
HP ZGX Nano G1n AI Station 128GB 4TB
Black Friday price: you save $365.71 until Dec 1. No code needed.
Paid in Bitcoin, Monero, USDT or another coin you choose.
Free insured shipping. US delivery in about 5 to 11 business days. Shipping details
Overview
The HP ZGX Nano G1n AI Station is HP's NVIDIA GB10 mini workstation for local AI development. The GB10 Grace Blackwell Superchip pairs a 20-core Arm CPU (10 Cortex-X925 and 10 Cortex-A725, 16MB L2) with an integrated Blackwell GPU, sharing 128GB of LPDDR5x-8533 memory; NVIDIA rates the platform at up to 1 petaFLOP of FP4 AI compute (ref.). This SKU (CZ2V8UT#ABA) has a 4TB PCIe Gen4 NVMe SSD and runs NVIDIA DGX OS with HP's ZGX Toolkit. Notebookcheck reviewed it as compact server power built on DGX Spark, and HP.com currently shows it out of stock.
The chassis is 150 x 150 x 51 mm without feet and starts at 2.76 lb. HP lists a 280W supply with up to 92% efficiency, feeding the unit through a USB-C power connector. Rear ports are three USB-C 20Gbps, one HDMI 2.1a, one 10GbE RJ45 and two 200Gbps QSFP ports on the NVIDIA ConnectX-7 controller, plus MediaTek Wi-Fi 7 (2x2) and Bluetooth 5.4. HP covers it with a 1-year limited parts and labor warranty.
The 128GB unified memory runs Llama 3.3 70B at 4-bit (about 42.5GB) with long context, 120B-class mixture-of-experts models, and NVIDIA cites up to 200B parameters for inference (ref.). The 4TB drive leaves room for a large local model library and datasets. Two units connect over QSFP for bigger models, and the full CUDA stack makes it a fit for fine-tuning and building agents that will later deploy on NVIDIA servers.
- NVIDIA GB10 Grace Blackwell
- 128GB LPDDR5x-8533 unified
- 4TB PCIe Gen4 NVMe SSD
- ConnectX-7, 2x 200G QSFP
Specifications
| Superchip | NVIDIA GB10 Grace Blackwell |
|---|---|
| CPU | 20-core Arm: 10 Cortex-X925 + 10 Cortex-A725, 16MB L2 |
| AI compute | Up to 1 petaFLOP FP4 (ref.) |
| Memory | 128GB LPDDR5x-8533, onboard |
| Storage | 4TB PCIe Gen4 NVMe M.2 SSD |
| OS | NVIDIA DGX OS |
| Power supply | 280W, up to 92% efficiency |
|---|---|
| Dimensions | 150 x 150 x 51 mm (without feet) |
| Weight | Starting at 2.76 lb |
| Warranty | 1 year limited, parts and labor |
| USB-C | 3x USB-C 20Gbps, 1x USB-C power |
|---|---|
| Video | 1x HDMI 2.1a |
| Ethernet | 1x RJ45 10GbE |
| Cluster | NVIDIA ConnectX-7, 2x QSFP 200Gbps |
| Wireless | Wi-Fi 7 (MediaTek MT7925), Bluetooth 5.4 |
Questions and answers
Still unsure? Visit the help center or message us.
How does the HP ZGX Nano G1n compare with the NVIDIA DGX Spark?
The HP ZGX Nano G1n AI Station uses the same GB10 Superchip and 128GB of unified memory. HP adds its own chassis, the ZGX Toolkit, support and warranty, and this model has a 4TB SSD.
Can two HP ZGX Nano units be clustered?
Yes. Each unit has two 200Gbps QSFP ports on an NVIDIA ConnectX-7 controller, so two units can be linked directly to run larger models.
Which operating system does it use?
It ships with NVIDIA DGX OS, a Linux distribution with the NVIDIA AI software stack, CUDA and drivers preinstalled.
How large a model fits in 128GB?
Llama 3.3 70B at 4-bit uses about 42.5GB, so it fits with room for context. NVIDIA rates GB10 systems for inference on models up to about 200B parameters.
Is this new?
Yes. It ships new and factory-sealed, with the maker's warranty, insured and tracked to your door.
Pairs well with
Picked to go with this product. Check sizes and connectors for your build before you order.







