MINISFORUM N5 MAX AI NAS

The MINISFORUM N5 MAX is a 10-bay AI NAS powered by the AMD Ryzen AI Max+ 395, a Strix Halo processor with 16 cores, 32 threads, a maximum clock of 5.1 GHz, and a 70W to 90W TDP range. Paired with AMD Radeon 8060S graphics, the platform delivers up to 126 TOPS of local AI performance. This is not a conventional NAS built around a low-power file-serving processor; the N5 MAX is designed to run 120B-class open models on the same box that stores the data, which removes the usual split between a storage appliance and a separate compute machine. It combines five hot-swap SATA bays and five M.2 NVMe slots for up to 200TB of hybrid storage, up to 128GB of unified LPDDR5x memory, dual 10GbE networking, and USB4 V2 at 80Gbps. MinisCloud OS ships preinstalled, and the chassis measures 199 × 202.4 × 252.3 mm at 5.8 kg with a built-in 250W power supply.The N5 MAX uses a two-tier storage design that separates bulk capacity from working speed. Five 3.5-inch or 2.5-inch SATA 3.0 bays accept drives up to 32TB each for 160TB of bulk capacity, intended for archives, project libraries, media assets, and long-term retention. Five M.2 NVMe slots add up to 40TB of flash: one M.2 2280 slot at PCIe 4.0 x4 supporting up to 8TB, one M.2 2230 or 2280 slot at PCIe 4.0 x1 supporting up to 8TB, and three M.2 2280 slots at PCIe 4.0 x1 supporting up to 8TB each. Total system capacity reaches 200TB. The NVMe tier is intended for cache, hot data, model files, and active project work, accelerating an HDD array that would otherwise be bound by spinning-disk seek times. Supported RAID levels are RAID 0, RAID 1, RAID 5 and RAIDZ1, and RAID 6 and RAIDZ2, and mixed strategies can place hot data on flash while cold backups sit on the SATA tier.Memory is onboard LPDDR5x unified memory with a maximum capacity of 128GB on a 256-bit bus. The default operating speed is 8000 MT/s, with 8533 MT/s available as a user-configurable overclocking option that MINISFORUM does not guarantee for sustained operation. In MINISFORUM benchmark testing using llama.cpp llama-bench with 99 GPU layers and Flash Attention enabled, the 128GB configuration produced the following results: gpt-oss-120b in MXFP4 quantization at 59.03GB reached 59.30 tokens per second text generation and 668.79 tokens per second prompt processing; Qwen3.5-122B-A10B in Q4_K_M at 69.11GB reached 31.76 tokens per second generation and 317.58 prompt processing; gpt-oss-20b in MXFP4 at 11.28GB reached 83.09 tokens per second; and Gemma 4 26B-A4B in Q4_K_M at 15.64GB reached 70.33 tokens per second. PP512 measures prompt-processing throughput and TG128 measures text-generation speed. Actual results vary with model version, quantization, context length, system configuration, and software environment, and the figures above were produced on the 128GB configuration rather than the 64GB one.Networking consists of two 10GbE RJ45 ports using Realtek RTL8127 controllers, which connect to a copper 10GbE switch without transceivers. The rear panel adds two USB4 V2 ports at 80Gbps with DisplayPort 2.0 Alt Mode and 15W Power Delivery, one USB 3.2 Gen2 port, one USB 3.2 Gen1 port, one USB 2.0 port, and one HDMI 2.1 FRL output supporting 8K at 60Hz or 4K at 144Hz. The front panel carries one USB4 port at 40Gbps and one USB 3.2 Gen2 port, plus a status LED, two NIC LEDs, and five per-drive LEDs. An internal USB 3.2 Gen2 header allows a boot device to be installed without consuming an external port or an M.2 slot. The system ships with MinisCloud OS, which includes ZFS snapshots, LZ4 compression with single-core speeds above 500 MB/s, multi-user isolation, per-account Docker container isolation, and the MinisOpenClaw local AI agent running Qwen3.6-35B by default; Windows 11 and Linux are also supported. A slide-out motherboard bay simplifies servicing.Common use cases for the MINISFORUM N5 MAX include running 120B-class open models such as gpt-oss-120b and Qwen3.5-122B-A10B entirely on-device, private AI assistants and agent workflows through MinisOpenClaw with no cloud API costs and no data leaving the machine, AI semantic search across a personal or team photo and document library through MinisPhotos, ZFS storage arrays with snapshot-based recovery for creative studios and small teams, media servers and archives where a 160TB HDD tier is accelerated by a 40TB NVMe cache tier, Docker-based self-hosted services with per-account container isolation, and 10GbE workgroup file serving for video and photo production. Buyers typically choose the N5 MAX over a conventional NAS when local AI inference and large-capacity storage need to live in the same box, and over a separate workstation-plus-NAS pairing when a single 5.8 kg chassis is preferable to maintaining two devices. Buyers who need only file serving without local AI will find the N5 Pro or N5 AIR a better value.