table (about 51.2 GB) kept on NVMe (1) · Official block FP8, about 126 GiB including the about 50 GiB FP8 n-gram (PLE) table, which is served from host RAM; KV cache bf16 (the QSA layers refuse fp8) (1) · Official checkpoint as downloaded: MXFP4 routed experts as Marlin W4A16, MXFP8 dense as Marlin W8A16; KV fp8 (1) · W4A16 INT4 g128 with the PLE table repacked INT4 g32: zankich/Qwen3.8-Flash-Next-W4A16-Merlin-INT4PLE-FP8KV; KV BF16 (1)

ExLlamaV3 (patched, pinned) (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 (vLLM fork, container image) (1) · Morrowmake/vllm-cmp170hx image (tag qwen38-recipe-1.0.0) plus one mounted file (vLLM #58235 vision-sink fix) (1) · Morrowmake/vllm-cmp170hx image with patch files mounted read-only (1) · Qwen-Flash-SM80 overlay on vllm/vllm-openai:qwen38-flash-next (digest-pinned), FlashInfer PCIe IPC AllReduce, GDS/cuFile over NVMe P2PDMA (1) · vLLM PR #56201 head (79a7108d) plus a 20-patch sm_80 series; image zanooda/vllm-sm80-ds41f:v41-sm80 (1) · vLLM v0.30.0 + zankich INT4-PLE plugin + six hash-guarded patch layers (1) · vLLM: haosdent/vllm branch dsv4-flash-a100 with a vendored copy of allover326's 7 patches (image dsv4-a100:devel) (1) · vLLM: hyd998877/vllm_170hx_glm53-flash-exl3-optimized with kaka86mm's overlay patches; exllamav3 1.4.8 (1) · vLLM: official vllm/vllm-openai:glm53-flash image plus promisezackr's 24 patches (SM8x base from 344303947/vllm170hx, pipeline parallel for GLM-5.3-Flash, DFlash2 under PP8) (1) · vLLM: the vllm/vllm-openai:qwen38-flash-next image with zebgop-ops patches 0001 to 0016 (never 0008-WITHDRAWN) and three ported files (1) · wtdcode/vllm-backport 3142ded1 plus a two-file patch and two tinybeast PLE plugins (venv, no container) (1)

0.1.dev20073+g8e685d198; repo commit 2f551bb (patch 0016, 2026-10-02) (1) · f8ea5bb-era base with the allover326 patch set as of mid-August 2026 (1) · final production build v30 (numbers measured on v24); repo commit 218b47f (1) · GNmac-lab at 2ad5741ab1; image ds41-cmp170hx:1.2.0-fidelity (1) · recipe 0.2.1; engine 18305948ac (2) · recipe 1.0.0, engine a0c57b5237 (1) · recipe 31318fd; patches 0001-0020 (1) · tag recipe-09 (309ef6d1); TabbyAPI deepseek-v41 (ce250ff) (1) · upstream dev 0f685aa9fd plus 9 patches; repo db01c256 (1) · v0.1.dev20051+g487ecf187 (the image's vLLM); repo commit 90ec72e (1) · v0.3.0 (commit 4d229f1, 2026-10-14); base vLLM 0.1.dev20073+g8e685d198 (1) · vLLM v0.30.0 (ced6857a); plugin zankich/vllm v0.29.0z (6609bfef) (1) · vllm-backport 3142ded1 (1)

8 GB (14)

1 (1) · 2 (4) · 3 (1) · 4 (6) · 8 (2)

PP2 (layer partition 25,23), no P2P (1) · PP2, MTP depth 3 (1) · PP3 (VLLM_PP_LAYER_PARTITION=15,15,13) with DSpark, 5 draft tokens; PCIe Gen2 x4, no P2P (1) · PP4 (VLLM_PP_LAYER_PARTITION=13,11,11,10), DFlash2 k=7; PCIe Gen2 x4, no P2P (1) · PP4, VLLM_PP_LAYER_PARTITION=12,12,12,12, MTP 3, FULL_AND_PIECEWISE CUDA graphs; PCIe Gen2 x4 per card, no P2P. Three cards: PP3 with 16,17,15 or 16,16,16 (1) · PP8 (VLLM_PP_LAYER_PARTITION=7,5,7,5,7,5,7,2) with DFlash2, 7 draft tokens; PCIe Gen2 x4, no P2P; host 2x Xeon E5-2690 v4, 96 GB DDR4 (1) · PP8, partition 5,5,5,5,5,5,5,5 (1) · single card (1) · TEP2 (TP2 plus expert parallel), MTP 6; one card at PCIe Gen2 x16 (x16 mod), one at Gen2 x4; GPU P2P required (1) · TP2 over NCCL (GPU P2P), DFlash2 K7 plus hybrid drafting (1) · TP4 (P2P on); expert parallel off (1) · TP4 with expert parallel (96 whole experts per card) (1) · TP4, P2P on (1) · TP5 across four CMP 170HX and one RTX PRO 6000 Blackwell (native backend, reduces through host memory); MTP draft on the PRO (1)

1,048,576 (2) · 1,048,576 (KV pool 6.62M tokens at utilization 0.90, about 6.3 concurrent 1M requests) (1) · 1,048,576 FP16 (1) · 262,144 (6) · 262,144 native; 524,288 with YaRN 2.0; 1,048,576 with YaRN 4.0 (needle verified on PP3) (1) · 392,960 per request; 655,360-token Q8 KV pool shared by two requests (1) · 524,288 (1) · 691,968 (largest stable context at --gpu-memory-utilization 0.965) (1)

80 - 100 (4) · 100 - 130 (4) · 130 - 300 (6)

None (2) · 1,500 - 2,100 (3) · 2,100 - 4,000 (4) · 4,000 - 12,000 (5)

64 GB (65,536 MiB), --gpu-memory-utilization 0.965 (1) · 64 GB, --gpu-memory-utilization 0.90 (1) · 64 GB, --gpu-memory-utilization 0.93 (0.94 is the ceiling) (1) · 64 GB, --gpu-memory-utilization 0.93 (v0.3.0) (1) · 64 GB, --gpu-memory-utilization 0.94, TB_TORCH_MEM_CAP_GIB=61.5 (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 (64,911 MiB usable), GPU_MEMORY_UTILIZATION 0.98 (1) · 64 GiB exposed, --gpu-memory-utilization 0.95 (1) · 64 GiB, --gpu-memory-utilization 0.90 (1) · 64 GiB, --gpu-memory-utilization 0.95 (fixed) (1) · 64 GiB; --gpu-memory-utilization 0.94 plus per-rank KV budgets; measured 34.0-34.3 GiB per 12-layer rank, 37.9 GiB on the drafter rank (1) · 64 GiB; weights 45.5 GiB plus 10 GiB KV per card; GPU_UTIL 0.94 (fixed) (1) · about 62.9 / 62.3 GiB idle, peaks about 64.4 / 63.0 GiB with vision (1) · gpu_split 58 GB per CMP, 70 GB on the PRO (1)

None (2) · 100 W (1) · 140 W (200 W observed-draw abort guard) (1) · 150 W per CMP (PRO 400 W) (1) · 180 W (4) · 180 W (author's cards); about 170 W with a modest memory OC in his Discord numbers (1) · 180 W (one card unstable at 250 W) (1) · 200 W (3)

None (3) · 580 or newer (measured on 610.57.04) (2) · 610.43.02 (nvidia-driver-610-open) (1) · 610.43.03 (open modules) (1) · 610.57.04 (1) · NVIDIA 580 or newer; P2P needs the Morrowmake/cmpunlocker build (1) · NVIDIA 610.43.02 (1) · NVIDIA open 610.43.02 from a cmpunlocker fork with a P2P patch (1) · NVIDIA open 610.43.03, CUDA 13.0 (1) · NVIDIA open 610.57.04 patched with Morrowmake/cmpunlocker 1f2c498 (profile 8gb) (1) · nvidia-open 610.43.02 + cmpunlocker v0.5 or newer (1)

None · cmpunlocker (64 GiB, 74 SMs) with a P2P-capable build · cmpunlocker (64 GiB, 74 SMs); Morrowmake/cmpunlocker for P2P · amoghmunikote/cmpunlocker 6c442eeb (patch 0016 SM reconfig, 74 SMs) plus Bayley P2P patches 0011/0013/0015 · bayley/cmpunlocker 5a7bb4b, sudo ./install.sh --p2p --no-iommu · cmpunlocker (64 GB) · cmpunlocker (64 GiB) · cmpunlocker (64 GiB); P2P-capable driver for the NCCL backend · cmpunlocker (64 GiB, 74 SMs) · cmpunlocker (64 GiB, 74 SMs) with a P2P-capable driver for TP4 · cmpunlocker 360acd7 (2026-08-06), 8 GB profile · cmpunlocker fork with P2P (without Morrowmake's TRAP31 patch) · cmpunlocker v0.5+ (sudo ./install.sh --profile=8gb --no-iommu) · cmpunlocker VRAM unlock (64 GB per card) · cmpunlocker VRAM unlock (64 GiB per card) · cmpunlocker VRAM unlock (65,536 MiB per card) · cmpunlocker with the BAR1 P2P patches (64 GiB BAR1 per card) · Morrowmake/cmpunlocker 1f2c498; modprobe RmForceEnableGen2=1, RMPcieLinkSpeed=0x1, ForceP2P=0x11 · Other

Reported · Proven · Other · None

alanoo (1) · anoane (1) · dkpoulsen (1) · dufrelic (1) · flipnburner (1) · mrmannyatx (1) · nathang6866 (2) · nestaa51 (1) · nguyenthimy2022kg-alt (1) · promisezackr (1) · tim-odonnell (1) · voidgiraffe (1) · Zanooda (1)

August 2026 (2) · September 2026 (3) · October 2026 (9)

https://github.com/alanoo81/deepseek-v41-flash-4x-cmp170hx (1) · https://github.com/alexvnesta/glm53-flash-cmp170hx-tp (1) · https://github.com/anoane/09-deepseek-v41-flash-1017-exl3-tp5-1m (1) · https://github.com/DUFRelic/qwen38-flash-next-recipe (1) · https://github.com/ElectroHobo/mimo-v26-flash-cmp170hx-recipe (1) · https://github.com/ElectroHobo/qwen38-flash-next-3bpw-1x-cmp170hx-recipe (1) · https://github.com/kaka86mm/glm53-flash-170hx (1) · https://github.com/mannydelrio/qwen38-flash-next-2x-cmp170hx (1) · https://github.com/Morrowmake/dsv41-flash-cmp170hx-recipe (1) · https://github.com/nguyenthimy2022kg-alt/Qwen-Flash-SM80-170HX (1) · https://github.com/promisezackr/glm53-flash-170hx-pp8 (1) · https://github.com/tim-odonnell/cmp170hx-deepseek-v4-flash (1) · https://github.com/Zanooda/deepseek-v4.1-flash-cmp170hx (1) · https://github.com/zebgop-ops/qwen38-flashnext-pp (1)