HP ZGX Fury G1n NVIDIA DGX Station GB300 AI Workstation
Email sales@sourceit.com.sg or use Request for Quotation below and our team will confirm the latest price, availability and lead time.
PRODUCT DESCRIPTION
HP ZGX Fury AI Station G1n NVIDIA DGX Station GB300 Grace Blackwell Ultra AI Workstation, 72-Core Grace CPU, 748GB Coherent Memory, 20 PFLOPS FP4, HP Z Turbo Drive PCIe Gen5 NVMe, Dual 400G ConnectX-8, Liquid-Cooled Rackable Tower (E6ZS0PT / E6ZS5PT) - 3 Years Local Warranty
HP ZGX Fury AI Station G1n β NVIDIA DGX Station GB300 AI Workstation with 3-Year Onsite Warranty
The HP ZGX Fury AI Station G1n is HP's build of the NVIDIA DGX Station GB300 platform: an NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip pairing a 72-core Grace CPU with a Blackwell Ultra GPU across 748 GB of coherent memory, delivering up to 20 petaFLOPS of FP4 AI compute for training, fine-tuning and deploying LLMs of up to 1 trillion parameters, AI agents, generative AI and scientific simulation β entirely within your own infrastructure. HP positions it for development and fine-tuning of 100-billion-class models and inference on trillion-parameter-class models from a single deskside or rack-mounted system.
Unlike other GB300 stations, the ZGX Fury G1n is backed in Singapore by HP's 3-year (3/3/3) limited warranty covering parts, labour and onsite repair, and it is available in factory-built configurations with an NVIDIA RTX PRO Blackwell display GPU and up to 12 TB of HP Z Turbo Drive PCIe Gen5 NVMe storage. It is the natural choice for enterprises, financial institutions, government agencies and research institutes in Singapore that already standardise on HP Z workstations and want data-centre-class AI with a familiar support model.
NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip β 20 PFLOPS of FP4 AI Compute
The GB300 Desktop Superchip joins one NVIDIA Grace CPU with 72 Arm Neoverse V2 cores to one NVIDIA Blackwell Ultra GPU over a 900 GB/s NVLink-C2C interconnect, so the CPU and GPU address one coherent memory space rather than shuttling data across PCIe. Per NVIDIA's DGX Station datasheet the Blackwell Ultra GPU delivers up to 20 PFLOPS of FP4 Tensor Core performance with sparsity, 10 PFLOPS FP8/FP6, 5 PFLOPS FP16/BF16, 2.5 PFLOPS TF32 and 80 TFLOPS FP32, with seven NVDEC and seven nvJPEG decoders for video and image pipelines. Multi-Instance GPU (MIG) partitions the GPU into up to seven isolated instances so a team can share one system for concurrent training, fine-tuning and inference jobs.
748 GB Coherent Unified Memory for Frontier-Scale Models
252 GB of HBM3e GPU memory at 7.1 TB/s and 496 GB of LPDDR5X CPU memory (four 128 GB SOCAMM LPDDR5X-6400 modules) at 396 GB/s unite over the 900 GB/s NVLink-C2C link into a single 748 GB coherent memory space β enough to load trillion-parameter models at FP4, run long-context RAG pipelines, fine-tune foundation models and drive digital-twin and engineering simulation without CPUβGPU offloading. NVIDIA notes that two linked DGX Station systems present more than 1.4 TB of coherent memory and up to 40 petaFLOPS of AI compute.
Factory Configurations β RTX PRO 2000 or RTX PRO 4000 Blackwell, 8 TB or 12 TB NVMe
Every ZGX Fury G1n shares the same GB300 Superchip, 748 GB coherent memory, 72-core Grace CPU, dual 400G ConnectX-8 networking and 3/3/3 warranty; the configurations differ in the professional display GPU and NVMe storage. E6ZS0PT#AB4 pairs an NVIDIA RTX PRO 2000 Blackwell (16 GB GDDR7) with 4 Γ 2 TB HP Z Turbo Drive PCIe Gen5 NVMe SSDs (8 TB). E6ZS5PT#AB4 steps up to an NVIDIA RTX PRO 4000 Blackwell SFF (24 GB GDDR7) with 2 Γ 2 TB + 2 Γ 4 TB NVMe (12 TB). A headless configuration with no discrete graphics card and 4 Γ 2 TB self-encrypting drives (SED) is available to order for remote-accessed or compliance-driven deployments β part number confirmed at quotation. Select your configuration from the dropdown above.
Dual 400G ConnectX-8 Networking, Gen5 / Gen6 NVMe and PCIe 5.0 Expansion
Two QSFP112 ports on the NVIDIA ConnectX-8 SuperNIC run at 400 Gb/s each (up to 800 Gb/s) for linking two systems, private AI clusters and NVMe-over-fabric storage, alongside a 10 GbE RJ45 port for site networking and a dedicated 1 GbE BMC port for out-of-band management. Two embedded M.2 PCIe Gen5 slots off the CPU hold the OS in software RAID 1, and two M.2 PCIe Gen6 slots off the ConnectX-8 hold data in software RAID 0 β all SED-capable. Three PCIe slots (one Gen5 x16 and two Gen5 x16 at x8 electrical) take the display GPU and further NICs or accelerators. Front I/O provides two USB-C 10 Gbps, two USB-A 5 Gbps and an audio jack; the rear adds four USB-A 10 Gbps, audio in/out and BMC-only micro-USB and Mini DisplayPort 2.0.
Liquid-Cooled Rackable Tower β Deskside Today, 5U Rack Tomorrow
HP builds the ZGX Fury G1n as a liquid-cooled rackable tower β 44.6 Γ 22.15 Γ 66.7 cm in standard tower orientation, starting at 41 kg β with front-to-back airflow and rack rails included, so it can sit beside a desk today and move into a standard 5U rack position later without a chassis change. HP specifies a dedicated 200β240 V AC circuit with a C20 inlet for the 1,600 W-class DGX Station platform; it is not intended for a standard 120 V outlet. NVIDIA-approved partner-developed BIOS and BMC firmware provide secure remote management for continuous 24/7 operation.
Ubuntu 24.04 LTS, NVIDIA AI Developer Tools, HP Z Runtime and HP Z Toolkit
The ZGX Fury ships with Ubuntu 24.04 LTS and the NVIDIA AI Developer Tools β drivers, CUDA and container tooling β preinstalled, the same NVIDIA software foundation used across the DGX family, so TensorRT-LLM, NVIDIA NeMo, NIM microservices, Triton, RAPIDS and NGC containers run natively alongside PyTorch, JAX, vLLM and Hugging Face Transformers. HP adds HP Z Runtime, a preinstalled command-line tool for pulling, serving and managing models, and the HP Z Toolkit, a free Visual Studio Code extension for an x86 Windows 11 or Ubuntu 24.04 client that handles discovery, pairing, remote access and SSH management. HP states the system is certified for Red Hat Enterprise Linux and that Windows support is coming; the units SourceIT supplies today ship with Ubuntu. NVIDIA AI Enterprise is available as a separate subscription.
Available in Singapore Through SourceIT
SourceIT is an authorised IT reseller in Singapore and supplies the HP ZGX Fury AI Station G1n with a full GST invoice, HP's 3-year onsite warranty and enterprise procurement assistance β email sales@sourceit.com.sg or use the Request for Quotation button for configuration options, lead times and pricing. For desktop-scale AI, see the NVIDIA DGX Spark or the HP ZGX Nano G1n; for alternative GB300 stations, see the MSI XpertStation WS300, ASUS ExpertCenter Pro ET900N G3, GIGABYTE W775-V10-L01 and Supermicro ARS-511GD-NB-LCC in the NVIDIA DGX Spark & DGX Station collection.
HP ZGX Fury AI Station G1n Datasheet β Technical Specifications
General
| Product Name | HP ZGX Fury AI Station G1n |
| Part Numbers (Singapore) | E6ZS0PT#AB4 (RTX PRO 2000 Blackwell, 4 Γ 2 TB); E6ZS5PT#AB4 (RTX PRO 4000 Blackwell SFF, 2 Γ 2 TB + 2 Γ 4 TB); no-discrete-GPU SED configuration to order |
| Category | Enterprise AI workstation / deskside AI supercomputer (NVIDIA DGX Station GB300) |
| Form Factor | Liquid-cooled rackable tower; 5U rack with included rails |
| Warranty | HP 3-year (3/3/3) limited warranty β 3 years parts, labour and onsite repair |
| Local Warranty (Singapore) | 3 Years HP onsite; procurement and GST invoice by SourceIT |
Processor & GPU
| Compute Platform | NVIDIA DGX Station architecture β NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip |
| CPU | 1 Γ NVIDIA Grace CPU, 72 Arm Neoverse V2 cores, soldered |
| GPU | 1 Γ NVIDIA Blackwell Ultra GPU |
| CPUβGPU Interconnect | NVIDIA NVLink-C2C, 900 GB/s (per NVIDIA DGX Station datasheet) |
| GPU Partitioning | Multi-Instance GPU (MIG) β up to 7 instances |
| Professional Display Graphics | E6ZS0PT: NVIDIA RTX PRO 2000 Blackwell (16 GB GDDR7); E6ZS5PT: NVIDIA RTX PRO 4000 Blackwell SFF (24 GB GDDR7); headless configuration: none |
AI Performance (per NVIDIA DGX Station datasheet)
| FP4 Tensor Core | Up to 20 PFLOPS (with sparsity) |
| FP8 / FP6 Tensor Core | 10 PFLOPS |
| FP16 / BF16 Tensor Core | 5 PFLOPS |
| TF32 Tensor Core / FP32 / FP64 | 2.5 PFLOPS / 80 TFLOPS / 1.3 TFLOPS |
| Model Support | AI models up to 1 trillion parameters (FP4), per NVIDIA and HP |
| Decoders | 7 Γ NVDEC, 7 Γ nvJPEG |
Memory
| Total Coherent Memory | 748 GB unified CPU + GPU memory space |
| GPU Memory | 252 GB HBM3e, 7.1 TB/s |
| CPU Memory | 496 GB LPDDR5X-6400, 396 GB/s |
| CPU Memory Modules | 4 Γ 128 GB SOCAMM LPDDR5X (onboard) |
Storage
| OS Storage | 2 Γ embedded M.2 PCIe Gen5 (from CPU), software RAID 1 through host OS |
| Data Storage | 2 Γ embedded M.2 PCIe Gen6 (from ConnectX-8), software RAID 0 through host OS |
| Installed Drives | E6ZS0PT: 4 Γ 2 TB HP Z Turbo Drive PCIe Gen5 M.2 NVMe (8 TB); E6ZS5PT: 2 Γ 2 TB + 2 Γ 4 TB (12 TB); headless configuration: 4 Γ 2 TB self-encrypting NVMe |
| Encryption | SED capable |
Networking
| High-Speed Networking | 2 Γ QSFP112, 400 Gb/s per port, NVIDIA ConnectX-8 SuperNIC (up to 800 Gb/s, Ethernet) |
| General Ethernet | 1 Γ RJ45 10 GbE |
| Management LAN | 1 Γ RJ45 1 GbE (BMC, system management only) |
I/O, Expansion & Display
| Front I/O | 2 Γ USB Type-C 10 Gbps, 2 Γ USB Type-A 5 Gbps, 1 Γ universal audio jack |
| Rear I/O | 4 Γ USB Type-A 10 Gbps, microphone, audio-in, line-out; 1 Γ micro USB (BMC); 1 Γ Mini DisplayPort 2.0 (BMC systems management only); Kensington lock slot |
| PCIe Slots | 3 β 1 Γ PCIe Gen5 x16; 2 Γ PCIe Gen5 x16 (x8 electrical) |
Software, Management & Security
| Operating System | Ubuntu 24.04 LTS with NVIDIA AI Developer Tools (arm64); Red Hat Enterprise Linux certified per HP |
| HP Software | HP Z Runtime (model CLI, preinstalled); HP Z Toolkit (VS Code extension for Windows 11 / Ubuntu 24.04 x86 clients) |
| NVIDIA AI Stack | CUDA, cuDNN, TensorRT / TensorRT-LLM, NVIDIA NIM, NVIDIA NeMo, Triton, RAPIDS, NGC containers; NVIDIA AI Enterprise by separate subscription |
| BIOS / BMC | NVIDIA-approved partner-developed BIOS and BMC firmware; secure remote management for continuous 24/7 operation |
Power, Cooling & Physical
| Cooling | Liquid cooled (integrated in chassis), front intake and rear exhaust |
| Power | 1,600 W total system power (NVIDIA DGX Station platform); dedicated 200β240 V AC circuit with C20 inlet required per HP, not for 120 V outlets |
| Dimensions (H Γ W Γ D) | 44.6 Γ 22.15 Γ 66.7 cm, standard tower orientation; rack rails included for 5U rack mounting |
| Weight | Starting at 41 kg |
Compatibility & Software Support
NVIDIA AI Software Stack
Ships with Ubuntu 24.04 LTS and the NVIDIA AI Developer Tools preinstalled β the same NVIDIA software foundation used across the DGX family. CUDA, TensorRT-LLM, NVIDIA NeMo, NIM microservices, Triton, RAPIDS and NGC containers run natively on the Grace Blackwell Ultra platform, alongside PyTorch, JAX, vLLM, Hugging Face Transformers and ONNX Runtime; Docker and Kubernetes cover containerised MLOps pipelines, and HP's Z Runtime CLI pulls, serves and manages models out of the box.
Arm64 Architecture β Plan Your Software Accordingly
The Grace CPU is an Arm Neoverse V2 processor, so the operating system, drivers and container images are arm64 (aarch64) builds. Mainstream AI frameworks and NVIDIA NGC containers are published for arm64, but proprietary x86-only applications will not run natively β check any commercial software you rely on before ordering. HP states that Windows support for the ZGX Fury is coming soon and NVIDIA has announced DGX Station for Windows; the units SourceIT supplies today ship with Ubuntu, and HP lists the system as Red Hat Enterprise Linux certified.
Clustering and Networking Compatibility
The two NVIDIA ConnectX-8 SuperNIC QSFP112 ports run at 400 Gb/s each (up to 800 Gb/s aggregate), so two systems can be linked for more than 1.4 TB of coherent memory and up to 40 PFLOPS, or connected to an NVMe-over-fabric storage array. QSFP112 transceivers or direct-attach cables are ordered separately to match your fabric β SourceIT quotes compatible optics or DACs with the system; the HP ZGX 0.4 m QSFP112 Ethernet DAC cable (C20MQAA) is HP's QSFP112 cable for the ZGX family. The 10 GbE RJ45 port covers site networking, and the dedicated management LAN keeps BMC traffic isolated from production data.
HP Z Ecosystem
The ZGX Fury G1n ships with HP Z Runtime and HP Z Toolkit and includes rack rails so a deployment can grow from deskside to a shared compute room. The optional RTX PRO 2000 / 4000 Blackwell display GPUs suit local visualisation, CAD and DCC alongside AI work; the headless SED configuration suits remote-accessed and regulated environments. The same HP Z Toolkit also manages the GB10-based HP ZGX Nano G1n, so a team can prototype on the Nano and scale to the Fury inside one toolchain.
Package Contents & Ordering Information
What Ships With the System
- HP ZGX Fury AI Station G1n with NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip
- NVMe storage and display GPU as per the selected configuration (E6ZS0PT#AB4 or E6ZS5PT#AB4; headless SED build to order)
- Rack rails for 5U rack mounting, power cord and HP documentation
- HP 3-year (3/3/3) limited warranty with onsite repair
QSFP112 transceivers or DAC cables β for example the HP ZGX 0.4 m QSFP112 Ethernet DAC cable for linking two units β are quoted separately.
Which Configuration?
- E6ZS0PT#AB4 β RTX PRO 2000 Blackwell / 4 Γ 2 TB (this listing, first dropdown option) β balanced build with professional display graphics for local visualisation
- E6ZS5PT#AB4 β RTX PRO 4000 Blackwell SFF / 2 Γ 2 TB + 2 Γ 4 TB (this listing, second dropdown option) β more on-board dataset capacity (12 TB) and the stronger visualisation GPU
- Headless SED β no graphics card / 4 Γ 2 TB SED (this listing, third dropdown option) β rack-adjacent or remote-accessed deployments, or where self-encrypting drives are mandated by security or compliance policy; part number confirmed at quotation
Other NVIDIA GB10 and DGX Station Systems at SourceIT
- MSI XpertStation WS300 β NVIDIA DGX Station GB300 platform, same 748 GB / 20 PFLOPS core compute
- ASUS ExpertCenter Pro ET900N G3 β NVIDIA DGX Station GB300 platform
- GIGABYTE W775-V10-L01 β NVIDIA DGX Station GB300 platform
- Supermicro Super AI Station ARS-511GD-NB-LCC β NVIDIA DGX Station GB300 platform
- HP ZGX Nano G1n 4TB and NVIDIA DGX Spark 4TB β desktop-scale GB10 systems (128 GB unified memory, 1 PFLOP FP4); also from ASUS, Dell, Lenovo, GIGABYTE, MSI, Acer and xFusion
Browse the NVIDIA DGX Spark & DGX Station collection and the AI workstations and AI servers collection. For configuration, research-grant or education enquiries, GST invoicing and lead times, email sales@sourceit.com.sg quoting E6ZS0PT or E6ZS5PT.
Datasheet, Downloads & Setup Guide
Official HP and NVIDIA documentation:
- HP ZGX Fury AI Station β official product overview
- HP ZGX Fury G1n E6ZS0PT β HP Singapore product page and specifications
- HP ZGX Fury G1n E6ZS5PT β HP Singapore product page and specifications
- HP ZGX Fury AI Station G1n QuickSpecs (PDF)
- HP ZGX Fury AI Station G1n datasheet (PDF)
- NVIDIA DGX Station (GB300) β official platform page (Singapore)
- NVIDIA DGX Station datasheet (PDF)
Frequently Asked Questions β HP ZGX Fury AI Station G1n
What is the HP ZGX Fury AI Station G1n?
HP's build of the NVIDIA DGX Station GB300 platform: a liquid-cooled rackable tower with the GB300 Grace Blackwell Ultra Desktop Superchip, 748 GB coherent memory and up to 20 PFLOPS FP4 for training, fine-tuning and serving AI models of up to 1 trillion parameters on premises, backed by HP's 3-year onsite warranty.
What is the difference between the ZGX Fury G1n configurations?
Only the display graphics card and the storage. E6ZS0PT#AB4 pairs an NVIDIA RTX PRO 2000 Blackwell with 4 Γ 2 TB NVMe; E6ZS5PT#AB4 steps up to an RTX PRO 4000 Blackwell SFF with 2 Γ 2 TB + 2 Γ 4 TB NVMe; the headless configuration has no discrete graphics card and 4 Γ 2 TB self-encrypting drives. The GB300 Superchip, 748 GB coherent memory, 72-core Grace CPU, ConnectX-8 networking and 3/3/3 warranty are identical across all three.
Why would I choose the headless SED configuration?
If the system is accessed remotely or sits in a shared compute room, a display GPU adds cost without value β and the self-encrypting drives give hardware-level encryption at rest, often required for regulated, government and financial-sector workloads in Singapore.
How does the ZGX Fury G1n compare to the NVIDIA DGX Spark and HP ZGX Nano?
Different class entirely: the ZGX Fury delivers up to 20 PFLOPS FP4 and 748 GB coherent memory for models up to 1 trillion parameters, versus the GB10-based DGX Spark and ZGX Nano at 1 PFLOP and 128 GB (about 200 billion parameters). Choose the Nano or Spark for individual developers, the ZGX Fury for teams and production AI.
HP ZGX Fury G1n vs MSI XpertStation WS300, ASUS ET900N G3, GIGABYTE W775 or Supermicro β which should I choose?
All are NVIDIA DGX Station GB300 systems with identical core compute. The ZGX Fury adds HP's published 3-year onsite warranty, a rackable liquid-cooled chassis with rails included, factory display-GPU options, up to 12 TB of HP Z Turbo Drive NVMe and the HP Z Runtime / Z Toolkit software. Talk to us about which fits your deployment and support expectations.
What AI models can it run?
HP cites development and fine-tuning of 100-billion-class models and inference on trillion-parameter-class models from the 748 GB coherent memory, which accommodates open-weight families such as Llama, Qwen, DeepSeek, Gemma, Mistral and gpt-oss at full scale.
Can multiple team members share one system?
Yes β Multi-Instance GPU (MIG) partitions the Blackwell Ultra GPU into up to seven isolated instances for concurrent users and workloads.
Can two ZGX Fury systems be linked?
Yes. The ConnectX-8 SuperNIC's two 400 Gb/s QSFP112 ports link two DGX Station-class systems for more than 1.4 TB of coherent memory and up to 40 PFLOPS, per NVIDIA. SourceIT quotes the cabling with the systems.
Does it run Windows?
Not today. The ZGX Fury ships with Ubuntu 24.04 LTS on an arm64 Grace CPU; HP states Windows support is coming soon and NVIDIA has announced DGX Station for Windows. Proprietary x86-only applications will not run natively.
Does it need a server room?
No rack is required β the ZGX Fury is liquid cooled inside its own tower chassis for 24/7 operation, and it ships with rails for a 5U rack position later. Plan a dedicated 200β240 V AC circuit with a C20 inlet, as HP specifies.
What warranty does HP provide in Singapore?
HP's 3-year (3/3/3) limited warranty β 3 years of parts, labour and onsite repair β as stated on HP Singapore's product pages for E6ZS0PT and E6ZS5PT.
How do I purchase the HP ZGX Fury G1n in Singapore?
SourceIT is an authorised IT reseller in Singapore and supplies the ZGX Fury G1n with GST invoice, HP onsite warranty and enterprise procurement assistance β contact sales@sourceit.com.sg or use the Request for Quotation button for configuration options, lead times and pricing.
