Beelink GTR9 Pro

BeelinkMini PC reference

Beelink GTR9 Pro

Beelink GTR9 Pro combines AMD Ryzen AI Max+ 395, Radeon 8060S and 128GB of onboard memory in a compact workstation. The large shared-memory capacity is relevant to memory-heavy creative work and local AI runtimes, while dual 10GbE can connect fast storage or other systems. Memory is fixed, and usable model capacity depends on the runtime, quantization and memory allocation.

Processor
AMD Ryzen AI Max+ 395
Graphics
Radeon 8060S
Memory
128 GB
Storage
2 TB
Configuration being compared128 GB RAM · 2 TB SSDNew
US stores · USD

Prices across stores

MiniPC.guru compares retailer listings. Purchases are completed on the retailer’s website.

View price history
RetailerAvailabilityListed priceLast checkedRetailer website
In stock$4,349View at Beelink
Not confirmedSee at storeCheck expiredView at Amazon

Availability reflects the latest store check. Shipping, tax, coupons and checkout totals are not included; confirm them at the store.

Specifications

Processor, graphics and installed memory/storage describe the selected configuration. Expansion limits and connectivity describe the model; optional hardware is identified separately. References are collected in About this data below.

Processor
AMD Ryzen AI Max+ 395
Graphics
Radeon 8060S
Memory
128 GB
Storage
2 TB
Ports & displays
2 × USB-A 3.2 (10Gbps); front USB-C (10Gbps data); 2 × rear USB4 (40Gbps, DisplayPort 1.4; USB PD explicitly unsupported); HDMI 2.1 and DisplayPort 2.1, each up to 8K/60Hz; SD 4.0 reader; 2 × 3.5mm audio; dual 10GbE RJ45; AC input. Up to four display outputs.
Ethernet, Wi-Fi & Bluetooth
2 × 10GbE. Wi-Fi 7 and Bluetooth 5.4 (MediaTek MT7925 wireless card).
RAM & storage upgrades
128GB onboard LPDDR5X-8000, not user-upgradeable. 2 × M.2 2280 PCIe 4.0 x4 SSD slots. Maximum supported SSD capacity is unconfirmed because the manufacturer page and its specification illustration publish different per-slot limits.
Power supply & power limits
Integrated 230W power supply; input 100–240V AC, 50/60Hz. This is a supply rating, not measured system consumption.
Dimensions & weight
180 × 180 × 90.8mm, excluding rubber feet; 2338.8g. Manufacturer dimension order; axes are not labeled.

Video

Plays on YouTube. Nothing is loaded from YouTube until you press play. YouTube

What owners say

Its two 10-gigabit ports are the reason to buy this machine and the reason owners send it back. Under mixed graphics and network load both Intel controllers hang at once, disappear from the bus, and return only if you unplug the machine from the wall; a reboot is not enough. One owner traced why: the two ports share a single firmware image and one management processor, so they fail together. Newer firmware stops the blue screens but not the hang. On Beelink's own forum, in a thread open since September 2025 with 98 participants, support first advised using Wi-Fi instead, and in August 2026 still answered that its engineers were working on it. Boards shipped from February 2026 replace the Intel controller with a Realtek one; several owners report the replacement is stable and at least one reports the identical failure on Realtek hardware, which points at the board rather than the chip. The rest is good. Idle draw is 16 to 25 W, a 120-billion-parameter model runs sustained at 125 to 128 W, and a review lab measured 39 to 41 dB(A) while it did, although one owner measured 52 and returned the machine. Memory and processor are soldered, and 96 of the 128 GB arrive assigned to graphics.

For whom
Local AI on a large unified memory pool, for someone who can work over Wi-Fi or USB networking, or who has confirmed which Ethernet controller their board carries.
Not for
Anything that depends on the wired network while the machine is under load, which is the feature it is sold on.

Our own summary of 5 published sources, reviewed Sep 17, 2026. We have not tested this Mini PC ourselves; the sources are listed under About this data.

Performance

5 sourced resultstok/s

Same model, quantization, context length, prompt, output length, runtime version and batch size. Record time to first token separately.

Externally reported measurements. MiniPC.guru has not independently reproduced these tests.

55.57 tok/sCommunity benchmark

gpt-oss-120b-tg128-r20.csv, max-performance campaign

By hogeheer499-commits · GitHub

Source published
May 7, 2026
Test performed
May 7, 2026
Added to MiniPC.guru
Sep 17, 2026
Source reviewed
Sep 17, 2026
Tested hardware
AMD Ryzen AI Max+ 395 with Radeon 8060S, 16 cores / 32 threads · Radeon 8060S Graphics, gfx1151, RADV STRIX_HALO · 128 GB LPDDR5X-8000 unified memory, about 124 GiB visible to the operating system, 512 MB assigned to the iGPU in firmware · not applicable: token generation does not read from disk. The source documents the memory but not the SSD capacity; Beelink sells the GTR9 Pro in a single 2 TB configuration
Software / version / OS
llama-bench, Vulkan backend · llama.cpp build 9049, commit 2496f9c14 · Ubuntu 24.04, kernel 6.19.4-061904-generic, Mesa 26.0.6
Power profile
tuned accelerator-performance; IOMMU disabled; firmware UMA set to 512 MB
Test conditions
gpt-oss 120B MXFP4 MoE, 116.8 billion parameters, 63.4 GB on disk. Quantisation MXFP4. n_gen 128, n_prompt 0, n_depth 0, so pure generation of 128 tokens at zero context depth; n_batch 2048, n_ubatch 512, n_threads 16, type_k and type_v f16, n_gpu_layers 999, flash_attn on, use_mmap off, AMD_VULKAN_ICD=RADV. Twenty repetitions, standard deviation 1.41 tok/s

Editorial review checks the source and documentation; it does not certify the measurement or mean we tested the device ourselves.

62.56 tok/sCommunity benchmark

qwen36-35b-b9049-clean-r20.csv, clean b9049 rerun

By hogeheer499-commits · GitHub

Source published
May 7, 2026
Test performed
May 7, 2026
Added to MiniPC.guru
Sep 17, 2026
Source reviewed
Sep 17, 2026
Tested hardware
AMD Ryzen AI Max+ 395 with Radeon 8060S, 16 cores / 32 threads · Radeon 8060S Graphics, gfx1151, RADV STRIX_HALO · 128 GB LPDDR5X-8000 unified memory, about 124 GiB visible to the operating system, 512 MB assigned to the iGPU in firmware · not applicable: token generation does not read from disk. The source documents the memory but not the SSD capacity; Beelink sells the GTR9 Pro in a single 2 TB configuration
Software / version / OS
llama-bench, Vulkan backend · llama.cpp build 9049, commit 2496f9c14 · Ubuntu 24.04, kernel 6.19.4-061904-generic, Mesa 26.0.6
Power profile
tuned accelerator-performance; IOMMU disabled; firmware UMA set to 512 MB
Test conditions
Qwen3.6 35B-A3B mixture of experts, 34.7 billion parameters, 22.1 GB on disk. Quantisation UD-Q4_K_M. n_gen 128, n_prompt 0, n_depth 0, so pure generation of 128 tokens at zero context depth; n_batch 2048, n_ubatch 512, n_threads 16, type_k and type_v f16, n_gpu_layers 999, flash_attn on, use_mmap off, AMD_VULKAN_ICD=RADV. Twenty repetitions, standard deviation 1.77 tok/s

Editorial review checks the source and documentation; it does not certify the measurement or mean we tested the device ourselves.

100.99 tok/sCommunity benchmark

qwen3-coder-q4ks-b9851-p512-n128-r50.csv, b9851 Vulkan sentinel

By hogeheer499-commits · GitHub

Source published
Jun 30, 2026
Test performed
Jun 30, 2026
Added to MiniPC.guru
Sep 17, 2026
Source reviewed
Sep 17, 2026
Tested hardware
AMD Ryzen AI Max+ 395 with Radeon 8060S, 16 cores / 32 threads · Radeon 8060S Graphics, gfx1151, RADV STRIX_HALO · 128 GB LPDDR5X-8000 unified memory, about 124 GiB visible to the operating system · not applicable: token generation does not read from disk. The source documents the memory but not the SSD capacity; Beelink sells the GTR9 Pro in a single 2 TB configuration
Software / version / OS
llama-bench from the official b9851 release binary, Vulkan backend · llama.cpp build 9851, commit 0eca4d490 · Ubuntu 24.04; the source assigns no Mesa version to this run and warns against inheriting an older system snapshot
Power profile
not reported by source: the repository does not record the power profile for the 30 June 2026 run
Test conditions
Qwen3-Coder 30B-A3B Instruct mixture of experts, 30.5 billion parameters, 17.5 GB on disk. Quantisation Q4_K_S. n_gen 128, n_prompt 0, n_depth 0, so pure generation of 128 tokens at zero context depth; n_batch 2048, n_ubatch 512, n_threads 16, type_k and type_v f16, n_gpu_layers 999, flash_attn on, explicit device Vulkan0, use_mmap off. Fifty repetitions, standard deviation 1.55 tok/s

Editorial review checks the source and documentation; it does not certify the measurement or mean we tested the device ourselves.

94.64 tok/sCommunity benchmark

qwen3-coder-30b-udq4kxl-b10687-p512-n128-r20.csv, b10687 Vulkan sentinel

By hogeheer499-commits · GitHub

Source published
Aug 30, 2026
Test performed
Aug 30, 2026
Added to MiniPC.guru
Sep 17, 2026
Source reviewed
Sep 17, 2026
Tested hardware
AMD Ryzen AI Max+ 395 with Radeon 8060S, 16 cores / 32 threads · Radeon 8060S Graphics, gfx1151, RADV STRIX_HALO · 128 GB LPDDR5X-8000 unified memory, about 124 GiB visible to the operating system · not applicable: token generation does not read from disk. The source documents the memory but not the SSD capacity; Beelink sells the GTR9 Pro in a single 2 TB configuration
Software / version / OS
llama-bench, Vulkan backend · llama.cpp build 10687, commit c841aee, compiled locally from source · Ubuntu 24.04, kernel 7.0.0-30-generic, Mesa 26.1.7
Power profile
desktop performance governor with DPM set to auto; the source records CPU-only background load during this run
Test conditions
Qwen3-Coder 30B-A3B Instruct mixture of experts. Quantisation UD-Q4_K_XL. n_gen 128, n_prompt 0, n_depth 0, so pure generation of 128 tokens at zero context depth; n_batch 2048, n_ubatch 512, n_threads 16, type_k and type_v f16, n_gpu_layers 999, flash_attn on. The use_mmap column is empty in this run's CSV. Twenty repetitions, standard deviation 0.40 tok/s

Editorial review checks the source and documentation; it does not certify the measurement or mean we tested the device ourselves.

18.43 tok/sCommunity benchmark

nemotron-3-super-120b-a12b-udiq4xs-b2016bf2-r3.csv, int-dot regression run

By hogeheer499-commits · GitHub

Source published
Jun 5, 2026
Test performed
Jun 5, 2026
Added to MiniPC.guru
Sep 17, 2026
Source reviewed
Sep 17, 2026
Tested hardware
AMD Ryzen AI Max+ 395 with Radeon 8060S, 16 cores / 32 threads · Radeon 8060S Graphics, gfx1151, RADV STRIX_HALO · 128 GB LPDDR5X-8000 unified memory, about 124 GiB visible to the operating system · not applicable: token generation does not read from disk. The source documents the memory but not the SSD capacity; Beelink sells the GTR9 Pro in a single 2 TB configuration
Software / version / OS
llama-bench, Vulkan backend · llama.cpp commit 2016bf2, compiled locally from source · Ubuntu 24.04; the source assigns no kernel or Mesa version to the 5 June 2026 run
Power profile
not reported by source: the repository does not record the power profile for the 5 June 2026 run
Test conditions
NVIDIA Nemotron 3 Super 120B-A12B mixture of experts, 120.7 billion parameters, 64.5 GB on disk. Quantisation UD-IQ4_XS at 4.25 bits per weight. n_gen 128, n_prompt 0, n_depth 0, so pure generation of 128 tokens at zero context depth; n_batch 2048, n_ubatch 512, n_threads 16, type_k and type_v f16, use_mmap on, n_gpu_layers and flash_attn left at the automatic defaults the tool prints as -1. Three repetitions, standard deviation 0.03 tok/s

Editorial review checks the source and documentation; it does not certify the measurement or mean we tested the device ourselves.

Price history

Beelink GTR9 Pro · 128 GB RAM · 2 TB SSD

Loading checks…

New condition · US stores · USD · Selected configuration only. Prices exclude shipping and tax. Each price stays on the chart until that listing's next check, for at most 36 hours, the same window the store prices use. Out-of-stock, failed and missing checks leave gaps.

About this data13 references

Specs and any benchmark results come from the sources below. We haven't tested this Mini PC ourselves.

Notes

  • upgrades: The page advertises 16TB total (8TB per slot), while the linked specification illustration states 4TB per slot. Neither ceiling is treated as validated compatibility here.
  • power: The short table lists 2.5A AC input while the illustration lists 3.5A. Input amperage is omitted; the separately stated 230W PSU rating is not measured PC consumption.

Sources

  1. Beelink GTR9 Pro — official specifications ↗
    • Manufacturer / specification source
    • Ports & displays · Reviewed Sep 15, 2026
    • Ethernet, Wi-Fi & Bluetooth · Reviewed Sep 15, 2026
    • RAM & storage upgrades · Reviewed Sep 15, 2026
    • Power supply & power limits · Reviewed Sep 15, 2026
    • Dimensions & weight · Reviewed Sep 15, 2026
  2. Selected configuration ↗
    • Selected configuration
  3. Beelink GTR9 Pro — manufacturer specification/connector illustration ↗
    • Ports & displays · Reviewed Sep 15, 2026
    • RAM & storage upgrades · Reviewed Sep 15, 2026
  4. gpt-oss-120b-tg128-r20.csv, max-performance campaign ↗
    • External benchmark · Reviewed Sep 17, 2026
  5. qwen36-35b-b9049-clean-r20.csv, clean b9049 rerun ↗
    • External benchmark · Reviewed Sep 17, 2026
  6. qwen3-coder-q4ks-b9851-p512-n128-r50.csv, b9851 Vulkan sentinel ↗
    • External benchmark · Reviewed Sep 17, 2026
  7. qwen3-coder-30b-udq4kxl-b10687-p512-n128-r20.csv, b10687 Vulkan sentinel ↗
    • External benchmark · Reviewed Sep 17, 2026
  8. nemotron-3-super-120b-a12b-udiq4xs-b2016bf2-r3.csv, int-dot regression run ↗
    • External benchmark · Reviewed Sep 17, 2026
  9. Beelink GTR9 Pro review: power, noise and verdict ↗
    • Owner review source · Reviewed Sep 17, 2026
  10. Beelink GTR9 Pro: tracing the dual 10GbE firmware hang ↗
    • Owner review source · Reviewed Sep 17, 2026
  11. GTR 9 Pro Ethernet malfunction under load ↗
    • Owner review source · Reviewed Sep 17, 2026
  12. Beelink GTR9 Pro: power modes, networking and first-week notes ↗
    • Owner review source · Reviewed Sep 17, 2026
  13. strix-halo-guide: evidence-backed AMD Strix Halo local-AI setup and benchmarks ↗
    • Owner review source · Reviewed Sep 17, 2026