Framework touts 128GB desktop configs

Framework is trying to turn a niche hardware spec into a buying moment. The spec is memory — not just raw RAM, but how much of it local AI workloads can actually use. In a new social push, the company highlighted that its 128GB Framework Desktop configs are still in stock and framed them as a practical option for people who need lots of memory for on-device inference. That only lands because Framework’s tiny desktop was already built around AMD’s Ryzen AI Max chips, which were designed to cram unusually large shared memory into a compact machine. (frame.work) ### What is Framework selling here? Framework Desktop is a 4.5-liter mini PC built around AMD Ryzen AI Max processors. The current lineup includes a 32GB Max 385 model, a 64GB Max+ 395 model, and a top 128GB Max+ 395 model. Framework’s own storefront makes the 128GB version a headline feature, right next to the claim that the box can run large local models like Llama 70B with up to 96GB of graphics-addressable memory. (fra([frame.work) Why does 128GB matter so much? For local AI, memory is often the wall before compute is. A lot of consumer GPUs are fast enough for inference but top out at 8GB, 16GB, or 24GB of VRAM. Framework’s pitch is that Ryzen AI Max sidesteps some of that constraint by using a unified memory pool — up to 128GB total, with up to 96GB assignable to the GPU on Windows and even more possible on Linux. Basically, it gives small-deskt(frame.work)ut jumping straight to workstation-class add-in GPUs. (frame.work) ### What is special about AMD’s chip here? The interesting part is the memory subsystem. Ryzen AI Max combines desktop-class Zen 5 CPU cores, a large Radeon 8060S integrated GPU, and a 256-bit LPDDR5x-8000 memory bus. Framework says that delivers 256GB/s of memory bandwidth — a big reason the machine can act more like a compact AI workstation than a normal mini PC. Think of it as an unu(frame.work)ing the GPU to live inside a much smaller bucket. (frame.work) ### Is this really about Apple? Partly — but the contrast needs care. Apple’s current Mac Studio lineup still includes a 256GB M3 Ultra configuration in the online store, priced at $5,999 with 1TB storage, and Apple’s 2025 tech specs list the M3 Ultra as configurable to 256GB unified memory. So the cleanest read is not “Apple removed 256GB,” but “Framework sees an opening among buyers comparing high-memory desktops for AI and workstation use.” (support.apple.com) ### Where does Framework think it wins? Price and positioning. Framework launched the top-end 128GB desktop at a starting price of $1,999, which is dramatically below Mac Studio configurations that reach 256GB memory. It also leans on standard PC parts — Mini-ITX board shape, Flex ATX power supply, M.2 storage, customizable front I/O — to make the machine feel less sealed-off than Apple’s desktop. (commu([support.apple.com)he-framework-desktop/65008)) ### What’s the catch? The memory is soldered. Framework says the giant 256-bit bus is what forces LPDDR5x onto the board, so this is not a modular-RAM story even if the rest of the system is unusually repairable and standards-based. That tradeoff is central to the whole product: you get the bandwidth and capacity up front, but you need to choose the right memory tier at purchase. (frame.work) ### Why push this now? Because AI PC marketing is getting more concrete. Last year, “AI PC” mostly meant NPUs and vague demos. Now buyers are asking a much simpler question — can this machine run the model I want, locally, without compromises? Framework’s answer is to point at the 128GB SKU and say yes, that one can. (frame.work) ### Bottom line This is l(frame.work)ner of the market — compact desktops for people who care more about memory-heavy local AI than brand prestige. If that niche keeps growing, 128GB stops looking like an oddball spec and starts looking like the whole point.