Recipe
qwen3-6-27b-fp8-rtx-3090-24gb-vllm-tp2
qwen3-6-27b-fp8-rtx-3090-24gb-vllm-tp2Record
- Status
- candidate
- Source
- localmaxxing
- Engine
- vllm
- Engine version
- 0.20.2
- Accelerators
- 2
- Tensor parallel
- 2
- chat
- unknown
- reasoning
- unknown
- tools
- unknown
- vision
- unknown
Hugging Face model card
Identity
https://huggingface.co/Qwen/Qwen3.6-27B-FP8- Repository
- Qwen/Qwen3.6-27B-FP8
- Status
- known
- Link type
- Exact Hub repository
Public Hugging Face repository confirmed by the Hub API.
Candidate: useful compatibility or speed evidence. The registry does not offer Run until promotion requirements are met.
Observed configuration
vllm
Evidence only · candidate · reference
Candidate evidence — not a Run contract
No tokenized launch fields. This is measured or documented compatibility, not a Docker launch.
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 8,192 | — | 77.8 | 355.4 | historical | qwen3-6-27b-fp8-rtx-3090-24gb-vllm-tp2-sweep |
Remaining fields
Identity, launch, related records, and measured speed are shown above. This is the rest of the normalized record.
- hardware count
- 2
- hardware id
- rtx-3090-24gb
- id
- qwen3-6-27b-fp8-rtx-3090-24gb-vllm-tp2
- model instance id
- qwen-qwen3-6-27b-fp8--fp8--33492
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- qwen3-6-27b-fp8-rtx-3090-24gb-vllm-tp2-sweep
- status
- candidate
capabilities
draft launch
- accelerator backend
- nvidia
- arguments
- --model, Qwen/Qwen3.6-27B-FP8, --tensor-parallel-size, 2, --host, 0.0.0.0, --port, 8000
- container port
- 8,000
- host port
- 8,000
- image
- vllm/vllm-openai@sha256:0a51ea5b4ae2dc5d81890e5173f54203d2a3ae0cfffe51b8fd2afd4391bfd967
- ipc
- host
- kind
- docker
- shm size
- 16g
- synthesized · generated at
- 2026-08-31T22:12:17Z
- synthesized · image provenance
- deepseek-fp8-rtx-pro-6000-blackwell-96gb-vllm-tp1
- synthesized · template
- vllm-openai-v1
mounts
| read only | target |
|---|---|
| No | /root/.cache/huggingface |
engine
- name
- vllm
- version
- 0.20.2
serving
- tensor parallel
- 2
Provenance & metadata (3)
facts
- capabilities.chat · provenance · captured at
- 2026-08-27T06:04:13.773Z
capabilities.chat · reason capability-not-verified
capabilities.chat · state unknown
- capabilities.reasoning · provenance · captured at
- 2026-08-27T06:04:13.773Z
capabilities.reasoning · reason capability-not-verified
capabilities.reasoning · state unknown
- capabilities.tools · provenance · captured at
- 2026-08-27T06:04:13.773Z
capabilities.tools · reason capability-not-verified
capabilities.tools · state unknown
- capabilities.vision · provenance · captured at
- 2026-08-27T06:04:13.773Z
capabilities.vision · reason capability-not-verified
capabilities.vision · state unknown
- description · provenance · captured at
- 2026-08-27T06:04:13.773Z
description · reason not-observed
description · state unknown
- engine.graph mode · provenance · captured at
- 2026-08-27T06:04:13.773Z
engine.graph mode · reason runtime-detail-not-published
engine.graph mode · state unknown
- metadata.localmaxxing.engine flags.attentionBackend · provenance · captured at
- 2026-08-27T06:04:13.773Z
metadata.localmaxxing.engine flags.attentionBackend · reason not-observed
metadata.localmaxxing.engine flags.attentionBackend · state unknown
- metadata.localmaxxing.engine flags.gpuLayers · provenance · captured at
- 2026-08-27T06:04:13.773Z
metadata.localmaxxing.engine flags.gpuLayers · reason not-observed
metadata.localmaxxing.engine flags.gpuLayers · state unknown
- serving.kv cache tokens · provenance · captured at
- 2026-08-27T06:04:13.773Z
serving.kv cache tokens · reason kv-cache-capacity-not-published
serving.kv cache tokens · state unknown
- serving.max concurrency · provenance · captured at
- 2026-08-27T06:04:13.773Z
serving.max concurrency · reason not-observed
serving.max concurrency · state unknown
- serving.max context tokens · provenance · captured at
- 2026-08-27T06:04:13.773Z
serving.max context tokens · reason not-observed
serving.max context tokens · state unknown
metadata
- localmaxxing · base lineage
- known
- localmaxxing · engine flags · flashAttn
- Yes
- localmaxxing · engine flags · kvCacheDtype
- auto
- localmaxxing · engine flags · mtpEnabled
- No
- localmaxxing · engine flags · specDecoding
- Yes
- localmaxxing · engine flags · tensorParallel
- 2
- localmaxxing · hardware label
- 2x RTX 3090 24GB
- localmaxxing · notes
- Lenovo ThinkStation P620 (TR Pro 5955WX, 64 GB DDR4-3200 ECC), 2x RTX 3090 (1x Gigabyte Waterforce Xtreme 3090 Ti + 1x NVIDIA Founders Edition 3090), both PCIe Gen4 x16, both 250 W power cap, ReBAR enabled (BAR1 32 GB each), no NVLink, NODE topology (GPU0 domain 40, GPU1 domain 60). Engine: vLLM 0.20.2 + torch 2.11.0 + CUDA 13.x runtime. Marlin FP8 dequant (Ampere has no native FP8 tensor cores). TP=2 across both cards via NCCL allreduce. MTP n=3 speculative decoding via Qwen-shipped mtp.safetensors weights, acceptance 56.12% (per-position 70.19/54.82/43.35), accept length 2.68. Methodology: vllm bench serve, random 512in/512out, 20 prompts, max-concurrency 1, --save-result. Hardware-delta context: bench-paired no-MTP run on same rig was 44.82 t/s, vs prior PC dual-3090 vLLM FP8 entry at 44.17 t/s (+1.5%, within noise) - confirming PCIe x16 vs x4 contributes ~zero to single-stream throughput. The +75% MTP uplift is all software, not hardware.
- localmaxxing · revision unpinned
- Yes
- localmaxxing · run id
- cmp4vlcfg000kph01ggawf644
provenance
- captured at
- 2026-08-27T06:04:13.773Z
sources
| captured at | kind | url |
|---|---|---|
| 2026-08-27T06:04:13.773Z | normalized-recipe | www.localmaxxing.com/en/runs/cmp4vlcfg000kph01ggawf644 ↗ |