Recipe
qwen3-8-27b-q8-0-radeon-ai-pro-r9700-32gb-llama-cpp-tp3
qwen3-8-27b-q8-0-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
- b10690-db2bf4143
- Accelerators
- 3
- Tensor parallel
- 3
- Context tokens
- 10,952
- chat
- unknown
- reasoning
- unknown
- tools
- unknown
- vision
- unknown
Hugging Face model card
Identity
https://huggingface.co/Qwen/Qwen3.8-27B- Repository
- Qwen/Qwen3.8-27B
- 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.
llama-server-mQwen3.8-27B-Q8_0.gguf--spec-typedraft-mtp-adaptive-smtensor-ts1/1-ngl99--flash-attnauto-ctkf16-ctvf16-c262144(GGML_CUDA_ALLREDUCE=internal+GGML_CUDA_P2P=1,pci=disable_acs_redirlive,2of3GPUsvisible)
| Flag | Value |
|---|---|
-m | Qwen3.8-27B-Q8_0.gguf |
--spec-type | draft-mtp-adaptive |
-sm | tensor |
-ts | 1/1 |
-ngl | 99 |
--flash-attn | auto |
-ctk | f16 |
-ctv | f16 |
-c | 262144 |
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 10,952 | 1,058.6 | 59.2 | 10,119.5 | observed | qwen3-8-27b-q8-0-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
- qwen3-8-27b-q8-0-radeon-ai-pro-r9700-32gb-llama-cpp-tp3
- model instance id
- qwen-qwen3-8-27b--q8-0
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- qwen3-8-27b-q8-0-radeon-ai-pro-r9700-32gb-llama-cpp-tp3-sweep
- status
- candidate
capabilities
engine
- name
- llama.cpp
- version
- b10690-db2bf4143
serving
- max context tokens
- 10,952
- 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
- 2026-08-31T23:03:15Z
serving.max concurrency · reason server-capacity-not-evidenced
serving.max concurrency · state unknown
metadata
- localmaxxing · backend
- rocm
- localmaxxing · batch size
- 1
- localmaxxing · hardware label
- Radeon AI Pro R9700
- localmaxxing · notes
- llama-benchy 0.4.0 depth-ladder (pp2520 tg240, no-cache, runs=2+warmup) at depth 8192. rdna-boosts rebased onto master 50f068fff (b10690, incl. PR #27742), ROCm 7.14 TheRock, gfx1201, draft-mtp-adaptive MTP, tensor split over 2 compute cards. New: P2P peer access (ACS redirect) + butterfly allreduce — RCCL allreduce costs 20% decode on small MTP payloads. +4.6% tg over prior best 56.61, +11.5% over pre-rebase 53.13.
- localmaxxing · observed command
- llama-server -m Qwen3.8-27B-Q8_0.gguf --spec-type draft-mtp-adaptive -sm tensor -ts 1/1 -ngl 99 --flash-attn auto -ctk f16 -ctv f16 -c 262144 (GGML_CUDA_ALLREDUCE=internal + GGML_CUDA_P2P=1, pci=disable_acs_redir live, 2 of 3 GPUs visible)
- localmaxxing · run id
- cmtdfljvx004dlm0125ofpfqd
- localmaxxing · tokenized · arguments
- llama-server, -m, Qwen3.8-27B-Q8_0.gguf, --spec-type, draft-mtp-adaptive, -sm, tensor, -ts, 1/1, -ngl, 99, --flash-attn, auto, -ctk, f16, -ctv, f16, -c, 262144, (GGML_CUDA_ALLREDUCE=internal, +, GGML_CUDA_P2P=1,, pci=disable_acs_redir, live,, 2, of, 3, GPUs, visible)
- 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/cmtdfljvx004dlm0125ofpfqd ↗ |