HP ZGX Nano G1n NVIDIA GB10 AI Station - 128GB / 4TB
PRODUCT DESCRIPTION
HP ZGX Nano G1n AI Station Personal AI Supercomputer, NVIDIA GB10 Grace Blackwell Superchip, 20-Core Arm CPU, 128GB Unified Memory, 4TB Self-Encrypting NVMe SSD, ConnectX-7 200GbE, Wi-Fi 7 (DD8J0PA) - 3 Years Onsite Local Warranty
HP ZGX Nano G1n AI Station — HP's GB10 Desktop AI Supercomputer with 3-Year Onsite Support in Singapore
The HP ZGX Nano G1n AI Station (part number DD8J0PA) is HP's build of the NVIDIA GB10 Grace Blackwell platform: a 150 × 150 × 51 mm mini desktop pairing a 20-core Arm CPU with a Blackwell GPU and 128 GB of coherent unified memory for up to 1,000 TOPS (1 petaFLOP) of FP4 AI performance. HP positions it as a personal AI supercomputer for prototyping, fine-tuning and running inference on models of up to 200 billion parameters on a single unit, so AI developers, data scientists and enterprise teams in Singapore can work on large language models locally rather than renting cloud GPUs.
What sets the HP version apart is the support model and the HP Z tooling. SourceIT supplies this DD8J0PA configuration with a 3-year onsite local warranty, serviced in Singapore, and a full GST invoice; HP adds the HP Z Toolkit (also called the ZGX Toolkit), a free Visual Studio Code extension that discovers the ZGX Nano on the network and manages projects, experiments and model serving from a Windows 11 or Ubuntu client. It suits organisations that already standardise on HP Z workstations and want desktop-scale AI with a familiar procurement and support path.
NVIDIA GB10 Grace Blackwell Superchip — 1 PetaFLOP of FP4 AI Compute
The GB10 Superchip combines a 20-core Arm CPU (10 Cortex-X925 performance cores and 10 Cortex-A725 efficiency cores) with an NVIDIA Blackwell GPU on a single package, connected by NVLink-C2C so CPU and GPU share one memory space. Fifth-generation Tensor Cores with native FP4 and FP8 precision deliver up to 1 PFLOP of sparse FP4 throughput; fourth-generation RT Cores accelerate rendering and simulation; and one NVENC and one NVDEC engine handle video for computer-vision and multimodal pipelines. NVIDIA rates the GB10 at a 140 W TDP for CPU and GPU combined.
128 GB Coherent Unified Memory at 273 GB/s — Models up to 200 Billion Parameters
Conventional workstations run out of VRAM long before a 70B or 120B model loads. The ZGX Nano's 128 GB of LPDDR5x-8533 is a single coherent pool shared by CPU and GPU at 273 GB/s, so open-weight model families such as Llama, Qwen, DeepSeek, Gemma, Mistral and gpt-oss load in full. A single unit runs inference on models up to 200 billion parameters (FP4), fine-tunes models up to 70 billion parameters, and gives data scientists a 128 GB workspace for RAPIDS, embeddings and large-dataset analytics without paging to disk. The memory is soldered and, per HP, not upgradeable.
4 TB Self-Encrypting NVMe SSD, ConnectX-7 200 GbE, 10 GbE, Wi-Fi 7 and Three 20 Gbps USB-C Ports
DD8J0PA ships with a 4 TB PCIe Gen4 x4 NVMe TLC M.2 2242 SSD with OPAL self-encryption (SED), rated by HP at up to 7,000 MB/s sequential read and 6,000 MB/s write — room for multiple multi-hundred-gigabyte checkpoints, datasets and container images with data at rest protected. Networking is where a GB10 system separates itself from an ordinary mini PC: an integrated NVIDIA ConnectX-7 controller drives two QSFP ports at 200 Gbps signalling each for linking units or attaching fast storage, alongside a 10 GbE RJ-45 port for the office network and a MediaTek MT7925 Wi-Fi 7 (2×2) and Bluetooth 5.4 module. Three USB 3.2 Type-C ports at 20 Gbps with DisplayPort 1.4a Alt Mode, HDMI 2.1a with multichannel audio and a USB Type-C power input complete the rear panel.
Link Two ZGX Nano Units for Models up to 405 Billion Parameters
Two ZGX Nano units connected over ConnectX-7 with a QSFP112 direct-attach cable work as one system on models of up to 405 billion parameters — the clustering configuration HP documents. HP's option kit is the HP ZGX 0.4m QSFP112 Ethernet DAC Cable (C20MQAA), which SourceIT stocks alongside the system. NVIDIA supports up to four GB10 systems for models of up to 700 billion parameters at the platform level; HP has not published a four-unit ZGX Nano configuration, so confirm larger topologies with SourceIT.
NVIDIA DGX OS 7, CUDA 13, DGX Dashboard, NVIDIA AI Workbench and the HP Z Toolkit
The ZGX Nano boots into NVIDIA DGX OS 7, an Ubuntu 24.04-based Linux build with the NVIDIA AI stack ready on day one. HP lists NVIDIA Connect, CUDA 13, the NVIDIA RTX Driver 580, DGX Dashboard and NVIDIA AI Workbench as preinstalled, and the same foundation runs TensorRT-LLM, NVIDIA NIM, NeMo, Triton, RAPIDS, NGC containers, PyTorch, JAX, vLLM, Ollama and Hugging Face tooling. NVIDIA's DGX Spark Playbooks provide step-by-step recipes for local LLM serving, fine-tuning, RAG and agents, NVIDIA AI Enterprise for DGX Spark is available for production deployments, and recent DGX OS updates add streamlined NemoClaw installation — NVIDIA's open-source reference stack that wraps OpenClaw agents with OpenShell security guardrails. HP's layer on top is the HP Z Toolkit, a free Visual Studio Code extension for a Windows 11 or Ubuntu 24.04 client that pairs with the ZGX Nano, tracks experiments in MLflow, serves models with Ollama and exports them for deployment; HP claims an average 45 percent time saving versus a do-it-yourself setup. HP Remote System Controller adds KVM and virtual-media access to BIOS level.
240 W USB-C Power, 1.25 kg, Quiet Enough for a Desk
Per HP's QuickSpecs the ZGX Nano runs from a 240 W external USB Type-C power adapter at 89 percent efficiency with active PFC, so it needs nothing more than a standard wall socket — no dedicated circuit, no rack, no liquid cooling. HP's declared noise is 22.0 dB sound pressure at idle and 27.6 dB during disk writes. The black chassis measures 15 × 15 × 5.1 cm without feet, weighs from 1.25 kg and carries EPEAT Gold registration; HP notes it is not qualified for rack mounting. An integrated TPM 2.0 (convertible to FIPS 140-2 mode) and the self-encrypting SSD protect data at rest.
Available in Singapore Through SourceIT
SourceIT is an authorised IT reseller in Singapore and supplies the HP ZGX Nano G1n AI Station (DD8J0PA) with a full GST invoice, 3-year onsite local warranty serviced in Singapore, and enterprise, research-grant and education procurement assistance. Pair it with the HP ZGX 0.4m QSFP112 DAC cable for a two-unit cluster, read the NVIDIA DGX Spark Setup and Enterprise Deployment Guide, and browse every GB10 and GB300 system in the NVIDIA DGX Spark & DGX Station collection; HP's step-up GB300 system is the HP ZGX Fury G1n.
HP ZGX Nano G1n AI Station Datasheet — Technical Specifications
General
| Product Name | HP ZGX Nano G1n AI Station — 128GB / 4TB |
| Part Number | DD8J0PA |
| Product Family | HP Z AI Stations (ZGX series), NVIDIA GB10 platform |
| Category | Personal AI supercomputer / AI developer workstation |
| Form Factor | Mini desktop, 150 × 150 mm footprint |
| Colour | Black |
| Local Warranty (Singapore) | 3-year onsite local warranty, serviced in Singapore. Supplied by SourceIT with a GST invoice |
Processor & GPU — NVIDIA GB10 Grace Blackwell Superchip
| Architecture | NVIDIA Grace Blackwell — CPU + GPU on one superchip, NVLink-C2C |
| CPU | 20-core Arm: 10 × Cortex-X925 performance cores + 10 × Cortex-A725 efficiency cores |
| GPU | Integrated NVIDIA Blackwell architecture GPU, Blackwell-generation CUDA cores |
| Tensor Cores | 5th generation, FP4 / FP8 optimised |
| RT Cores | 4th generation |
| GB10 TDP | 140 W (CPU + GPU), per NVIDIA |
AI Performance
| Peak AI Compute | Up to 1,000 TOPS (1 PFLOP) FP4 Tensor performance with sparsity |
| Inference | AI models up to 200 billion parameters on a single unit, per HP |
| Fine-Tuning | AI models up to 70 billion parameters on a single unit, per NVIDIA |
| 2 Linked Units | AI models up to 405 billion parameters over ConnectX-7 with a QSFP cable, per HP |
| Up to 4 Linked Units | NVIDIA platform figure: up to 700 billion parameters; not published by HP for ZGX Nano |
Memory
| System Memory | 128 GB LPDDR5x-8533 coherent unified system memory, shared by CPU and GPU |
| Memory Bandwidth | 273 GB/s |
| Upgradeability | Soldered — not upgradeable |
Storage
| SSD | 4 TB PCIe Gen4 x4 NVMe TLC M.2 2242, SED OPAL self-encrypting |
| SSD Performance | Up to 7,000 MB/s sequential read, 6,000 MB/s sequential write; 2,400 TBW endurance (HP QuickSpecs) |
Networking
| High-Speed NIC | NVIDIA ConnectX-7 200 GbE controller — 2 × QSFP112 ports, 200 Gbps signalling per port, IEEE 802.3ck |
| Ethernet | 1 × RJ-45 10 GbE (Realtek RTL8127-CG; 1 / 2.5 / 5 / 10 Gbps) |
| Wi-Fi | MediaTek MT7925 Wi-Fi 7 (802.11be, 2×2 dual-band), WPA3 |
| Bluetooth | Bluetooth 5.4 |
I/O & Display
| Rear Ports | 1 × USB Type-C power input (240 W); 3 × USB 3.2 Type-C 20 Gbps with DisplayPort 1.4a Alt Mode (30 W total output); 1 × HDMI 2.1a; 1 × RJ-45 10 GbE; 2 × QSFP 200 GbE |
| Audio | HDMI multichannel audio output |
Software & Security
| Operating System | NVIDIA DGX OS 7 (Ubuntu 24.04 based), arm64. Microsoft Windows is not supported |
| Preinstalled Software (HP) | NVIDIA Connect, CUDA 13, RTX Driver 580, DGX Dashboard, NVIDIA AI Workbench |
| NVIDIA AI Stack | CUDA, cuDNN, TensorRT / TensorRT-LLM, NVIDIA NIM, NVIDIA NeMo, Triton, RAPIDS, NGC containers, PyTorch, TensorFlow, JAX, vLLM, Ollama |
| HP Tooling | HP Z Toolkit (ZGX Toolkit) VS Code extension for Windows 11 / Ubuntu clients; HP Remote System Controller |
| Security | Integrated TPM 2.0 (convertible to FIPS 140-2 certified mode); SED OPAL storage encryption; AMI UEFI BIOS |
Power, Physical, Acoustics & Certifications
| Power Supply | 240 W external USB Type-C AC adapter, 89% efficiency, active PFC; 90–264 V AC (HP QuickSpecs; option kit CM5A0AA) |
| Dimensions (W × D × H) | 15 × 15 × 5.1 cm without feet; 15 × 15 × 5.45 cm with feet |
| Weight | Starting at 1.25 kg |
| Declared Noise (ISO 7779 / 9296) | Idle 3.4 B sound power / 22.0 dB sound pressure; random disk writes 3.9 B / 27.6 dB |
| Ecolabels / Compliance | EPEAT Gold (IEEE 1680.1); RoHS; TAA compliant |
Compatibility & Software Support
What Runs on the HP ZGX Nano
Anything in the mainstream NVIDIA and open-source AI ecosystem published for arm64: Hugging Face Transformers, vLLM, Ollama, llama.cpp, LangChain and LlamaIndex RAG pipelines, NVIDIA NIM, NeMo, TensorRT-LLM, ComfyUI and FLUX-class image generation, RAPIDS, and the NVIDIA Isaac, Metropolis and Holoscan frameworks. NVIDIA's DGX Spark Playbooks apply directly because the ZGX Nano runs the same DGX OS as every GB10 system.
Arm64 Architecture — Plan Your Software Accordingly
The GB10's CPU is Arm-based, so DGX OS 7, drivers and container images are arm64 builds. Every major AI framework ships arm64 builds, but proprietary x86-only applications will not run natively, and HP states that Microsoft Windows is not supported. The intended workflow is a Windows 11 or Ubuntu client driving the ZGX Nano through the HP Z Toolkit, SSH, Jupyter or the DGX Dashboard.
Clustering and Networking
The two ConnectX-7 QSFP ports link two ZGX Nano units back-to-back with a single QSFP112 direct-attach cable — HP's option kit is the HP ZGX 0.4m QSFP112 Ethernet DAC Cable (C20MQAA) — to pool 256 GB of unified memory for models up to 405 billion parameters. The same NIC supports RDMA to NVMe-over-fabric storage; the 10 GbE RJ-45 handles site networking. HP also offers a radio-free ZGX Nano variant for secure and regulated environments; ask SourceIT if you require it.
NVIDIA GB10 Platform Siblings
The ZGX Nano shares its GB10 Superchip, 128 GB unified memory and 1 PFLOP FP4 with the NVIDIA DGX Spark and the GB10 desktops from ASUS, Dell, Lenovo, GIGABYTE, MSI, Acer and xFusion. They differ in chassis, storage tier, warranty term, bundled tooling and price; the HP unit is distinguished by the HP Z Toolkit, Remote System Controller support and the 3-year onsite local warranty SourceIT supplies it with in Singapore.
Package Contents & Ordering Information
What Ships in the Box
- HP ZGX Nano G1n AI Station (DD8J0PA) with NVIDIA GB10 Grace Blackwell Superchip, 128 GB unified memory and 4 TB self-encrypting NVMe SSD
- 240 W USB Type-C AC adapter and power cord
- NVIDIA DGX OS 7 preinstalled with the NVIDIA AI software stack
- HP documentation (no keyboard or mouse included)
The HP ZGX 0.4m QSFP112 Ethernet DAC Cable (C20MQAA) for linking two units is sold separately.
HP ZGX Nano G1n Configurations
- HP ZGX Nano G1n AI Station 4TB (DD8J0PA) (this listing) — NVIDIA GB10, 128 GB unified memory, 4 TB SED NVMe, DGX OS 7, 3-year onsite local warranty
- HP also lists 1 TB and 2 TB configurations and a radio-free variant for regulated environments in other regions — email SourceIT to check Singapore availability
Other NVIDIA GB10 and DGX Station Systems at SourceIT
- NVIDIA DGX Spark 4TB (940-54242-0007-000) and DGX Spark Founders Edition 4TB (940-54242-0009-000) — NVIDIA's first-party GB10 systems
- ASUS Ascent GX10 — 1TB, 2TB and 4TB
- Dell Pro Max with GB10 2TB and 4TB
- Lenovo ThinkStation PGX 1TB and 4TB
- GIGABYTE AI TOP ATOM — 4TB PCIe 5.0, 4TB PCIe 4.0, 1TB PCIe 4.0
- MSI EdgeXpert MS-C931 4TB
- Acer Altos BrainSphere GB10 F1 4TB
- xFusion FusionXpark GB10 — 1TB, 2TB, 4TB
- HP ZGX Fury G1n (NVIDIA DGX Station GB300) — HP's step-up system with 748 GB coherent memory and up to 20 PFLOPS; GB300 alternatives: MSI XpertStation WS300, ASUS ET900N G3, GIGABYTE W775-V10-L01, Supermicro ARS-511GD-NB-LCC
Browse the NVIDIA DGX Spark & DGX Station collection and the wider AI workstations and AI servers collection. For volume, research-grant or education pricing, GST invoicing and lead times, email sales@sourceit.com.sg quoting DD8J0PA.
Datasheet, Downloads & Setup Guide
Official HP and NVIDIA documentation and resources for the ZGX Nano G1n AI Station:
- HP ZGX Nano AI Station — official HP product page
- HP ZGX Nano G1n AI Station QuickSpecs (HP document c09212373)
- HP ZGX Nano G1n AI Station datasheet (PDF)
- NVIDIA DGX Spark — GB10 platform page and specifications (Singapore)
- NVIDIA DGX Spark Playbooks — step-by-step recipes for LLMs, fine-tuning, RAG and agents
- NVIDIA DGX Spark Setup and Enterprise Deployment Guide — SourceIT
Frequently Asked Questions — HP ZGX Nano G1n AI Station
What is the HP ZGX Nano G1n AI Station?
HP's personal AI supercomputer on the NVIDIA GB10 Grace Blackwell platform: a 15 × 15 × 5.1 cm mini desktop with a 20-core Arm CPU, Blackwell GPU, 128 GB of unified memory and up to 1,000 TOPS of FP4 AI performance, running NVIDIA DGX OS 7 with the HP Z Toolkit.
What is the configuration of DD8J0PA?
NVIDIA GB10 Grace Blackwell Superchip, 128 GB LPDDR5x-8533 unified memory, 4 TB PCIe Gen4 NVMe SED SSD, ConnectX-7 with two QSFP 200 GbE ports, 10 GbE, Wi-Fi 7, three 20 Gbps USB-C ports, HDMI 2.1a, 240 W USB-C adapter and NVIDIA DGX OS 7, supplied by SourceIT with a 3-year onsite local warranty.
Is the HP ZGX Nano the same as the NVIDIA DGX Spark?
Both use the NVIDIA GB10 platform with 128 GB unified memory, 1 PFLOP FP4 and NVIDIA DGX OS, so raw AI performance is the same. The HP ZGX Nano G1n adds the HP Z Toolkit and Remote System Controller support in an HP black chassis and is supplied in Singapore with a 3-year onsite local warranty; the DGX Spark is NVIDIA's gold Founders Edition hardware.
How does the ZGX Nano compare to the other GB10 desktops from ASUS, Dell, Lenovo, GIGABYTE, MSI, Acer and xFusion?
Identical GB10 compute and memory; they differ in chassis, storage tiers, bundled tooling, warranty term and price. SourceIT stocks all of them; the HP unit suits teams that want HP Z tooling and onsite support.
What AI models can a single ZGX Nano run?
Inference on models up to 200 billion parameters and fine-tuning up to 70 billion — in practice Llama, Qwen, DeepSeek, Gemma, Mistral and gpt-oss-class LLMs and FLUX-class image generation from the 128 GB unified memory. Two linked units reach 405 billion parameters.
Can I link two or more ZGX Nano units together?
Yes. HP documents linking two ZGX Nano units through the ConnectX-7 QSFP ports with the HP ZGX 0.4m QSFP112 DAC cable (C20MQAA), sold separately, for models up to 405 billion parameters. NVIDIA supports up to four GB10 systems for 700-billion-parameter models at the platform level, but HP has not published a four-unit ZGX Nano configuration.
What operating system and software come preinstalled?
NVIDIA DGX OS 7 (Ubuntu 24.04 based) with CUDA 13, the NVIDIA RTX Driver 580, DGX Dashboard and NVIDIA AI Workbench; the full NVIDIA AI stack runs natively on top. The HP Z Toolkit is a free download for your Windows 11 or Ubuntu client, and NVIDIA AI Enterprise for DGX Spark is available for production deployments.
Does the ZGX Nano run Windows or x86 software?
No. HP states that Microsoft Windows is not supported. The Grace CPU is Arm64, so the ZGX Nano runs arm64 Linux software and containers, accessed from a Windows 11 or Ubuntu client over the network or with a monitor and keyboard attached directly.
Does it need special power, cooling or a server room?
No. It runs from a 240 W external USB-C adapter on a standard wall socket, is actively air-cooled and HP's declared noise is 22.0 dB at idle — quiet enough for an open-plan office. HP does not support rack mounting.
Can the memory or storage be upgraded?
The 128 GB LPDDR5x unified memory is soldered and HP lists the product as not upgradeable. Storage is a single M.2 2242 SSD; HP publishes a service video for SSD replacement, and any drive change should be discussed with SourceIT to keep warranty cover intact.
Who is the HP ZGX Nano for?
AI developers prototyping and fine-tuning models before scaling up; researchers and university labs needing on-premise compute; data scientists running large-memory analytics; and Singapore enterprises in finance, healthcare and government that must keep data on-premises and prefer HP's onsite support.
What warranty do I get in Singapore?
3-year onsite local warranty, serviced in Singapore. Supplied by SourceIT with a GST invoice.
How do I buy the HP ZGX Nano G1n in Singapore?
Order online on this page or email sales@sourceit.com.sg quoting DD8J0PA. SourceIT is an authorised IT reseller in Singapore and supplies the ZGX Nano with GST invoice, onsite local warranty and the HP QSFP112 cable for clustering.
