NVIDIA and MSI Create the Compact DGX Spark Platform

MSI has unveiled an unusual device for AI specialists—the portable Expert Edge system. Built on the NVIDIA Grace Blackwell architecture, it combines supercomputer-level computational power with a form factor small enough to be tucked into a backpack.

The device is based on a 20-core ARM processor and a graphics accelerator from NVIDIA. A key feature of the hardware is its substantial 128 GB of unified memory, allowing users to load and work with large language and graphic models for which the VRAM of standard graphics cards is often insufficient.

For peripheral connectivity, it features four USB 3.2 ports, an HDMI output, and a 10 Gigabit Ethernet jack. Of particular interest is the inclusion of a high-speed ConnectX-7 interface with 200 Gb/s throughput, which allows two such systems to be directly linked for distributed computing without the need for additional switches.

The device runs on the specialized DGXOS, built on Ubuntu. The system comes pre-loaded with a suite of tools for a quick start in development—from standard environments like Visual Studio Code to pre-installed guides for setting up popular frameworks such as Comfy UI and Open Web UI. According to testers, deploying a working environment for AI experiments takes just a few minutes.

Interestingly, enthusiasts have tested the system's capabilities beyond just AI. Thanks to the full-fledged desktop OS, they managed to run video game console emulators, including PS3 and Xbox. While this isn't the primary purpose of the Expert Edge, the tests demonstrated stable performance even with such demanding tasks. The emergence of systems like this points to a growing demand for personalized computing power in the field of AI.