Local AIAI workstation review
Framework Desktop review: the home AI machine you can actually open
128 GB of shared memory in a small box, running large AI models that a normal graphics card can't fit. A premium-priced system with accessible parts, though its memory cannot be upgraded later.
Is the Framework Desktop (Ryzen AI Max+ 395) worth buying?
Only if you specifically want to run large AI models at home. The Framework Desktop's 128 GB of shared memory holds models a normal graphics card can't, and several parts are serviceable. Check the current price for the 128 GB configuration. If mid-size models are all you need, compare a 32 GB Mac mini or Beelink SER8 before paying for memory you will not use.
The pick if you want to run the large models at home and you want a machine you can service yourself. Check the full configuration price; the memory is soldered. No measured speed or noise result is available here.
Where it's good
- 128 GB of shared memory runs models most machines can't load at all
- Standard mini-ITX case — swap the storage, fans, even the mainboard later
- Capacity is the reason to compare this configuration
- Framework's track record on parts and repairability is real
Where it bites
- A premium-priced system; check the total for the 128 GB configuration
- The 128 GB memory is soldered — a limit of the AMD chip, but still a limit
- Performance depends on the model, backend and workload
- Overkill if you only ever run mid-size models
Some links on this site may be affiliate links. As an Amazon Associate I earn from qualifying purchases. How this works.
Evidence: editorial research, not a documented hands-on test. The retained score is an editorial opinion; speed, noise and durability have not been substantiated by recorded tests here.
The Framework Desktop is a small computer built around AMD’s Ryzen AI Max+ 395 chip, which comes paired with 128 GB of memory shared between the processor and a large built-in graphics chip. That shared pool is the whole point: it lets you load AI models that exceed the memory capacity of a 24 GB graphics card.
Buying-page check — 12 October 2026 (US): the Framework Desktop DIY configurator showed a selectable Ryzen AI Max+ 395 / 128 GB option with a $3,449 estimate, before storage, operating system, CPU fan, power cable and customization. The 64 GB option was out of stock. The 128 GB estimate is not a complete-system or delivered quote; finish the configuration and check availability before comparing totals. This is a browser observation, not an API-verified offer or a measured performance result.
Availability follow-up — 17 October 2026 (US): the same Framework configurator now showed the Max+ 395 / 64 GB option selectable at $1,959, alongside the Max+ 395 / 128 GB option at $3,449. The Max 385 / 32 GB option was disabled and marked out of stock. These are system-option prices before separately selected storage and other extras, not complete configured totals. The October 12 observation above remains historical; availability can change. This check did not test checkout or hardware performance.
What the extra memory buys you
Most people running local AI are working with 24–32 GB and mid-size models. This machine is for the tier above that — the large models that start to feel like the paid services. Usable capacity depends on model and runtime overhead. We have no recorded response-time test to support a promise of answers in a few seconds.
It’s not the fastest way to run a model that does fit in 24 GB — a dedicated graphics card can win that race. The reason to choose this memory tier is capacity: room for models that exceed a smaller card’s memory.
The Framework part
The Framework uses a mini-ITX mainboard and has serviceable parts. The storage is user-replaceable, the fans are standard, and Framework sells the mainboard on its own if you want to reuse the case later or move the guts into something else.
The memory is the exception — it’s soldered, because the chip requires it. So the 128 GB decision is permanent. 128 GB is the top memory option for the Max+ 395 configuration covered here.
Use the model-memory planning guide to budget the shared memory pool for the OS, model and runtime overhead. The advertised capacity is not dedicated VRAM and does not guarantee model fit or response speed.
Who should buy it — and who shouldn’t
Buy it if you specifically want to run large models at home, you’d rather buy once than upgrade, and repairability matters to you.
Buy something else if mid-size models are all you need — a 32 GB Mac mini is worth comparing for cost and power use — or if speed on smaller models matters more than size, in which case a used RTX 3090 in a PC you already own is the option to price as a complete usable system.
The specs, in plain English
- CPU
- AMD Ryzen AI Max+ 395 (16C/32T)
- Graphics
- Radeon 8060S (integrated, 40 CU)
- RAM
- 128 GB LPDDR5X (soldered)Same 128 GB shared-memory trick as the GMKtec — enough to run large models at home. Also soldered; that's a limitation of the chip, not Framework.
- Storage
- User-replaceable NVMe (2x M.2 2280)
- Networking
- 5GbE, Wi-Fi 7
- Case
- 4.5 L, standard mini-ITX, user-serviceableUnlike most mini PCs, you can open it, swap the storage and fans, and it uses standard parts. You're not buying a sealed box.
The bottom line
If running the big models at home is the goal and you want to buy once, this is the machine. If you mostly run mid-size models, save the money and get a 32 GB Mac mini or a Beelink SER8 — you won't use what this offers.