▚ myminingrig

AI PC · mini-pc

Beelink GTR9 Pro (Ryzen AI Max+ 395)

The premium Strix Halo build - dual 10GbE and a vapour chamber - and, since a 2026 price reshuffle, the value pick too.

Memory
128 GB
Bandwidth
256 GB/s
AI compute
50 TOPS
£ / GB
£15

What it runs

With 128 GB of memory you can load, at 4-bit, a model up to roughly

~252B params

Verdict

The best-built Strix Halo box - dual 10GbE, quiet, premium - and after a 2026 price reshuffle it now undercuts the GMKtec it once cost double.

BEST FOR

  • + a premium build with fast networking
  • + AI server clustering
  • + the best-value 128GB Strix Halo right now

NOT FOR

  • - anyone who needs CUDA
  • - buyers who don't need the dual 10GbE

The Beelink GTR9 Pro is the same Ryzen AI Max+ 395 and 128GB you’ll find in cheaper boxes, dressed in a much nicer suit. Dual 10-gigabit networking, a full vapour chamber, a built-in power supply and a metal chassis that could pass for a Mac. It’s the premium take on Strix Halo - and after a 2026 price reshuffle, it’s no longer the pricey one: it now ships around $1,999 while a memory shortage pushed the once-cheaper GMKtec above it.

The take

The build is the best of the Strix Halo bunch, and the pitch is clustering. Two 10GbE ports and dual USB4 mean you can lash several together into a little AI cluster, the cooling is a proper vapour chamber, and there’s a 230W supply built in so there’s no brick on the desk. All lovely, and now the price cooperates too. The GTR9 Pro launched at a steep Western premium, north of four thousand for the 128GB model, more than double what the GMKtec then asked. That has flipped: it now ships around $1,999, while the 2026 memory shortage pushed the GMKtec’s 128GB SKU up to $3,400-3,650. So the box that used to be the dear one is, for now, the cheaper of the two, and it’s the one with the better build and the dual 10GbE. Prices in this corner move fast, so check both on the day, but as things stand the Beelink is the one to beat.

What it’ll actually run

It runs exactly what any 128GB Strix Halo box runs, because it’s the same silicon. Hand most of the 128GB to the Radeon 8060S and a 70B at Q4 fits with context to spare, MoE models like gpt-oss-120b sit comfortably, and the everyday 8B to 32B range is quick. Beelink quotes a Qwen 32B at Q8 around 6 tokens a second in LM Studio, which is about right for a dense model at that precision. MoE models do far better.

The speed ceiling is bandwidth, same story as its siblings: near 256 GB/s on paper, lower in the real world, so dense large models answer at a measured pace. Worth knowing: owners found the 128GB memory can run in a throttled mode until you unlock it in the BIOS, so check your configuration. It’s Windows 11 out of the box, runs Linux well - owners report Ollama and the Lemonade server working nicely on Ubuntu - and it’s quiet, though reviewers note it isn’t silent when you push it to full power.

Who should buy it

Buy it if you want the nicest-built Strix Halo machine, and doubly so now the price has fallen into line. For clustering boxes or moving big models around a home lab the dual 10GbE earns its keep, and at around $1,999 you get the premium build and the networking for less than the plainer GMKtec currently costs. As things stand it’s the pick of the 128GB Strix Halo boxes.

The caveats are smaller than they were. Prices in this corner move fast, so confirm the GMKtec hasn’t dropped back below it on a sale before you commit. And if you need CUDA, no AMD machine is your answer - a GB10 box like the DGX Spark is the Nvidia route to big unified memory. Great hardware, and now at a fair price.

Settings people actually run

The configs owners land on, pulled from the community. A sensible starting point, not gospel - tune to your own kit.

Biggest model that fits

gpt-oss-120b MoE (~60GB); BIOS memory split set high

Same 128GB pool as any Strix Halo box; hand most of it to the GPU.

Dense large model

Qwen 32B Q8 / Llama 70B Q4

Beelink quotes Qwen 32B Q8 around 6 tokens/s; MoE models are much quicker.

Unlock full bandwidth

Check the memory mode in the BIOS

Owners found 128GB can sit in a throttled mode until unlocked; verify your configuration.

Clustering

Dual 10GbE + dual USB4 for AI server clusters

The reason to pick this over a cheaper box: fast networking for lashing several together.

What owners report

Real first-hand experience gathered from owners and the community.

  • “

    An owner on Ubuntu 24.04 ran a 96GB VRAM / 32GB system split, with idle around 15W and AI-load draw up to about 180W at the wall; Ollama and the Lemonade AMD server both worked well.

    Beelink store owner review

  • “

    The GTR9 Pro's 128GB memory was found to run in a throttled 1:4 mode reporting about 72 GB/s until unlocked, so BIOS configuration matters for bandwidth.

    Beelink community forum

  • “

    Reviewers rate the build and features highly but note it isn't silent at full 140-160W performance. At review time the same chip sold for far less as the GMKtec EVO-X2, though a 2026 memory shortage has since reversed that - the GMKtec's 128GB SKU now costs more than the GTR9 Pro.

    ServeTheHome (build/noise); price reversal via grounded sweep 2026-09

Fact-checked 22 Sept 20263 claims verified against primary sources.
4 claim(s) we couldn't fully verify
  • · memory_bandwidth_gbs 256 - Beelink doesn't publish a bandwidth figure; 256 GB/s is the Strix Halo theoretical, real-world is lower, and owners found a throttled mode reporting about 72 GB/s until unlocked in the BIOS.
  • · Price around USD 1,999 / GBP 1,899 (Sept 2026) - Grounded Sept 2026 sweep: now shipping (was preorder) at ~USD 1,999, down from the ~4,349 Western preorder premium. Meanwhile the memory shortage pushed the GMKtec above it, reversing the value order - body/verdict/FAQ reframed. Highly channel-dependent and volatile; GBP is an estimate.
  • · power_w 140 - 140W is Beelink's full-performance figure; owners measured up to about 180W at the wall under AI load. Built-in 230W supply.
  • · affiliate ASIN B0GQXDCKN1 - A live Amazon US listing for the GTR9 Pro 128GB; not cross-checked against every regional catalogue.

Hands-on reviews we drew on

We don't just copy the spec sheet. These are the teardowns and hands-on reviews behind this page - worth watching in their own right.

Common questions

Is the Beelink GTR9 Pro still pricier than the GMKtec EVO-X2?+

Not any more - this flipped in 2026. The GTR9 Pro launched at a steep Western premium, north of four thousand against the GMKtec's ~1,800, but it now ships around 1,999 dollars while a memory shortage pushed the GMKtec's 128GB SKU up to ~3,400-3,650. Same chip, same 128GB, so the Beelink now gives you the nicer build and dual 10GbE for less than the box that used to undercut it.

What can the GTR9 Pro run?+

Exactly what any 128GB Strix Halo box runs. Hand most of the 128GB to the Radeon 8060S and a 70B at Q4 fits with context to spare, MoE models like gpt-oss-120b sit comfortably, and the everyday 8B to 32B range is quick.

Is it good for AI clustering?+

That's its party trick. Dual 10GbE ports and dual USB4 let you lash several units into a small AI cluster, which is the main reason to pick this over a cheaper box with the same silicon.

Does it run silently?+

It's quiet, with a proper vapour chamber and dual fans, but reviewers note it isn't silent when you push it to full 140W. For always-on ticking over it's fine; under sustained load you'll hear it.