Spending $7,399 on a small desktop calls for a better explanation than “it has AI.” The Minisforum MS-S1 MAX-P495 has one: 192GB of unified memory, with up to 160GB available to its integrated graphics. If the local AI model you want to use keeps running out of memory, that’s a much more useful starting point than a processor badge.
The US store currently lists the Ryzen AI Max+ PRO 495 configuration with 192GB of RAM and a 2TB SSD for $7,399. It’s a compact workstation aimed at people who want to run demanding models on their own hardware. The buying question is whether you need that much memory in one machine, and whether your software can use it well.
What the extra memory actually buys
Unified memory is a shared pool that the CPU and integrated GPU can access. The P495’s advertised 160GB graphics allocation comes out of its 192GB total. You don’t get 192GB of system RAM plus a separate 160GB graphics card.
Compared with the earlier MS-S1 MAX’s maximum 128GB configuration, that’s 50% more total memory. Minisforum says the maximum graphics allocation rises from 96GB to 160GB. For a workload that exceeds the older machine’s available graphics memory, the difference can determine whether it fits on one system.
Capacity lets you hold more data; processing power determines how quickly you can work through it. More memory won’t automatically make a smaller model generate answers faster. Nor does having 160GB available to integrated graphics make this equivalent to a dedicated GPU with the same memory capacity.
AMD specifies LPDDR5X-8533 memory on a 256-bit interface. The previous MS-S1 MAX used LPDDR5X-8000, so the rated transfer speed increase is about 6.7%, much smaller than the jump in capacity. That’s a useful clue about what kind of upgrade this is.
What the local AI demonstrations show
In its October 8 announcement, Minisforum reports running DeepSeek V4 Flash 284B at Q4 quantization on one system at approximately 12 tokens per second.
“284B” refers to the model’s parameter count; it isn’t a measure of answer quality or speed. Q4 describes a reduced-precision representation that lowers memory requirements. The model file, working memory and conversation context all need room, so a large RAM figure alone doesn’t establish which settings will run comfortably.
Minisforum also quotes approximately 16 tokens per second for Qwen3.5-397B on two linked machines. Its four-system example uses DeepSeek-R1 671B Q4, with a stated 380GB model size. The Qwen result uses two PCs, while the DeepSeek-R1 demonstration uses four.
The ports are worth a closer look
The P495 measures 222.1 × 206.3 × 77.1mm, or roughly 8.7 × 8.1 × 3 inches. Its internal 320W power supply helps keep a separate power brick off the desk.
There are two 10Gb Ethernet ports for fast wired networking, two rear USB4 v2 ports rated at 80Gbps and two front USB4 ports rated at 40Gbps. Wi-Fi 7 and Bluetooth 5.4 are also listed. Fast connections are useful when moving large model files or working with network storage, although they don’t guarantee fast distributed AI inference.
Expansion comes with a detail that’s easy to miss: the internal PCIe slot is physically x16 but electrically PCIe 4.0 x4. The two M.2 storage slots aren’t identical either. One uses PCIe 4.0 x4 and the other PCIe 4.0 x1. Check those limits before planning add-in hardware or assuming both SSDs will have the same bandwidth.
Who should consider it?
The Ryzen AI Max+ PRO 495 supplies 16 Zen 5 CPU cores, 32 threads, boost speeds up to 5.2GHz and Radeon 8065S graphics with 40 compute units. The NPU is rated at up to 55 TOPS, or trillion operations per second. That peak rating isn’t a chatbot speed test, and it doesn’t tell you how much of your chosen application will run on the NPU.
This is a more convincing purchase for someone who can name the workload that needs the extra memory than for someone simply shopping for a powerful mini PC. A developer keeping a large model local has a clear reason to investigate it. For everyday desktop work, the price is hard to justify.
If your models already fit comfortably on a smaller-memory PC, capacity alone is a weak reason to spend $7,399. For workloads that exceed the older machine’s available memory, though, 192GB offers a concrete benefit: room to keep more of the work on one compact system.
Minisforum MS-S1 MAX-P495: 192GB RAM + 2TB SSD
Price: US$7,399
Ryzen AI Max+ PRO 495 configuration. Check current delivery timing before ordering.
Where to buy: Check the 192GB configuration at Minisforum
