Recipe
qwen3-6-27b-q8-0-rocmfpx-radeon-ai-pro-r9700-32gb-llama-cpp-tp3
qwen3-6-27b-q8-0-rocmfpx-radeon-ai-pro-r9700-32gb-llama-cpp-tp3Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.
Record
- Status
- candidate
- Source
- localmaxxing
- Engine
- llama.cpp
- Engine version
- ROCmFPX r9700-gfx1201-lab@45bcff5
- Accelerators
- 3
- Tensor parallel
- 3
- Context tokens
- 2,048
- Max concurrency
- 4
- chat
- unknown
- reasoning
- unknown
- tools
- unknown
- vision
- unknown
Hugging Face model card
Identity
https://huggingface.co/Qwen/Qwen3.6-27B- Repository
- Qwen/Qwen3.6-27B
- 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
llama.cpp
Evidence only · candidate · reference
Candidate evidence — not a Run contract
Observed source tokens
Mechanical split of the source command. Unverified against the engine CLI. The registry does not offer Run for this recipe.
3x:ROCmFPXllama-server(r9700-gfx1201-lab@45bcff5)--deviceROCm$i-ngl99--jinja--prio3-mQwen3.6-27B-Q8_0_ROCMFPX.gguf--flash-attnon--no-mmap-ctkf16-ctvf16-b2048-ub512-np4-c8192
| Flag | Value |
|---|---|
--device | ROCm$i |
-ngl | 99 |
--prio | 3 |
-m | Qwen3.6-27B-Q8_0_ROCMFPX.gguf |
--flash-attn | on |
-ctk | f16 |
-ctv | f16 |
-b | 2048 |
-ub | 512 |
-np | 4 |
-c | 8192 |
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 787 | 248.9 | 140.9 | 1,008.3 | observed | qwen3-6-27b-q8-0-rocmfpx-radeon-ai-pro-r9700-32gb-llama-cpp-tp3-sweep |
Remaining fields
Identity, launch, related records, and measured speed are shown above. This is the rest of the normalized record.
- hardware count
- 3
- hardware id
- radeon-ai-pro-r9700-32gb
- id
- qwen3-6-27b-q8-0-rocmfpx-radeon-ai-pro-r9700-32gb-llama-cpp-tp3
- model instance id
- qwen-qwen3-6-27b--q8-0-rocmfpx
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- qwen3-6-27b-q8-0-rocmfpx-radeon-ai-pro-r9700-32gb-llama-cpp-tp3-sweep
- status
- candidate
capabilities
engine
- name
- llama.cpp
- version
- ROCmFPX r9700-gfx1201-lab@45bcff5
serving
- max concurrency
- 4
- max context tokens
- 2,048
- tensor parallel
- 3
Provenance & metadata (3)
facts
- capabilities.chat · provenance · captured at
- 2026-08-30T09:10:02Z
capabilities.chat · reason capability-not-verified
capabilities.chat · state unknown
- capabilities.reasoning · provenance · captured at
- 2026-08-30T09:10:02Z
capabilities.reasoning · reason capability-not-verified
capabilities.reasoning · state unknown
- capabilities.tools · provenance · captured at
- 2026-08-30T09:10:02Z
capabilities.tools · reason capability-not-verified
capabilities.tools · state unknown
- capabilities.vision · provenance · captured at
- 2026-08-30T09:10:02Z
capabilities.vision · reason capability-not-verified
capabilities.vision · state unknown
- engine.graph mode · provenance · captured at
- 2026-08-30T09:10:02Z
engine.graph mode · reason runtime-detail-not-published
engine.graph mode · state unknown
- serving.kv cache tokens · provenance · captured at
- 2026-08-30T09:10:02Z
serving.kv cache tokens · reason kv-cache-capacity-not-published
serving.kv cache tokens · state unknown
- serving.max concurrency · provenance · captured at
- 2026-09-01T10:20:39Z
serving.max concurrency · reason explicit-source-server-capacity
serving.max concurrency · state known
- serving.max context tokens · provenance · captured at
- 2026-09-01T10:20:39Z
serving.max context tokens · reason explicit-source-context-limit
serving.max context tokens · state known
metadata
- localmaxxing · backend
- vulkan
- localmaxxing · batch size
- 12
- localmaxxing · hardware label
- Radeon AI Pro R9700
- localmaxxing · notes
- Custom Q8_0_ROCMFPX quant (1337Hero/Qwen3.6-27B-Q8_0-ROCMFPX-GGUF) on the custom ROCmFPX HIP llama-server, native gfx1201. 3 independent single-card replicas (one per R9700, --device ROCm$i), client round-robin, 4 slots each. Aggregate tok/s, median of 3 rounds, cold prefill. Quant is 2.94% smaller than upstream Q8_0, which is what lets it fit one card with ~1GB headroom.
- localmaxxing · observed command
- 3x: ROCmFPX llama-server (r9700-gfx1201-lab @45bcff5) --device ROCm$i -ngl 99 --jinja --prio 3 -m Qwen3.6-27B-Q8_0_ROCMFPX.gguf --flash-attn on --no-mmap -ctk f16 -ctv f16 -b 2048 -ub 512 -np 4 -c 8192
- localmaxxing · run id
- cmro9s3kb0285lg01odsdmnwq
- localmaxxing · tokenized · arguments
- 3x:, ROCmFPX, llama-server, (r9700-gfx1201-lab, @45bcff5), --device, ROCm$i, -ngl, 99, --jinja, --prio, 3, -m, Qwen3.6-27B-Q8_0_ROCMFPX.gguf, --flash-attn, on, --no-mmap, -ctk, f16, -ctv, f16, -b, 2048, -ub, 512, -np, 4, -c, 8192
- localmaxxing · tokenized · fidelity
- faithful
provenance
- captured at
- 2026-08-30T09:10:02Z
sources
| captured at | kind | url |
|---|---|---|
| 2026-08-30T09:10:02Z | normalized-recipe | www.localmaxxing.com/en/runs/cmro9s3kb0285lg01odsdmnwq ↗ |