Overview
Most small-form-factor PCs ask you to accept a bargain: you get the size, and you give up the graphics muscle. The Framework Desktop is built to break that deal. It wraps AMD’s Ryzen AI Max+ 395, a 16-core Zen 5 chip with an integrated Radeon 8060S GPU, in a roughly 4.5-liter case, then lets that GPU draw on a huge pool of shared memory.
The memory is the real story. Configurations go up to 128GB of LPDDR5x, and the GPU can claim a large share of it as video memory. That puts a class of workloads, mainly large local AI models, within reach of a box the size of a shoebox. Typical desktop graphics cards top out far below that capacity unless you spend workstation money.
A note on method: this is an analytical review based on the platform’s published specifications and architecture, not a run through a custom benchmark suite. Where performance depends on the workload, I explain why rather than quote numbers I haven’t measured.
At announcement, pricing started around $1,099 for the 32GB Ryzen AI Max 385 configuration and climbed to roughly $1,999 for the 128GB Max+ 395 model. Check Framework’s store for current figures, since component pricing has moved around across the industry.

Design and Hardware
Framework’s reputation rests on repairable, modular laptops, and the desktop carries that philosophy into a different shape. The mainboard follows a Mini-ITX layout, and the case is assembled with minimal tools. Storage sits in standard M.2 slots, the power supply is a standard FlexATX unit, and the cooling hardware is off-the-shelf in spirit, so a dead fan is a five-minute swap rather than a warranty claim.
The front of the case takes Framework’s expansion cards, the same swappable port modules used in its laptops. You choose whether the front panel carries extra USB-A, USB-C, or other connectivity. Tiles on the front panel can be arranged in different patterns, a small touch that makes the case feel like a Framework product rather than a generic mini PC. The back carries the fixed essentials, including fast wired networking and modern USB4 ports.
The design has one major exception to the modular rule: the memory is soldered. That is a technical necessity of the platform, not a Framework preference. The Ryzen AI Max chips reach their bandwidth by pairing the processor with LPDDR5x placed right next to it, and socketed memory cannot match the signal quality at those speeds. It is a reasonable engineering trade, but it means the RAM you buy on day one is the RAM you keep.
Performance in practice
The CPU side is straightforward. Sixteen Zen 5 cores with 32 threads handle compiling, video encoding, virtual machines, and heavy multitasking with the confidence of a proper desktop processor. Sustained power limits matter more here than in a laptop, because the desktop chassis can hold the chip at higher wattage for longer without throttling from heat soak.
The GPU is where expectations need calibrating. The Radeon 8060S is the largest integrated graphics part AMD has shipped, and it is capable at 1080p and 1440p gaming at sensible settings. It does not replace a high-end discrete card for 4K gaming or demanding ray tracing. If your main goal is the fastest frame rates per dollar, a conventional tower with a dedicated GPU wins that contest.
The case for this machine is capacity, not peak speed. A discrete GPU with 16GB or 24GB of VRAM forces you to shrink or split large models. The Framework Desktop lets you load a quantized model in the tens of billions of parameters into memory the GPU can address directly. It will not generate tokens as fast as a top-tier graphics card, because memory bandwidth is the limiting factor, but it can run models that a typical consumer GPU cannot load at all. For developers experimenting with local inference, that distinction is the entire value proposition.

Pros
- Very large shared memory pool. Up to 128GB that the GPU can draw from opens up local AI work, large datasets, and heavy creative projects that cap out quickly on consumer graphics cards.
- Serious CPU performance in a tiny case. Sixteen Zen 5 cores in a 4.5-liter chassis is a combination few small desktops offer, and the cooling headroom helps it sustain load.
- Repairable and standard where it can be. M.2 storage, a FlexATX power supply, and a Mini-ITX-style board mean upgrades and replacements follow familiar PC conventions.
- Modular front ports. Expansion cards let you shape the connectivity to your desk instead of accepting whatever the manufacturer chose.
- Good Linux footing. AMD’s open graphics driver stack and Framework’s enthusiast audience make this a natural machine for Linux users and developers.
- Efficient for what it does. A single chip doing the job of a CPU plus a large-memory GPU draws far less power than a comparable tower, which also means less noise and heat in a small room.
Cons
- Soldered memory. You cannot upgrade it later. Buying 32GB now and adding more in two years is not an option, so the right configuration has to be chosen up front.
- Memory bandwidth limits AI speed. Capacity is generous, but throughput is modest next to a high-end dedicated GPU. Large models will load, but they will not feel fast.
- Integrated graphics ceiling. It is strong for integrated, but a mid-range discrete card still beats it in gaming, and the gap widens at higher resolutions.
- Price climbs quickly. The top configuration costs enough that it competes with well-equipped towers, and the entry model offers less of the memory advantage that makes the platform interesting.
- Limited GPU upgrade path. The mainboard includes a PCIe slot, but the case is tight and the power supply is small, so this is not a machine for dropping in a power-hungry graphics card.
- Competition exists. Several mini PCs use the same chip, and some come in more conventional prebuilt packages. Framework’s edge is repairability and design, which not every buyer values.
Who Should Buy It
Three kinds of buyers get the most from this machine. The first is the developer or researcher who wants to run large language models locally for privacy, cost, or experimentation. The second is the creative professional who needs plenty of memory for video, 3D, or simulation work in a quiet, compact box. The third is the Linux enthusiast who wants a small, repairable workstation and would rather support a company that publishes its design choices.
Gamers should think harder. If playing at 1440p with moderate settings is the goal, the Framework Desktop is capable and pleasantly small. If maximum performance is the goal, the money is better spent on a tower with a dedicated card.
Buyers who want a plug-and-forget appliance may also be better served elsewhere. Framework’s appeal grows with how much you care about assembling, servicing, and tuning your own hardware.
Verdict
The Framework Desktop is a focused machine, and that focus is its strength. It does not try to beat a gaming tower at gaming or a rack server at throughput. It offers an unusual mix: a fast 16-core processor, a capable integrated GPU, and a memory pool large enough to run models that ordinary consumer hardware cannot touch, all in a case small enough to sit beside a monitor.

Recommendation: buy it if local AI work, large-memory workloads, or a quiet compact Linux workstation are your priorities, and choose the 128GB configuration if AI is the reason you are here. The 32GB model makes less sense, because it gives up the platform’s defining advantage. Skip it if you want the best gaming performance per dollar or need the freedom to upgrade memory later.
The soldered RAM is the one decision you cannot undo. Whatever capacity you pick today is the ceiling for the life of the machine, and models keep growing.





