Workstation support
A clearly identified business configuration, documented components, and on-site service can matter more than the lowest purchase price on a professional workstation.
Compact business workstation
HP Z2 Mini G1a for local LLMs
The Z2 Mini G1a combines Ryzen AI Max+ PRO 395 and Radeon 8060S graphics with an internal 300 W power supply, Thunderbolt 4, and workstation support. A 128 GB configuration is the capacity-first choice for many quantized 70B models.
Bottom line
Why choose the Z2 Mini over a consumer mini PC?
Clean desktop setup
The power supply is internal. Two Thunderbolt 4 ports, mini DisplayPort, USB-C/A, 2.5GbE, and Wi-Fi 7 cover a professional desk well.
Large shared pool
The Radeon 8060S can use a large portion of LPDDR5X memory. That enables more model capacity than typical consumer GPUs, though not their peak memory bandwidth.
Specific complete system
128 GB / 1 TB for large models
The linked BN8E8UA configuration emphasizes memory capacity and support. Its 1 TB drive is a modest start for a large model library, so plan the second M.2 slot early.
BN8E8UA#ABA · 128 GB / 1 TB
HP Z2 Mini G1a
- Ryzen AI Max+ PRO 395Radeon 8060S and XDNA 2 NPU
- 128 GB LPDDR5X-8533soldered unified memory
- 1 TB NVMetwo M.2 2280 slots
- Windows 11 Prothree-year workstation warranty
Best use
Support-led 70B workstation
Good for: quantized 70B models, professional local RAG, and desks where Thunderbolt 4, an internal power supply, and on-site support justify the price premium.
Less suitable for: buyers seeking the lowest cost per gigabyte, a large included model drive, or CUDA-only software.
HP configuration specifications* Paid link. We may earn a commission if you buy; your price is unchanged. As an Amazon Associate we earn from qualifying purchases. Recheck the product number, memory, storage, seller, and warranty before buying.
Practical fit
What 128 GB changes
| Workload | Fit | What to check |
|---|---|---|
| Quantized 32B | ample memory headroom | choose runtime for preferred speed and compatibility |
| Quantized 70B | many configurations fit | UMA allocation, context cache, and OS reserve |
| RAG plus embeddings | room for parallel services | model-library storage can exceed the included 1 TB |
| CUDA-only workflow | poor fit | verify a Radeon backend or choose an NVIDIA system |
| Unattended business use | support package is useful | confirm the warranty attached to the exact product number |
Before ordering
Four concrete checks
Product number
Do not compare “Z2 Mini G1a” alone. Product number, processor, memory, SSD, operating system, and warranty must all match.
UMA and runtime
Check the maximum graphics-memory allocation and Vulkan or HIP/ROCm support in the application you intend to use.
SSD expansion
Two M.2 slots are useful, but installed occupancy can vary. Budget for a second model drive when maintaining several large quantizations.
Service package
Warranty and on-site terms depend on the product number and seller. Confirm the included duration in the actual offer details.
Alternatives
Smaller chassis, faster networking, or more GPU speed?
The EVO-X2 emphasizes a flatter consumer system, the GTR9 Pro dual 10GbE, and an RTX 5090 tower high inference speed inside a 32 GB VRAM envelope.