HP ZGX Nano G1n

The HP ZGX Nano G1n is a niche monster built for local AI development, packing a 1,000 TOPS NPU and 128GB of unified memory into a tiny box. CPU and RAM scores are the best we've ever recorded in this category, but the integrated GPU is a letdown for anything outside AI. If you're prototyping models and find one near the $6,160 mark, it's a great buy. If you want to game or edit video, look literally anywhere else.

The CPU and memory are the absolute best we've tracked in this category, sitting at the 100th percentile. That 20-core Grace chip and 128GB of unified LPDDR5x RAM chew through data-heavy workloads without breaking a sweat. The 4TB NVMe drive is also a standout, landing in the 98th percentile for storage speed and capacity. But the GPU score tells a different story. At the 11th percentile, the integrated Blackwell graphics are clearly not built for traditional gaming or 3D rendering benchmarks. This machine is all about the NPU and those 1,000 TOPS of AI compute. For model inferencing and fine-tuning, it's a monster. For playing Cyberpunk, it's a very expensive paperweight.

Stacked against traditional workstations like the Dell Tower Plus or the CLX SET, the ZGX Nano is in a different league for AI but falls flat for general use. Those machines will run circles around it in gaming and GPU rendering benchmarks. The Lenovo Legion and ASUS ROG systems are gaming-first beasts with discrete RTX cards that the Nano simply can't match in frame rates. But none of them can touch the Nano's 1,000 TOPS NPU or unified memory architecture for large language models. The MSI EdgeXpert sits somewhere in the middle, but it's still a generalist. The ZGX Nano is a specialist, and it wins its niche by a mile while losing everywhere else.

Value here is entirely dependent on your workload. If you're renting cloud GPUs for AI prototyping, the math on a $6,160 machine pays for itself fast. But the price spread is brutal. We saw listings from $6,160 all the way up to $10,207. That's a $4,047 gap for the same hardware. If you're buying, hunt for the lower end of that range. At $10k, you're creeping into territory where a multi-GPU workstation might make more sense for broader tasks. For pure AI development, the lower price is a solid deal. For anything else, it's a tough sell.

HP and NVIDIA teamed up to build something genuinely different with the ZGX Nano G1n. This isn't a desktop for spreadsheets or gaming. It's a local AI powerhouse crammed into a 1.25kg box, built around the wild GB10 Grace Blackwell Superchip. You get 20 CPU cores, a Blackwell GPU, a dedicated AI NPU cranking out 1,000 TOPS, and 128GB of unified memory. The idea is simple: give developers and researchers a machine that can prototype, fine-tune, and run large models right on their desk without renting cloud GPUs by the hour.

It ships with NVIDIA DGX OS, a curated open-source stack, and dual 200G QSFP112 ports for linking multiple units together. The spec sheet is bonkers for a mini PC. But that exotic hardware comes with a very specific focus, and a price tag that swings wildly from $6,160 to over $10,000 depending on where you look. This is a purpose-built tool, and if your purpose doesn't involve AI development, you're looking at the wrong machine.

Skip this entirely if you need a machine for gaming, video editing, or general office work. The integrated GPU is one of the weakest we've seen for traditional graphics tasks, and the specialized DGX OS will just get in your way. If you're not actively working with AI models, a standard desktop or workstation will serve you better for less money.

This is for AI developers, researchers, and data scientists who need a local sandbox for large models. If you're fine-tuning LLMs, running inference, or prototyping with an open-source stack, the ZGX Nano is a compact, quiet, and absurdly capable tool. It's not a gaming PC, not a video editing rig, and not a general-purpose workstation. Know what you're getting into.