HP ZGX Nano vs ZGX Fury: GB10 vs GB300 in Singapore

HP sells two AI stations in Singapore built on NVIDIA Grace Blackwell silicon: the HP ZGX Nano G1n, a 150 mm desk unit on the GB10 Superchip with 128 GB of memory and 1 petaFLOP of FP4 compute at S$11,499 list, and the HP ZGX Fury G1n, a liquid-cooled DGX Station tower on the GB300 Superchip with 748 GB of coherent memory and up to 20 petaFLOPS, price on request. They are not two sizes of the same thing; the Fury has roughly six times the memory and twenty times the compute, and needs its own power circuit. This post compares them on memory, PFLOPS, model-size ceiling, power, placement and price, says when to step up, and weighs a cluster of Nanos against one Fury. For the full HP ZGX Nano feature list see our ZGX Nano G1n guide; for the ASUS equivalents, the ASUS Agent Computer guide.

HP ZGX Nano G1n vs HP ZGX Fury G1n: side by side

Memory: 128 GB unified vs 748 GB coherent

Memory sets the model ceiling, and this is the widest gap in the table. The Nano's 128 GB of LPDDR5X is shared between CPU and GPU at 273 GB/s, and HP's own guidance is up to 200B parameters on one unit, which in practice means a 70B model at 8-bit with room for context, or a 120B-class mixture-of-experts model at 4-bit. The Fury's 748 GB is two tiers: 252 GB of HBM3e at 7.1 TB/s on the GPU, enough to hold a 200B model at FP4 entirely in GPU memory, plus 496 GB of LPDDR5X behind NVLink-C2C that lets a 400B-plus model spill into CPU memory and keep running. StorageReview's September 2026 test of the ASUS build of the same GB300 platform served a 433 GB GLM-5.2 checkpoint at 139 tokens per second on one tower; a Nano cannot load that model at all. If your target model is under 100B parameters, the Nano's ceiling is not the problem. If it is a frontier-scale open model, only the Fury runs it on one box.

Compute: 1 PFLOP vs 20 PFLOPS, and what that buys

The twenty-fold FP4 figure is a headline, but the operational difference is concurrency. A GB10 unit serves one developer, or a handful of users, interactively; StorageReview's GB300 testing used 128 concurrent streams, and on GPT-OSS-120B the GB300 platform produced about 9,300 tokens per second, 18 times a DGX Spark-class unit. For a single researcher iterating on prompts and fine-tunes, the Nano's 1 PFLOP is a working tool. For a department-wide endpoint, a RAG service over internal documents with dozens of simultaneous users, or batch generation against a deadline, the Fury's throughput is the difference between a shared service and a queue. HBM bandwidth compounds this: the Fury's 7.1 TB/s against the Nano's 273 GB/s is why the same 70B model answers faster on the Fury even when it fits comfortably on the Nano.

Power and placement: desk vs deskside

The Nano runs from a 240 W USB-C adapter on any desk socket, is rated by HP at 22 dBA idle and 27.6 dBA under load, and weighs from 1.25 kg; StorageReview measured peak draw at about 228 W. The Fury is a 41 kg liquid-cooled tower with a 1,600 W Titanium supply behind a C20 inlet, which in a Singapore office means a dedicated circuit on the 230 V mains confirmed by your electrician, and a two-person lift or a trolley. HP rates the Fury's cooling for office use and includes rails for a 5U rack slot when the office is no longer the right home. Our DGX Station installation checklist covers the circuit, placement and lifting plan; the Nano needs none of it.

Price: S$11,499 list vs a quote per configuration

The HP ZGX Nano G1n 4 TB lists at S$11,499 on our store; promotional pricing may apply, so check the product page or request a quote before deciding. The HP ZGX Fury G1n is price on request, typically quoted per configuration: the RTX PRO 2000 or RTX PRO 4000 display card or none, the drive mix (four 2 TB, or 2 TB plus 4 TB, with a self-encrypting option), transceivers or DAC cables, and the support term. HP has not published a list price, and a Singapore quote depends on allocation. The ZGX Fury G1n guide sets out the three store configurations; the DGX Station GB300 buyer's guide explains how GB300 quoting works across all five OEMs.

When to step up from the ZGX Nano to the ZGX Fury

Step up when any of the following is true. The model you need to serve does not fit in 128 GB even at 4-bit, or fits with no room for a useful context window. More than a few people will hit the endpoint at the same time, and latency under load matters. You are fine-tuning models above the 30B to 70B class rather than running LoRA adapters on small ones. You need a single coherent memory pool rather than a cluster, because your serving stack does not split a model cleanly over two machines. Or you need Red Hat Enterprise Linux certification, a rack-mountable chassis or a Windows configuration, none of which the Nano offers. Stay on the Nano when the work is one developer's prototyping, an edge inference node, a Secure and Regulated air-gapped deployment where the radio-free Nano variant is the point, or a budget set per seat rather than per department. Many Singapore teams buy both: Nanos on developer desks, one Fury as the shared serving node, managed from the same HP ZGX Toolkit extension.

Two or four ZGX Nanos vs one ZGX Fury

HP states that two Nanos linked over their QSFP ports can run models up to 405B parameters, and SourceIT sells the HP 0.4 m QSFP112 DAC cable at S$365 for the link. Two Nanos pool 256 GB and four pool 512 GB at a published list price per unit, where a GB300 quote depends on configuration and allocation, which makes the cluster the right answer for a small group that occasionally needs a 405B model and mostly runs 70B. The trade-offs are real: a clustered model runs over 200 GbE links rather than 7.1 TB/s HBM, so throughput on the large model is a fraction of a single coherent GB300; concurrency across the cluster is still bounded by GB10 compute; and you administer two or four machines. The Fury wins when the large model is the daily workload, when many users share it, or when a single system image matters for compliance. The DGX Station vs DGX Spark post works through the same decision across every GB10 and GB300 unit we sell, and the GX10 clustering guide shows what a two-unit GB10 link involves in practice.

Buying the HP ZGX Nano or ZGX Fury in Singapore

SourceIT stocks the ZGX Nano G1n 4 TB and lists the ZGX Fury G1n in three configurations, with HP's local warranty on both; confirm the term for your SKU when requesting a quote. Use Request for Quotation on the product page, call (65) 6978 3502, or email Sales@sourceit.com.sg. We are GeBIZ and Vendors@Gov registered for government and institutional procurement. If you are deciding between the Nano and NVIDIA's own unit instead, read ZGX Nano vs DGX Spark; the full range is in the AI workstations and servers collection.

Frequently asked questions

What is the difference between the HP ZGX Nano and ZGX Fury? The Nano is a GB10 desk unit with 128 GB of memory and 1 PFLOP of FP4 compute; the Fury is a GB300 DGX Station tower with 748 GB of coherent memory and up to 20 PFLOPS, liquid-cooled and needing a dedicated power circuit.

How much do the HP ZGX Nano and ZGX Fury cost in Singapore? The ZGX Nano G1n 4 TB lists at S$11,499, with promotional pricing that may apply on the product page. The ZGX Fury G1n is price on request, quoted per configuration.

What is the largest model each can run? HP guides up to 200B parameters on one Nano and 405B on two linked; the Fury handles trillion-parameter inference at FP4 and fine-tuning of 100B-class models.

Is a cluster of ZGX Nanos as fast as one ZGX Fury? No. Two Nanos pool 256 GB over 200 GbE links, but throughput on a large model and concurrency remain far below one GB300 with 7.1 TB/s HBM. The cluster wins on price and flexibility.

Does the ZGX Fury need special power in Singapore? Yes, a dedicated circuit for its 1,600 W supply, confirmed by your electrician. The Nano uses a 240 W USB-C adapter on any desk socket.

Can I manage a ZGX Nano and a ZGX Fury together? Yes. HP's ZGX Toolkit discovers and syncs across ZGX systems from a Visual Studio Code client on an x86 laptop running Windows 11 or Ubuntu 24.04.