Framework’s first computer has been improved for games, and local artificial intelligence inference

rangeThe company known as repairable standard laptops, just released the first desktop computer. It is a small desktop computer that punches its weight.

The most interesting part is what is inside the device. Framework is one of the first companies to use AMD from Strix Halo Architecture, also known as Ryzen Ai Max processors. It is the one -in -one processing unit that promises some serious performance.

In other words, Framework has just designed a personal computer for two types of customers: people looking for a very small game computer, or people who want to run large language models on their computers.

From the outside, the frame surface looks like a game more than a dangerous computer. It is a small 4.5 -liter computer built on a small main plate, making it smaller than PlayStation 5 or Xbox Series X.

It has a customizable front panel with 21 switching plastic square tiles. When purchasing a frame desktop on the company’s website, you can select the colors and patterns of tiles to create your front panel.

In addition to the usual ports that you usually get with a mini-board in ITX, you will find the iconic expansion cards of Framework at the bottom of the device-Antin in the front, and two at the back. You can choose between a wide range of stereotypes, such as USB-C or USB-A, headphone socket, SD card reader, or even storage expansion card.

The interior is very simple: there is the main plate with the accelerated treatment unit from AMD, fan, heat sink, power source, and two M.2 2280 NVME SSD storage.

Amd’s Strix Halo APU is welding on a main panel. Framework provides two different compositions CPU vehicles, 32 graphics cores and 40MB of Temporary storage memory.

But where is the ram? This will definitely be the most divided option because Framework provides 32 GB to 128 GB of welding RAM. You will not be able to buy more random access memory or promote it on the road.

“There is one place that we had to stay away from personal computer standards, which is on memory. books On the company’s blog.

He added: “We spent months working with AMD to explore ways to overcome this, but we finally decided that it is not technically possible to land in the standard memory when highly productive with a 256 -bit memory carrier.”

However, the presence of up to 128 GB of unified memory opens many possibilities when it comes to large language models. Llama 3.3 70B can work without any breaks using OLLAMA, Llama.cPP and other open source tools for local artificial intelligence work.

Other open -weight models of Mistral, Nous, Hermes or Deepseek should also work well. Framework also sells the main plate without condition. For example, the company built a mini -building with four panels that operate on the frame of the frame of the frame in parallel with the artificial intelligence test.

Framework starts from $ 1099, while the higher version costs $ 1,999. Like other frameworks, the company is supported by Windows in addition to famous Linux distributions such as Ubuntu, Fedora or her cousin who focuses on games.

Pre -orders are now open, but charges will start only in the early Q3 2025.