HP Z2 Mini G1a Workstation

The Z2 Mini G1a is the first Z2 Mini to go AMD, and the 'a' suffix marks a genuine change of philosophy rather than a spec bump. Instead of squeezing a discrete pro GPU into a tiny chassis, HP has leaned on AMD's Ryzen AI Max+ PRO 395 — 16 Zen 5 cores with Radeon 8060S graphics sharing a single 128GB DDR5 pool — and run it at a 150 W TDP, more than double what the same silicon gets in HP's ZBook Ultra G1a laptop. That headroom is the whole point: this is the desktop version of the chip, allowed to actually stretch its legs.

The headline capability is memory. Up to 96GB of that shared pool can be handed to the integrated GPU as VRAM, which is well beyond what discrete cards in this class offer and explains why StorageReview was able to run GPT-OSS 120B locally on a box with no add-in GPU at all. There's also a 50 TOPS NPU alongside it. That combination points squarely at a specific buyer: engineers, AEC professionals and AI developers who need large models or large datasets resident in GPU memory, in a machine barely bigger than an ATX power supply that can hang off the back of a monitor.

At $5,021 as tested — with 2TB of dual HP Z Turbo NVMe storage and a three-year warranty — this is not a general-purpose small desktop, and it shouldn't be evaluated as one.

Judged on what it's actually for, the Z2 Mini G1a is one of the more interesting workstations of its generation. The unified-memory approach sidesteps the single most frustrating constraint in small-form-factor professional machines, and doing it at 150 W rather than the mobile chip's 70 W means the sustained performance should hold up rather than sagging after the first few minutes. Reviewers report solid single- and multi-core results, generous GPU performance, modular Flex I/O and the usual battery of HP professional management features.

The compromises are real, though. Significant fan noise under load is the predictable cost of pushing this much silicon through a chassis this small, and it matters more than usual in a machine designed to sit on your desk at eye level. Connectivity is described as slightly quirky, and the bundled software load is heavier than a $5,000 professional system deserves. There's also a structural caveat worth flagging: CPU, GPU and memory scale together, so the cheaper 32GB configuration drops you to a Ryzen AI Max Pro 380 with 8040S graphics. You cannot buy the big memory pool cheaply, and the big memory pool is the reason to buy this machine.

The deeper strategic question is that integrated graphics, however capacious, is still integrated graphics — and unlike a tower, there's no path to adding a discrete GPU later if your workload turns out to want raw compute more than capacity. Buyers should be confident their workloads are memory-bound before committing.

Where it fits: if you are running large local models or memory-hungry professional applications and floor space is scarce, there is very little else that does this, and it earns a strong recommendation. If you just want a fast small workstation, the price and the noise make the calculus much less flattering. Four stars, with the reservation that the rating is specific to the AI-and-memory use case this machine was clearly built around.