What can the N5 Pro run?
The N5 Pro runs Jellyfin with hardware transcoding on its Radeon iGPU (Plex's AMD support is "as is"), full Docker via MinisCloud and real virtual machines. On its AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads) with 16 GB (to 96 GB), the sweet spot is Jellyfin, Immich, Home Assistant, Nextcloud.
A five-bay NAS built around a laptop-class AI processor: the Ryzen AI 9 HX PRO 370 with Radeon 890M graphics and a 50 TOPS NPU, two DDR5 SO-DIMM slots to 96 GB with ECC support, one 10GbE and one 5GbE port, and a U.2-capable SSD tier. It is the rare NAS where memory, not the CPU, is the upgrade that matters: fill both slots and local AI runs about twice as fast.
As an Amazon Associate we earn from qualifying purchases on Amazon.ca.
Everything the N5 Pro can run, at a glance
Rated from this model’s real AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads) and its 16 GB–96 GB of RAM. “Limited” and “No” are honest: an ARM box does not virtualize, and no hardware transcode is no hardware transcode.
| App | Runs? | Performance | What it needs on this model |
|---|---|---|---|
| Plex | Yes | fair | Radeon iGPU; Plex AMD support is 'as is' |
| Jellyfin | Yes | good | HW transcode on the Radeon iGPU (VA-API) |
| Immich | Yes | good | 8 GB+ handles ML search |
| Home Assistant | Yes | good | Container or full HA OS in a VM |
| Nextcloud | Yes | good | Runs well in Docker |
| Frigate | Yes | fair | Radeon iGPU decode; ROCm detection is unofficial, so add an accelerator |
| Pi-hole / AdGuard Home | Yes | good | Lightweight container |
| Paperless-ngx | Yes | good | Runs well in Docker |
| Vaultwarden | Yes | good | Lightweight container |
| Minecraft server | Yes | good | Enough RAM for a world |
| Docker (general) | Yes | good | Full container platform (MinisCloud) |
| Virtual machines | Yes | good | Enough RAM for a VM |
| Surveillance recording | Yes | good | No vendor NVR app; run Frigate in Docker |
| Time Machine / backup target | Yes | good | Any model |

App by app on the N5 Pro
Plex
The N5 Pro runs Plex, and its AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads) has a Radeon iGPU with a hardware video engine, but Plex says its AMD support is provided "as is" and recommends Intel Quick Sync or NVIDIA. Direct-play over 1 × 10GbE and 1 × 5GbE is flawless; for a Plex household that transcodes a lot, Jellyfin (below) makes better use of this chip.
Jellyfin
Jellyfin is the media server that suits the N5 Pro: it runs in MinisCloud's Docker and hardware-transcodes on the Radeon iGPU through VA-API, which Jellyfin documents as fully supported on Linux for AMD graphics. Pair it with 1 × 10GbE and 1 × 5GbE and it serves a whole household.
Immich
The N5 Pro makes a capable Immich server: a self-hosted Google Photos with face and object search. It runs as a Docker stack under MinisCloud, and its 16 GB out of the box and room for 96 GB covers the 8 GB the machine-learning wants. No upgrade needed to get going.
Home Assistant
On the N5 Pro you get both Home Assistant paths: the lightweight Container in MinisCloud's Docker, or the full HA OS in a virtual machine, since the 12-core AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads) supports VMs and can lend a guest some of its 96 GB ceiling.
Nextcloud
For a full private cloud (files, calendar, contacts, office) the N5 Pro hosts Nextcloud in MinisCloud's Docker. The 12-core AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads) and 16 GB out of the box and room for 96 GB keep it snappy for a family; it wants more headroom than a simple file share, so do not starve it.
Frigate
Frigate runs on the N5 Pro, and the Radeon iGPU can decode the camera streams, but Frigate calls ROCm on integrated GPUs unofficial ("does work with most of them"), so for detection a USB or M.2 accelerator is the dependable choice. Several cameras over 1 × 10GbE and 1 × 5GbE are well within reach.
Pi-hole / AdGuard Home
The N5 Pro makes a fine home for Pi-hole or AdGuard Home: the container barely registers next to storage, and because the box is already on all day, it is a natural network-wide DNS blocker for everything on your 1 × 10GbE and 1 × 5GbE.
Paperless-ngx
Paperless-ngx runs in MinisCloud's Docker on the N5 Pro; the 12-core AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads) handles a household's OCR in bursts, with 16 GB out of the box and room for 96 GB to keep indexing quick.
Vaultwarden
Run Vaultwarden on the N5 Pro and your passwords live on hardware you control: it is a tiny, Bitwarden-compatible vault that works with the official apps and browser extensions, using almost none of MinisCloud's resources.
Minecraft server
The N5 Pro can host a Minecraft world in MinisCloud's Docker: a world wants ~4 GB, and this model offers 16 GB out of the box and room for 96 GB. The AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads) and that RAM keep a small server smooth.
Docker (general)
The N5 Pro runs a full container platform through MinisCloud, which is what makes it more than a 5-bay disk box: Immich, Nextcloud, Pi-hole and the rest all run here. How many you stack comes down to 16 GB out of the box and room for 96 GB on the 12-core AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads).
Virtual machines
The N5 Pro is a real VM host: the 12-core AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads) virtualizes and 16 GB out of the box and room for 96 GB lets you hand a guest a proper slice while the NAS keeps serving over 1 × 10GbE and 1 × 5GbE. Fill it toward 96 GB if VMs are the point.
Surveillance recording
Minisforum documents no camera-recording app for MinisCloud OS, so on the N5 Pro surveillance means an NVR in Docker, such as Frigate, on the 12-core AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads). Size the 5 bays by cameras × retention, and use surveillance-tuned drives (WD Purple, SkyHawk) for the continuous writes.
Time Machine / backup target
Backups are the N5 Pro's day job: it serves as a Time Machine and Windows/phone backup target over SMB, and its 1 × 10GbE and 1 × 5GbE link keeps whole-machine backups quick. Cap each user's quota so the 5 bays are not swallowed by old snapshots.
Local AI on the N5 Pro
Expect 4.9 to 8.2 tok/s from Llama 3.1 8B on the N5 Pro at a 4k conversation, which is comfortable reading speed. The model takes 5.9 GB of the 14.0 GB left free in 16 GB, so there is headroom for other apps.
The limit is memory bandwidth: a language model reads all its weights for every word. 5600 MT/s DDR5 SO-DIMM (ECC support) in the N5 Pro moves 44.8 GB/s on one stick, which is what we assume because Minisforum does not say whether the stock RAM is one stick or two. That figure, not the CPU core count, is what sets the pace. Its 50 TOPS NPU does not help: Ollama does not use NPUs.
The largest model we would run day to day on the N5 Pro is Gemma 3 4B (7.7 to 13 tok/s). Best upgrade: a matching second RAM stick. It fills the empty memory channel, so chat speed rises about 2x, and doubles memory to 32 GB. That is enough for Qwen3 30B-A3B.
Two other AI jobs a NAS does: Frigate camera detection (ROCm on the AMD iGPU (unofficial in Frigate, works on most)) and Immich photo search (GPU-accelerated through ROCm (experimental in Immich)).
| Llama 3.1 8B (Q4, 4k context, Ollama's default) | 4.9 to 8.2 tok/s, comfortable reading speed (medium confidence) |
|---|---|
| Largest comfortable model (5+ tok/s) | Gemma 3 4B |
| AI photo search (Immich) | GPU-accelerated through ROCm (experimental in Immich) |
| AI camera detection (Frigate) | ROCm on the AMD iGPU (unofficial in Frigate, works on most) |
| Best single upgrade | Second RAM stick (to 32 GB, about 2x faster) |
Try other models on the N5 Pro in the AI NAS calculator › · Compatible RAM for the N5 Pro ›
The N5 Pro at a glance
| CPU | AMD Ryzen AI 9 HX PRO 370 (12 cores, 24 threads) |
|---|---|
| RAM | 16 GB DDR5 SO-DIMM 5600, ECC supported (2 slots), up to 96 GB |
| Bays | 5 |
| Container platform | Yes, MinisCloud |
| Virtual machines | Yes |
| Plex/Jellyfin transcode | Radeon iGPU: Jellyfin hardware transcode; Plex AMD support is "as is" |
| OS | MinisCloud OS |
Read more
- What can I run? The full matrix
- Best drives for the N5 Pro
- RAM for the N5 Pro
- N5 Pro capacity calculator
- What it costs to run (power calculator)
- AI on a NAS: what actually matters
Frequently asked questions
Can the N5 Pro run Plex with 4K transcoding?
With caveats. The N5 Pro's Radeon iGPU can hardware-transcode, and Jellyfin supports it fully on Linux through VA-API. Plex says its AMD support is provided "as is" and recommends Intel Quick Sync or NVIDIA, so for Plex-first households an Intel box is the safer pick.
Can the N5 Pro run Docker?
Yes. MinisCloud includes a full container platform, so the N5 Pro runs the whole self-hosted world (Pi-hole, Vaultwarden, Nextcloud and more), limited only by its 16–96 GB of RAM.
Can the N5 Pro run virtual machines?
Yes. The N5 Pro supports hardware virtualization, and with up to 96 GB of RAM it can give a Windows or Linux guest a useful slice while still serving files.
Can the N5 Pro run Immich?
Yes. With its 8 GB+ of RAM, the N5 Pro runs Immich's machine-learning photo search comfortably in Docker, a strong self-hosted Google Photos replacement.
Can the N5 Pro run Home Assistant?
Yes. Home Assistant Container runs in Docker on the N5 Pro and barely touches the CPU, and you can run the full Home Assistant OS in a VM if you prefer add-ons.
Is the N5 Pro a good home server, or just storage?
It is a genuine home server: full Docker (and VMs) turn it into a platform for Plex, photos, automation and more, not just a file box. RAM is the main lever on how much you stack.
Can the N5 Pro run a local AI chatbot like Ollama?
Yes. Llama 3.1 8B at the default Q4 runs at about 4.9 to 8.2 tok/s on the N5 Pro at Ollama's default 4k context, and the largest model we would run day to day is Gemma 3 4B. Memory bandwidth, not an NPU, sets that speed.

Devin Chua works out which drives, RAM and NVMe cache fit which NAS model at nasdrives.ca, and what the RAID choice means for usable capacity, checked against what is in stock on Amazon.ca.