64GB max RAM · 6 bays · 2 x 2.5GbE
AINAS Score 58
No verified price

AI Ready45Drives
Why AI Ready: AMD Ryzen 5 5500GT or Ryzen 7 5700G (Zen 3 APU, Radeon iGPU, no NPU) with up to 64GB UDIMM and one single-slot full-height PCIe slot that takes a compact GPU, which meets the ready bar at the low end.
Updated
Can the 45Drives HL8 run local AI?
Partly. The 45Drives HL8 runs local models on AMD Ryzen 5 5500GT or Ryzen 7 5700G alone, and its 64GB ceiling allows 32B class models comfortably, and 70B class models at 4-bit with little headroom (Estimated), but replies will be slow without graphics acceleration.
A graphics card can be added through its PCIe x16 slot, which is the biggest single speed upgrade for local models.
Local LLM support is not yet verified.
The HL8 suits a buyer who wants an 8-bay compact desktop NAS that can also be used for local AI experiments. It takes up to 64GB of DDR4 UDIMM, with 16GB included, and the memory is expandable. It has one PCIe x16 slot for a single-slot, full-height GPU, plus 2 M.2 slots, 2.5GbE and a 500W power supply. It runs Rocky Linux with Houston UI and supports containers. Llama 3.3 70B fits at Q4 (Estimated), and every model up to 32B fits at both Q4 and Q8 (Estimated). The AINAS Score is 58. The AI capability rows are Unverified, so treat AI use as a project you set up yourself.
The main thing that rules out the HL8 for some buyers is that no local AI use is confirmed on it. Local LLM, Ollama, Open WebUI, RAG, AI photos and surveillance are all marked Unverified, and the notes say no evidence of use was found. Anyone who needs a NAS with AI software proven to work should look for a vendor-documented local LLM capability, a verified container platform, and an NPU or a confirmed GPU path. The HL8 has no NPU. Its GPU slot is limited to a card under 7.9in, powered from the slot only, so large GPUs will not fit.
No verified price
We have not verified a listing for the 45Drives HL8 at a retailer we track. The specifications below stand on their own; a price appears here as soon as a listing is confirmed.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.co.uk or Amazon.com, as applicable, at the time of purchase will apply to the purchase of this product.
Each capability shows how it was confirmed: Verified by AINAS, Vendor confirmed, Community confirmed or Unverified.
| Capability | Supported | Evidence | Note |
|---|---|---|---|
| Local LLM | Limited | Unverified | Ryzen 5000G APU with up to 64GB and a single-slot PCIe GPU option; no vendor or community evidence of local LLM use on HL4/HL8 found. |
| Ollama | Limited | Unverified | Rocky Linux with Houston UI; containers via cockpit-podman per 45Drives staff. No Ollama evidence on this platform. |
| Open WebUI | Limited | Unverified | Runnable as a container in principle; no platform evidence found. |
| RAG | Limited | Unverified | No evidence found. |
| AI search | No | Unverified | No AI search feature in Houston UI documented. |
| GPU expansion | Limited | Community confirmed | 45HomeLab staff on the official forum: single-slot, full-height card, length under 7.9in, slot power only. NASCompares lists the slot as PCIe 4.0 x16. |
| Hardware acceleration | Limited | Community confirmed | Ryzen 5700G iGPU present, but an HL4 owner reports Plex transcoding fell back to CPU; forum suggests adding a single-slot Intel Arc A310. |
| AI photos | Limited | Unverified | No bundled photo app; would need self-hosted containers. No evidence found. |
| AI video | No | Unverified | No AI video feature documented. |
| Surveillance AI | Limited | Unverified | No bundled surveillance software; self-hosted only. No evidence found. |
Calculated from the HL8's memory and each model's published size at each quantization. Rows marked Estimated are calculations, not benchmarks. Check any model.
| Model | Best quantization | Runs? | Memory needed | Expected performance | Verification |
|---|---|---|---|---|---|
| Llama 3.2 3BLlama, 3.21B | Q8 | Yes | 13GB | Not measured | Estimated |
| Llama 3.1 8BLlama, 8.03B | Q8 | Yes | 19GB | Not measured | Estimated |
| Qwen3 8BQwen, 8.19B | Q8 | Yes | 19GB | Not measured | Estimated |
| Gemma 3 12BGemma, 11.8B | Q8 | Yes | 24GB | Not measured | Estimated |
| Qwen3 14BQwen, 14.8B | Q8 | Yes | 28GB | Not measured | Estimated |
| Phi-4 14BPhi, 14.7B | Q8 | Yes | 28GB | Not measured | Estimated |
| Mistral Small 3 24BMistral, 23.6B | Q8 | Yes | 38GB | Not measured | Estimated |
| Gemma 3 27BGemma, 27B | Q8 | Yes | 44GB | Not measured | Estimated |
| Qwen3 30B A3BQwen, 30.5B | Q8 | Yes | 47GB | Not measured | Estimated |
| Qwen3 32BQwen, 32.8B | Q8 | Yes | 50GB | Not measured | Estimated |
| DeepSeek-R1 32B (Qwen distill)DeepSeek, 32.8B | Q8 | Yes | 50GB | Not measured | Estimated |
| Llama 3.3 70BLlama, 70.6B | Q4 | Yes | 60GB | Not measured | Estimated |
| Llama 4 Scout 17Bx16ELlama, 109B | None fits | No | 89GB at Q4 | Not measured | Estimated |
| Qwen2.5 72BQwen, 72.7B | Q3 | Yes | 54GB | Not measured | Estimated |
| Mistral Large 2 123BMistral, 123B | Q2 | Yes | 62GB | Not measured | Estimated |
One verdict per workload, from the HL8's use-case scores and capability rows. Unknown means we cannot confirm it, not that it probably works.
We cannot confirm the HL8 for local LLM work. The capability row is Unverified, and the notes say no vendor or community evidence of local LLM use on this platform was found.
We cannot confirm the HL8 for private RAG. The RAG, Open WebUI and Ollama rows are all Unverified, with the notes saying Open WebUI could run as a container in principle but no platform evidence exists.
The HL8 is a poor fit for AI search. The capability row says no, and the notes say Houston UI documents no AI search feature (Unverified).
We cannot confirm the HL8 for AI photos. There is no bundled photo app, so you would need self-hosted containers, and the row is Unverified with no evidence found.
We cannot confirm the HL8 for surveillance. No surveillance software is bundled and it would be self-hosted only, and the row is Unverified with no evidence found.
The HL8 is an OK home lab machine. It has 8 bays, 2 M.2 slots, 2.5GbE, expandable memory up to 64GB and container support through Rocky Linux with Houston UI.
The HL8 is an OK choice for business use. It has 8 SATA bays and a 500W power supply, but no ECC memory is listed and networking is a single 2.5GbE port.
The HL8 is OK for creative AI. The PCIe x16 slot takes a single-slot, full-height GPU (Community confirmed, limited), and hardware acceleration is limited because Plex transcoding reportedly fell back to the CPU.
Out of the box the HL8 has 16GB; the official ceiling is 64GB of DDR4 UDIMM. A memory upgrade is the cheapest way to make it more useful for AI.
Working backwards from 64GB: reserve about 8GB for the operating system and storage services, and at roughly 0.6GB per billion parameters at 4-bit, the largest model that fits is around 93B (Estimated).
It can. A 1x PCIe x16 (single-slot, full-height card only) is the route, and a card with 16GB or more of its own memory changes what the HL8 can run. Measure the space and check the 500W power budget before ordering.
For a personal library, yes. A compact embedding model and a small chat model will run, but the HL8 is better as the document store behind a RAG system whose model runs on stronger hardware.
Built-in networking: 2.5GbE. Adequate for a household, slower for bulk indexing.
| Configuration | RAM | SSD | Notes |
|---|---|---|---|
| Fully Built, Ryzen 5 5500GT, 16GB, 500GB boot | 16GB | 500GB | |
| Fully Built, Ryzen 5 5500GT, 32GB, 500GB boot | 32GB | 500GB | |
| Fully Built, Ryzen 5 5500GT, 64GB, 500GB boot | 64GB | 500GB |
The closest 45Drives model by bay count and AINAS Score. Only the specifications that differ are listed.
Its strongest AINAS profiles are RAG at 60 and Local LLM at 59, which points to anyone building private search and question answering over their own documents.
Look elsewhere if you are teams moving large files over the network every day.
Other AI Ready systems closest in overall score.
Workloads it suits
Yes at Q4, as an estimate. Llama 3.3 70B needs 43GB at Q4 and the HL8 takes up to 64GB, so it fits (Estimated). At Q8 it needs 75GB, so it does not fit (Estimated). This is a calculation of memory, not a benchmark, and no local LLM use on the HL8 is confirmed.
Yes. The HL8 uses DDR4 UDIMM and takes up to 64GB, with 16GB included. The configurations sold have 16GB, 32GB and 64GB. The memory sits in standard UDIMM slots rather than being soldered, so you can change it later. No ECC memory is listed for the HL8.
Yes, within limits. It has one PCIe x16 slot for a single-slot, full-height card. Staff on the official forum say the card must be under 7.9in and run on slot power only. That is Community confirmed, and the notes call the support limited.
Yes, 2. One is used for the boot drive and one for an M.2 SATA controller, according to a third-party review. Treat that detail with caution, because 45Drives itself only lists the boot drive, at 500GB or 1TB.
Yes, in the form of containers. The HL8 runs Rocky Linux with Houston UI, and 45Drives staff say containers are managed through the cockpit-podman add-on, which is Podman rather than Docker. Ollama and Open WebUI on this platform are Unverified.
No. 45Drives describes the HL8's memory as DDR4 UDIMM and does not state ECC support, so the HL8 is recorded without ECC. One third-party spec table lists 16GB of DDR4-3200 ECC, but the same review's text only says DDR4 UDIMM, which leaves that claim unconfirmed. Buyers who need error-corrected memory for business file storage should treat the HL8 as a non-ECC system and look for a system that lists ECC memory in its specification.
Partly. Ollama support is not yet verified. See the capability table above for the source.
Partly. Open WebUI support is not yet verified. See the capability table above for the source.
Partly. AI photos support is not yet verified. See the capability table above for the source.
Partly. Surveillance AI support is not yet verified. See the capability table above for the source.
Partly. Hardware acceleration support is confirmed by community reports. See the capability table above for the source.
It has 2.5GbE built in, and the 1x PCIe x16 (single-slot, full-height card only) can take a faster network card. The fastest port is 2.5GbE.
Every field records its source. Specifications were last checked on . Spot an error? Send a correction with a manufacturer link.