128GB max RAM · 6 bays · 2 x 10GbE
AINAS Score 75
£774
Amazon UK · Verified

AI Ready45Drives
Why AI Ready: AMD EPYC 7252 to 7452 with no NPU or integrated GPU, PCIe 4.0 x16 slots taking a GPU up to 10.25in, and up to 512GB DDR4 from the store; 45Drives does not document GPU inference on this model, so it stays ready under the Xeon or EPYC rule.
Updated
Can the 45Drives HL15 run local AI?
Yes, the 45Drives HL15 can run local language models. It pairs AMD EPYC 7252, 7272, 7282 or 7452 with up to 512GB of memory, which suits 70B class models at higher-quality quantization, or large mixture-of-experts models at 4-bit (Estimated).
It also has a GPU path (PCIe x16 slot), so a graphics card can be added later for much faster inference.
Local LLM support is confirmed by community reports.
The HL15 suits buyers who want a 15-bay storage server that can also host local models and containers. It has an AMD EPYC CPU (8 cores on the 7252), 16GB of DDR4 ECC-capable RDIMM or LRDIMM memory included and up to 512GB available. It has 2 x 10GbE and 2 OCuLink ports. Bays are 15 and there are 2 M.2 slots. It runs Rocky Linux 9 with Houston UI and manages containers through Podman. On model memory, the largest model listed on this page, Qwen3 235B A22B, fits at Q4 (Estimated). It scores 74 on the AINAS Score.
The main catch is the GPU path. The EPYC options have no integrated GPU and no NPU, so acceleration needs an added card. 45Drives does not document GPU inference on this model. The chassis takes GPUs up to 10.25in, confirmed by vendor staff, but no GPU is sold as a listed option. The working LLM setup we found is a community owner build on Proxmox, not the Rocky Linux 9 and Houston UI setup it comes with. Buyers who want a supported, ready-to-run AI box should look for a vendor-documented GPU option, an integrated GPU or an NPU. Buyers who are happy to fit and configure their own card will not be held back.
No verified price
We have not verified a listing for the 45Drives HL15 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 | Yes | Community confirmed | Owner build dated 2026-08-16 adds an RTX 4070 Ti and runs Ollama with GPU passthrough on Proxmox. No 45Drives statement on LLM use. |
| Ollama | Yes | Community confirmed | Same owner build runs Ollama in an LXC container. Not the stock OS. |
| Open WebUI | Limited | Community confirmed | Owner build in the HL15 v1 form factor on the same ROMED8-2T board reports Open WebUI; seen in a search extract only. |
| RAG | Limited | Unverified | Runnable as containers in principle; no platform evidence found. |
| AI search | No | Unverified | No AI search feature documented in Houston UI. |
| GPU expansion | Yes | Vendor confirmed | 45HomeLab staff state the 4U chassis fits GPUs up to 10.25in. No GPU is sold as a listed option in the fetched store data. |
| Hardware acceleration | Limited | Community confirmed | EPYC options list no integrated GPU, so acceleration needs an added card. |
| AI photos | Limited | Community confirmed | Same owner build runs Immich in an LXC container. No bundled photo app. |
| AI video | No | Unverified | No AI video feature documented. |
| Surveillance AI | Limited | Unverified | No bundled surveillance software; self-hosted containers only, no evidence found. |
Calculated from the HL15'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 | Q8 | Yes | 98GB | Not measured | Estimated |
| Llama 4 Scout 17Bx16ELlama, 109B | Q8 | Yes | 149GB | Not measured | Estimated |
| Qwen2.5 72BQwen, 72.7B | Q8 | Yes | 101GB | Not measured | Estimated |
| Mistral Large 2 123BMistral, 123B | Q8 | Yes | 164GB | Not measured | Estimated |
One verdict per workload, from the HL15's use-case scores and capability rows. Unknown means we cannot confirm it, not that it probably works.
The HL15 is a good fit for local LLMs. Community owners confirm Ollama with GPU passthrough on Proxmox, though 45Drives makes no statement on LLM use, and up to 512GB of memory holds large models (Estimated).
Private RAG is an acceptable fit on the HL15. The RAG row is unverified and limited, with no platform evidence, while Open WebUI is limited but community confirmed by an owner build on the same board.
AI search is a poor fit for the HL15. No AI search feature is documented in Houston UI, and that capability row is unverified and marked as not supported.
AI photos work well on the HL15, with a caveat. An owner build runs Immich in an LXC container, community confirmed, but support is limited because there is no bundled photo app.
We cannot confirm surveillance on the HL15. It ships with no bundled surveillance software, only self-hosted containers, and the capability row is unverified with no evidence found.
The HL15 is a good home lab machine. It has 15 bays, 2 x 10GbE, 2 M.2 slots, up to 512GB of ECC-capable DDR4 and Podman container management through Houston UI.
The HL15 is a good fit for business use. It pairs an EPYC CPU with ECC-capable RDIMM or LRDIMM memory, 15 drive bays, 2 x 10GbE and a 1000W power supply in a rackmount chassis.
Creative AI is a good fit if you add a card. Vendor staff confirm GPU expansion up to 10.25in, but hardware acceleration is limited because the EPYC options have no integrated GPU.
The HL15 ships with 16GB of DDR4 RDIMM or LRDIMM and supports up to 512GB with ECC. Upgrading to the maximum lifts the model ceiling from 7B to 8B class models to 70B class models at higher-quality quantization, or large mixture-of-experts models at 4-bit (both Estimated).
As a rule of thumb, a 4-bit model needs about 0.6GB per billion parameters plus room for context. After leaving around 8GB for the NAS operating system, 512GB leaves space for a model of roughly 840B parameters at most (Estimated).
ECC memory support is a plus for always-on systems that also hold important files.
Yes. The HL15 has a PCIe 4.0 x16 slots on ASRock ROMED8-2T; exact count not readable on the store listing, some drop to x8 when NVMe or OCuLink is in use that can take a graphics card. A discrete GPU with its own memory runs models far faster than CPU inference and lets the NAS serve several users at once. Check card length, height and power draw against the chassis and the 1000W power supply first.
It also has 2 OCuLink ports, a second route to an external GPU dock.
With limits. The HL15 can host a small RAG stack such as AnythingLLM or Open WebUI with a compact embedding model, but memory and compute restrict how large the answering model can be. It suits a personal document library more than a team knowledge base.
Networking is 2 x 10GbE. That is ample for indexing large libraries and for several people querying at once.
| Configuration | RAM | SSD | Notes |
|---|---|---|---|
| Fully Built, EPYC 7252, 16GB | 16GB | None | |
| Fully Built, EPYC 7252, 64GB | 64GB | None | |
| Fully Built, EPYC 7252, 128GB | 128GB | None |
The closest 45Drives model by bay count and AINAS Score. Only the specifications that differ are listed.
The HL15 scores best for Home Lab (82/100) and RAG (79/100). It suits home lab owners who want one box for storage, containers, media and a few AI services.
Other AI Ready systems closest in overall score.
Workloads it suits
Yes, on memory alone. Qwen3 235B A22B needs 142GB at Q4 and the HL15 takes up to 512GB, so it fits at Q4 (Estimated). Q8 needs 250GB and also fits (Estimated). This is a calculation, not a benchmark. Running it in practice also needs software set up on the system, and a GPU would have to be added by you.
Yes. Vendor staff state the 4U chassis fits GPUs up to 10.25in, and the board has PCIe 4.0 x16 slots. No GPU is sold as a listed option, so you would supply your own. Some slots drop to x8 when NVMe or OCuLink is in use.
Yes. The memory is expandable rather than soldered. It includes 16GB and the largest configuration option is 512GB as DDR4 RDIMM or LRDIMM. The memory is ECC-capable, though 45Drives does not state ECC in words, so confidence is medium.
It has 2 OCuLink PCIe 4.0 x4 ports, so the connection is physically there. We found no confirmed external GPU setup over OCuLink. Using OCuLink can also drop some PCIe slots to x8.
Not Docker itself. Houston UI manages Podman containers through cockpit-podman, and the HL15 runs Rocky Linux 9. Container support is recorded as present, with medium confidence. Community owners run Ollama and Immich in LXC containers on their own Proxmox builds, not on Rocky Linux 9 with Houston UI.
Three, all Fully Built with an AMD EPYC 7252: one with 16GB of memory, one with 64GB and one with 128GB. The CPU options also include the EPYC 7272, 7282 and 7452, and the largest memory option is 512GB as 8x 64GB DIMMs of DDR4 RDIMM or LRDIMM. The boot drive size is not recorded for these builds. Every build has 15 bays, 2 M.2 slots that take PCIe 4.0 x4 or SATA drives, 2 x 10GbE, 2 OCuLink ports and a 1000W power supply.
Yes. Ollama support is confirmed by community reports. See the capability table above for the source.
Partly. Open WebUI support is confirmed by community reports. See the capability table above for the source.
Partly. AI photos support is confirmed by community reports. 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 x 10GbE built in, and the PCIe 4.0 x16 slots on ASRock ROMED8-2T; exact count not readable on the store listing, some drop to x8 when NVMe or OCuLink is in use can take a faster network card. The fastest port is 2 x 10GbE.
Every field records its source. Specifications were last checked on . Spot an error? Send a correction with a manufacturer link.