MINISFORUM MS-A2

It suits you if you run virtual machines, containers or a storage server and want fast links to a switch or NAS: the two SFP+ ports, the two 2.5GbE ports and memory up to 96 GB carry the homelab case. Developers compiling large projects get sixteen cores and thirty-two threads.

Skip it if you want to game on it as delivered, because the AMD Radeon 610M is a minimal display adapter and no graphics card is fitted. If you plan on several monitors, check first: the one confirmed video output is a single HDMI port.

Everyday use is well suited on every built model: sixteen cores with 32 GB to 96 GB of DDR5 is far beyond what browsing and office work need.

Coding is well suited, with thirty-two threads for parallel builds and two SODIMM slots taking up to 96 GB.

Gaming is a poor fit: the Radeon 610M is integrated graphics for driving a screen, and the PCIe slot could take a discrete card, but none is installed.

Homelab work is well suited, with four network ports, three storage slots including U.2 and a 96 GB memory ceiling.

Local AI depends on memory. For local AI, check the exact model release, quantization, context length, runtime overhead and memory available after the operating system. System RAM and GPU VRAM are different limits; shared-memory allocation depends on firmware, drivers and the backend. CPU inference or partial offloading can change what loads and its speed. Installed capacity alone does not establish compatibility or useful generation speed. Check the current Ollama or LM Studio hardware documentation for the exact GPU and operating system. That judgment rests on capacity, and generation speed is not measured.

Workstation use is well suited thanks to the PCIe card slot with 8 lanes.

The barebones kit is not assessed until built, since it is sold without memory or storage.

All five versions share the Ryzen 9 9955HX and Radeon 610M; memory, storage and what comes in the box differ. The four built models ship new with Windows 11 Pro on NVMe storage: 32 GB with 1000 GB, 64 GB with 1000 GB, 64 GB with 2000 GB, and 96 GB with 2000 GB. For local AI, check the exact model release, quantization, context length, runtime overhead and memory available after the operating system. System RAM and GPU VRAM are different limits; shared-memory allocation depends on firmware, drivers and the backend. CPU inference or partial offloading can change what loads and its speed. Installed capacity alone does not establish compatibility or useful generation speed. Check the current Ollama or LM Studio hardware documentation for the exact GPU and operating system. The barebones kit makes sense if you already own SODIMMs and NVMe drives.

Memory sits in two SODIMM slots and is not soldered, so the 32 GB and 64 GB models can grow to the 96 GB maximum.

Storage has three M.2 slots taking 2280 or 22110 drives, plus U.2 support. The first slot runs at PCIe Gen 4 and the second at PCIe Gen 3, so the first is the place for your fastest drive. There are no SATA bays. For expansion cards, the PCIe slot offers 8 lanes and can take a discrete graphics card.

Networking is the standout: two 10GbE SFP+ ports and two 2.5GbE ports. Wi-Fi 6E and Bluetooth 5.2 are built in. Peripherals get two USB-C ports and three USB-A ports, and there are no USB4 or Thunderbolt ports.

For screens, count on one HDMI output; there is no DisplayPort output, and MINISFORUM does not state whether the USB-C ports carry video, the HDMI version or the maximum number of displays. Power comes from an external 19V 12.63A supply.

The biggest gap is graphics: the Radeon 610M is no gaming chip and no card comes fitted.

MINISFORUM does not state the memory speed or bandwidth, ECC support, graphics memory, M.2 SATA support, HDMI version, maximum displays, USB-C DisplayPort alt mode, OCuLink, an SD card reader, the TDP or a VESA mount. For the barebones kit it also does not state the operating system, and memory and storage must be added before it runs.

The Apple Mac mini (M6) swaps the Ryzen 9 9955HX for an Apple M6, and its memory tops out at 32 GB against 96 GB here, so it only makes sense if your projects and models fit that ceiling.

The MINISFORUM MS-02 Ultra uses an Intel Core Ultra 9 285HX and takes up to 256 GB, so look at it if 96 GB would cap your virtual machines or AI models. Stay with the MS-A2 if 96 GB, its network ports and the PCIe slot cover your plans.