MINISFORUM MS-A2 vs MINISFORUM UM890 Pro

For everyday desktop work and for coding, every built configuration of both machines is well suited. The difference is scale: the MS-A2 brings 16 cores and 32 threads, and the UM890 Pro brings 8 cores and 16 threads. Both machines are well suited to coding, and the MS-A2 brings twice the cores and threads of the UM890 Pro.

Gaming reverses the order. The MS-A2's Radeon 610M is a minimal display adapter, so the MS-A2 is a poor fit for games as sold, although its card slot can take a discrete graphics card later. The UM890 Pro's Radeon 780M makes it workable for esports and older titles at 1080p, and its OCuLink port can drive an external GPU enclosure.

For local language models, memory capacity decides which model sizes fit. 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. MINISFORUM does not state memory bandwidth for either machine, so these figures cannot tell you how fast either one generates text.

The MS-A2 is well suited to a homelab, with 10GbE networking and three storage slots including U.2. The UM890 Pro is only workable there because its networking stops at 2.5GbE. As a workstation the MS-A2 is well suited thanks to its card slot with 8 lanes. The UM890 Pro is a poor fit for that role because it has no x8 or x16 card slot.

Both machines take memory in two SODIMM slots, unsoldered, up to 96 GB. A 32 GB model of either can therefore grow to 96 GB by changing modules.

Storage is where they part. The MS-A2 has three M.2 slots that take 2280 and 22110 drives, plus U.2 support. Its first slot runs at PCIe 4.0 and its second at PCIe 3.0, so put a PCIe 4.0 drive in the first slot. An M.2 drive in the second slot runs at PCIe 3.0 speed. The MS-A2 has no SATA bays, and its physical x16 slot, wired with 8 lanes, can take an add-in card. The UM890 Pro has two M.2 2280 slots, and MINISFORUM does not state their PCIe generation. It has no card slot, and OCuLink is its route to external hardware.

The MS-A2 has four network ports: two 10GbE SFP+ and two 2.5GbE. It also has Wi-Fi 6E and Bluetooth 5.2, on a wireless module that MINISFORUM includes. The UM890 Pro has two 2.5GbE ports. MINISFORUM does not state its Wi-Fi standard or Bluetooth version, or whether a wireless module is included.

For USB, the MS-A2 has two USB-C and three USB-A ports, with no USB4 or Thunderbolt. The UM890 Pro has two USB4 ports, two USB-C ports and four USB-A ports.

For screens, count the fixed outputs. The MS-A2 has one HDMI output and no DisplayPort, and the UM890 Pro has one HDMI output and one DisplayPort. MINISFORUM does not state the HDMI or DisplayPort versions, or the maximum number of displays, for either machine. Confirm these before planning a multi-monitor desk.

The fair matches are like for like. Both machines come as a 32 GB model with 1000 GB of NVMe storage and as a 64 GB model with 1000 GB. Both ship with Windows 11 Pro. The barebones kits also pair up, and neither can be judged until you add memory and storage. Only the MS-A2 is offered with 64 GB and 2000 GB of storage, or with 96 GB and 2000 GB. Set those against a UM890 Pro that you upgrade yourself toward its 96 GB ceiling.

If neither machine fits, the alternative is the Apple Mac mini (M6). It uses Apple M6 in place of the AMD Ryzen 9 9955HX, and its memory tops out at 32 GB rather than 96 GB. Consider it only if you need neither the memory headroom nor the slots described above. 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.