Skip to content

Beelink GTR9 Pro (Ryzen AI Max+ 395) AI benchmarks

Explore LLM and image benchmark results for Beelink GTR9 Pro (Ryzen AI Max+ 395), with hardware specifications and the software settings used in each test.

Hardware specifications

CPU
AMD Ryzen AI Max+ 395
GPU / SoC
AMD Radeon 8060S (integrated)
Memory
128 GB unified
64 GB assigned to GPU in BIOS
INT8 compute
~59 TOPS
Memory bandwidth
256 GB/s
Storage
2 TB SSD
OS
Ubuntu 24.04
Driver
amdgpu 6.16.6, ROCm 7.1.1, Mesa RADV 25.2.8

Benchmark results

ModelBenchmarkConfigurationGTR9 ProOlares One
Qwen3.8-27BIFBench
Evaluation time
llama.cpp Vulkan, UD-Q4_K_XL33.1 h8.7 h
Qwen3.8-27BInference speed
Token generation, single request
llama.cpp Vulkan, UD-Q4_K_XL28 tok/s93 tok/s
Qwen-Image-2.1 BaseText to image
Per image
ComfyUI, int81.6 min12 s
Qwen-Image-2.1 TurboImage edit
Per image
ComfyUI, int847 s5.2 s

Test configurations

Qwen3.8-27B

IFBench

llama.cpp Vulkan, UD-Q4_K_XL

Engine and build
llama.cpp b11718 (547208dca), official prebuilt llama-b11718-bin-ubuntu-vulkan-x64, Vulkan, Mesa RADV 25.2.8
Launch arguments
--mmproj mmproj-F16.gguf -c 104448 -ngl all -fa on -ctk q8_0 -ctv q8_0 --jinja -np 1 --spec-type draft-mtp --spec-draft-n-max 5

Inference speed

llama.cpp Vulkan, UD-Q4_K_XL

Engine and build
llama.cpp b11718 (547208dca), official prebuilt llama-b11718-bin-ubuntu-vulkan-x64, Vulkan, Mesa RADV 25.2.8
Launch arguments
--mmproj mmproj-F16.gguf -c 104448 -ngl all -fa on -ctk q8_0 -ctv q8_0 --jinja -np 4 -kvu --spec-type draft-mtp --spec-draft-n-max 5

Qwen-Image-2.1

ComfyUI, int8

Qwen-Image-2.1 Base: Text to image

Qwen-Image-2.1 Turbo: Image edit

Engine and build
ComfyUI GitHub archive of commit 63e5bf594339f676b4b15e072529873e3448409e, GitHub archive of the commit in comfyui, venv/comfyui (uv pip), torch 2.14.1+rocm7.2, triton-rocm 3.8.0, torchvision 0.29.1+rocm7.2, torchaudio 2.11.0+rocm7.2 wheels from download.pytorch.org installed with --no-deps, then requirements.txt, ROCm, torch 2.14.1+rocm7.2, Python 3.12.3
Launch arguments
--listen 127.0.0.1 --port 8288 --extra-model-paths-config extra_model_paths.yaml --output-directory images --input-directory inputs --disable-dynamic-vram, COMFYUI_ENABLE_MIOPEN=1

ComfyUI, NVFP4

Qwen-Image-2.1 Base: Text to image

Qwen-Image-2.1 Turbo: Image edit

Engine and build
ComfyUI GitHub archive of commit 63e5bf594339f676b4b15e072529873e3448409e, GitHub archive of the commit in comfyui, venv/comfyui (uv pip), torch 2.14.1+rocm7.2, triton-rocm 3.8.0, torchvision 0.29.1+rocm7.2, torchaudio 2.11.0+rocm7.2 wheels from download.pytorch.org installed with --no-deps, then requirements.txt, ROCm, torch 2.14.1+rocm7.2, Python 3.12.3
Launch arguments
--listen 127.0.0.1 --port 8288 --extra-model-paths-config extra_model_paths.yaml --output-directory images --input-directory inputs --disable-dynamic-vram, COMFYUI_ENABLE_MIOPEN=1