Skip to content
All hardware
HPAMD Strix HaloDesk tier · £2k – £7k

Z2 Mini G1a

The corporate Strix Halo — a 2.7-litre workstation with tier-1 warranty, in stock in the UK at £2,663.99.

Price

£2,663.99 inc VAT

hp.com/gb-en list (128 GB/2 TB) — seen at £2,397 with promo codes

verified 2026-07 · supply & lease options in every proposal

HP Z2 Mini G1a workstation — compact chassis, front three-quarter viewHP
128 GB
Unified memory

LPDDR5X-8533 (HP-quoted)

256 GB/s
Memory bandwidth

~212 GB/s measured

50 TOPS
XDNA 2 NPU

~126 platform TOPS

300 W
External PSU

2.7 L chassis

The machine

HP's Z2 Mini is the Strix Halo for businesses that buy through procurement: Ryzen AI Max+ PRO 395, 128 GB of LPDDR5X (HP quotes 8533 MT/s — a spec bump over the platform's usual 8000), 2 TB NVMe, Wolf security stack and 3-year warranty options, with genuine UK stock on hp.com.

It costs more than the Chinese mini-PCs but it is the only Strix box with a tier-1 OEM support story — usually the deciding factor when the machine is going into an office rather than a lab.

vs its siblings: The only Strix Halo with a tier-1 OEM support story and clean UK supply.

Memory, to scale

128 GB model-visible · bandwidth is the speed limit

Unified LPDDR5X

128 GB · 256 GB/s

LPDDR5x

For scale

DGX Spark — 128 GB @ 273 GB/s

RTX PRO 6000 — 96 GB @ 1.79 TB/s

Mac Studio M3 Ultra — 512 GB @ 819 GB/s

DGX Station GB300 — 748 GB coherent

Capability

What it actually runs

Declared from research and benchmarks, not computed marketing — tokens-per-second figures are cited where a real measurement exists.

  • gpt-oss-120bMXFP4with headroom30–55 tok/s (typ. ~45 on Linux, verified)
  • Qwen3.5 122BQ4fits~65 GB — inside the Linux VRAM window
  • Qwen3.6 35BQ4with headroom~66 tok/s — the triage engine
  • Qwen3-235B-classQ3fits~11 tok/s — possible, not pleasant
  • Llama-70B-class denseQ4fits~5 tok/s — bandwidth-starved, avoid
Specification

The full sheet

Compute

APU
AMD Ryzen AI Max+ PRO 395 — 16× Zen 5 to 5.1 GHz
GPU
Radeon 8060S — 40 CU RDNA 3.5
NPU
XDNA 2 — 50 TOPS

Memory

Unified memory
128 GB LPDDR5X-8533 (soldered)
VRAM allocation
~96 GB (Windows) · ~110 GB+ (Linux GTT)

Storage & I/O

SSD
2 TB NVMe
Support
3-year warranty options · ISV certifications · Wolf Pro Security

Software

Serving
Vulkan (best all-round) · ROCm 7.x · llama.cpp · LM Studio · Lemonade
Honest note
A MoE machine — dense 70B crawls (~5 tok/s); fine-tuning impractical

Where it shines

  • Tier-1 OEM warranty and UK stock — procurement-friendly
  • gpt-oss-120b at 30–55 tok/s for under £2.7k
  • 2.7 L, quiet, office-appropriate
  • Slightly faster memory than the platform standard

The trade-offs

  • 256 GB/s bandwidth — dense models and long prompts are slow
  • ROCm still trails CUDA; Vulkan is the reliable path
  • BIOS locks down some tuning the enthusiast boxes allow
  • 192 GB Gorgon Halo refresh lands Q3 2026 — buy knowing that

Buy this box for

The first self-hosted box in a business that buys warrantiesEmail triage, drafting and RAG on 120B-class sparse modelsPiloting local AI before committing to CUDA-class silicon
The platform

Understanding AMD Strix Halo

Ryzen AI Max+ 395

AMD's Ryzen AI Max+ 395 ('Strix Halo') puts 16 Zen 5 cores, a 40-CU Radeon 8060S GPU and up to 128 GB of unified LPDDR5X-8000 on one package. Framework, HP, GMKtec, Beelink, Corsair and AMD itself all ship it as a compact desktop, from roughly £1,700 to £2,700 in the UK.

It is a mixture-of-experts machine: gpt-oss-120b runs at 30–55 tok/s and even Qwen3-235B squeezes in at 3-bit (~11 tok/s), but dense 70B models crawl at ~5 tok/s because 256 GB/s of bandwidth is the ceiling. The software stack (Vulkan / ROCm / llama.cpp / LM Studio) is genuinely usable in 2026 but remains a step behind CUDA for production serving and fine-tuning.

It is the right first box for proving local AI on real workloads before committing to bigger silicon — and the refresh is close: AMD's Ryzen AI Max 400 series ('Gorgon Halo', announced May 2026) lifts the ceiling to 192 GB with OEM systems from Q3 2026.

Sources & verification

Specifications and prices verified 2026-07 against the sources below. The memory shortage is repricing this market monthly — we re-verify at quote.

Compare against the rest of the catalogue, or have us spec it against your workloads.

Back to catalogue
Make it real

Spec the Z2 Mini G1a for your workload

We'll size it against your real token volumes, quote supply and lease options, and stand up the models with evals that prove they do the job — or tell you honestly that a different box (or the API) wins.