Drilldown: Recipe

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) · GLM 5.3 Flash on 8x 8 GB (promisezackr PP8 NVFP4 DFlash2) (1)

DeepSeek-V4.1-Flash-1017 (3) · GLM-5.3-Flash (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) · Official checkpoint (FP8 dense, FP4 experts); software FP8 on sm_80 (2) · Official checkpoint: dense MXFP8 through an SM80 Marlin W8A16 path, routed experts MXFP4 through W4A16; KV FP8 (1)

Custom SM80 vLLM (patched from public commit 79a7108d9) (1) · vLLM PR #56201 head (79a7108d) plus the two Deepinfer-SM80 patches and 30 drop-in modules (2) · 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)

Deepinfer-SM80 2a828fc lineage; vLLM v0.28.1rc0 base (1) · Deepinfer-SM80, vLLM v0.28.1rc0 base (1) · reports 0.0.0; patches/public-base-to-runtime.patch (109 files) (1) · v0.1.dev20051+g487ecf187 (the image's vLLM); repo commit 90ec72e (1)

8 GB (4)

6 (2) · 8 (2)

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) · TP2 x PP3, partition 14,14,12 (1) · TP4 x PP2, partition 20,20 (1) · TP6 (virtual-head padding), PP1 (1)

1,048,576 (3) · 1,048,576 (KV pool 6.62M tokens at utilization 0.90, about 6.3 concurrent 1M requests) (1)

78.55 (1) · 123 (1) · 153.35 (1) · 160.8 (1)

None (2) · 3,463 (1) · 4,037 (1)

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

None · 180 W · 200 W · 100 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

None (4)

None (3) · cmpunlocker VRAM unlock (64 GB per card) (1)

Reported (4)

HalfVulpes (2) · promisezackr (1) · wgkbfcp (1)

August 2026 (1) · October 2026 (3)

https://github.com/HalfVulpes/Deepinfer-SM80 (2) · https://github.com/promisezackr/glm53-flash-170hx-pp8 (1) · https://github.com/wgkbfcp/deepseek-v4.1-flash-1017-cmp170hx-recipe (1)