Framework Unveils Gaming-Optimized Desktop PC with AI Capabilities

Framework, the folks famous for their modular, fixable laptops, have just dropped their first desktop PC. And let me tell you, this little machine packs a punch despite its size.

What really catches the eye is what's under the hood. Framework is among the first to jump on AMD's newly announced Strix Halo architecture, or as it's also called, the Ryzen AI Max processors. This all-in-one processing unit is set to deliver some serious power.

In short, Framework has crafted this PC with two groups in mind: those hunting for a tiny gaming rig, and those eager to run hefty language models right on their own machines.

From the outside, you might mistake the Framework Desktop for a toy rather than a serious piece of tech. It's a compact 4.5L unit, built around a mini-ITX mainboard, making it smaller than a PlayStation 5 or an Xbox Series X.

The front panel is where you can get creative, with 21 swappable plastic square tiles. When you order from Framework's site, you can pick out tile colors and patterns to make your front panel uniquely yours.

Alongside the typical ports you'd expect on a mini-ITX mainboard, you'll spot Framework's signature expansion cards at the bottom—two in the front, two in the back. You've got a bunch of options to choose from, like USB-C or USB-A ports, a headphone jack, an SD card reader, or even a storage expansion card.

The AMD Strix Halo APU is soldered onto the mainboard. Framework offers two setups—the AMD Ryzen AI Max 385 and the AMD Ryzen AI Max+ 395. The top model boasts 16 CPU cores, 40 graphics cores, and 80MB of cache, while the entry-level version has 8 CPU cores, 32 graphics cores, and 40MB of cache.

Now, about the RAM—here's where things get a bit controversial. Framework's gone with 32GB to 128GB of soldered-in RAM, meaning you can't add more or upgrade later.

"There is one place we did have to step away from PC norms, though, which is on memory. To enable the massive 256GB/s memory bandwidth that Ryzen AI Max delivers, the LPDDR5x is soldered," Framework CEO Nirav Patel explained on the company's blog.

"We spent months working with AMD to explore ways around this, but ultimately determined that it wasn’t technically feasible to land modular memory at high throughput with the 256-bit memory bus," he added.

Still, having up to 128GB of unified memory opens up a world of possibilities for running large language models. For instance, Llama 3.3 70B can run smoothly using tools like Ollama, llama.cpp, and other open-source options for local AI tasks.

Other open-weight models from Mistral, Nous, Hermes, or DeepSeek should also run without a hitch. Framework even offers the mainboard without a case. They've built a mini-rack with four Framework Desktop mainboards running in parallel for AI testing.

You can preorder now, but don't expect shipments until early Q3 2025.