ASUS Ascent GX10 Ultra-Small AI Supercomputer NVIDIA GB10 - 64GB / 1TB
PRODUCT DESCRIPTION
ASUS Ascent GX10 Ultra-Small AI Supercomputer, NVIDIA GB10 Grace Blackwell Superchip, 20-Core Arm CPU, 64GB Unified Memory, 1TB NVMe PCIe 4.0 SSD, ConnectX-7 200GbE, Wi-Fi 7 (GX10-GG0058BN) - 1 Year Onsite Warranty
ASUS Ascent GX10 64GB / 1TB — NVIDIA GB10 Petaflop AI Supercomputer in a 1.48 kg Desktop
The ASUS Ascent GX10 (GX10-GG0058BN, ASUS part number 90MS0371-M001U0) is ASUS's build of the NVIDIA DGX Spark platform: an ultra-small AI supercomputer powered by the NVIDIA GB10 Grace Blackwell Superchip, pairing a 20-core Arm CPU with an integrated NVIDIA Blackwell GPU and 64 GB of coherent unified LPDDR5X-8533 memory. The GB10 platform delivers up to 1 petaFLOP of FP4 AI performance for local LLM inference, fine-tuning and agentic AI on a single unit — on a desk, from a wall socket, with no cloud GPU bill and no data leaving your premises. This configuration fits a 1TB M.2 NVMe PCIe 4.0 SSD inside ASUS's Stellar Grey metal chassis.
For Singapore developers, students, research labs and enterprise AI teams, the 64 GB model is the lowest-memory GX10 — the same GB10 compute, ConnectX-7 networking and NVIDIA DGX software stack as the 128 GB models, with half the unified memory. SourceIT, an authorised ASUS and IT reseller in Singapore, supplies it with a full GST invoice, ASUS onsite warranty and enterprise procurement support.
NVIDIA GB10 Grace Blackwell Superchip — 1 PetaFLOP FP4, 20 Arm Cores, 5th-Gen Tensor Cores
The GB10 Superchip places a 20-core Arm CPU (10 Cortex-X925 performance cores plus 10 Cortex-A725 efficiency cores) and an NVIDIA Blackwell GPU on one package, joined by NVLink-C2C so both share a single memory space instead of shuttling data across a PCIe bus. Fifth-generation Tensor Cores with native FP4 precision deliver up to 1 PFLOP of sparse AI throughput for LLM inference and fine-tuning; fourth-generation RT Cores handle rendering, simulation and synthetic-data generation; and NVIDIA DLSS 4 is supported for AI-accelerated graphics. ASUS rates the chassis to MIL-STD 810H testing, and the toolless design opens without a screwdriver.
64 GB Unified Memory — the Same GB10 Platform at a Lower Memory Tier
The GX10-GG0058BN carries 64 GB of LPDDR5X-8533 unified memory, one coherent pool shared by CPU and GPU — ASUS offers the GX10 with 64 GB or 128 GB. Compute (GB10 Superchip, 1 PFLOP FP4), ConnectX-7 networking, I/O and chassis are the same as the 128 GB models; the difference is the size of model that fits resident in memory. NVIDIA's 200-billion-parameter inference and 70-billion-parameter fine-tuning figures for the platform refer to the 128 GB configuration, so choose a 128 GB GX10 if you plan to work with models at that scale on one unit. The 64 GB model suits developers prototyping with small and mid-size open models, RAG pipelines, computer vision and agent workloads who want the NVIDIA DGX software stack at a lower entry price.
1TB NVMe PCIe 4.0 SSD, ConnectX-7 200 GbE SmartNIC, 10G Ethernet, Wi-Fi 7
This model ships with a 1 TB M.2 NVMe PCIe 4.0 SSD in the single M.2 slot. Networking is what separates the GX10 from an ordinary mini PC: an integrated NVIDIA ConnectX-7 SmartNIC provides 200 GbE connectivity with RDMA for linking units together and moving datasets at line rate, plus hardware acceleration for TLS, IPsec and MACsec and IEEE 1588v2 PTP time synchronisation. A 10G Ethernet RJ-45 port handles the office network — the right port for pulling model weights — while Wi-Fi 7 (2x2) and Bluetooth 5.4 LE cover wireless.
Three USB 3.2 Gen 2x2 Type-C Ports with DisplayPort 2.1, HDMI 2.1a, USB-C PD 3.1 EPR Power
Three rear USB 3.2 Gen 2x2 Type-C ports run at 20 Gbps and each carries DisplayPort 2.1 Alt Mode, so one cable can drive a monitor and peripherals; a dedicated HDMI 2.1a port adds a fourth display output. A fourth Type-C port is the power input, taking USB PD 3.1 EPR from the supplied 240 W adapter, and a Kensington lock slot secures the unit on a shared desk or lab bench.
QuietFlow Cooling — Three Fans and Dual Vapor Chambers for Sustained AI Workloads
ASUS's QuietFlow thermal design combines three fans with dual vapor chambers and claims 1.6× more efficient thermal coverage than comparable compact systems, so the GB10 holds full performance through hours-long fine-tuning runs and always-on agent workloads without the noise of a rack GPU server. The whole unit measures 150 × 150 × 51 mm and weighs 1.48 kg — smaller than most desktop power supplies.
NVIDIA DGX OS, NemoClaw and the NVIDIA AI Software Stack — Built for Local Agentic AI
The GX10 boots into NVIDIA DGX OS, an Ubuntu-based Linux build with the NVIDIA AI software stack preinstalled: CUDA, TensorRT, NVIDIA NIM microservices and Blueprints, PyTorch, TensorFlow, Jupyter and Ollama. ASUS positions it for agentic AI without perpetual token costs — NVIDIA NemoClaw wraps OpenClaw agents with OpenShell policy-based privacy and security guardrails and runs open models such as Nemotron locally; Hermes Agent and other open-source agent frameworks are supported; NVIDIA Sync manages multi-unit clusters; and NVIDIA PAIR routes inference across RTX GPUs, Macs and Windows PCs through a single local endpoint. Because it is the same software foundation as DGX Spark, DGX Station and DGX Cloud, models prototyped here migrate to larger NVIDIA infrastructure with minimal code changes.
Available in Singapore Through SourceIT
SourceIT is an authorised ASUS reseller in Singapore and supplies the ASUS Ascent GX10 64GB / 1TB with a full GST invoice, ASUS 1-year onsite warranty, and enterprise, research-grant and education procurement assistance. The GX10 is also available with 128 GB of memory in 1TB, 2TB and 4TB on the same GB10 platform. Browse the ASUS Ascent GX10 collection and every GB10 and GB300 system we stock in the NVIDIA DGX Spark & DGX Station collection.
ASUS Ascent GX10 64GB / 1TB (GX10-GG0058BN) Datasheet — Technical Specifications
General
| Product Name | ASUS Ascent GX10 Ultra-Small AI Supercomputer — 64GB / 1TB |
| SKU / Part Number | GX10-GG0058BN / 90MS0371-M001U0 |
| Category | Personal AI supercomputer / AI developer desktop (NVIDIA DGX Spark GB10 platform) |
| Form Factor | Ultra-compact desktop, 150 × 150 mm footprint; toolless chassis |
| Colour / Finish | Stellar Grey metal chassis |
| Durability | MIL-STD 810H multiple tests passed, per ASUS |
| Manufacturer Warranty | ASUS 1-Year Onsite Warranty |
| Local Warranty (Singapore) | Supplied by SourceIT with GST invoice and local support |
Processor & GPU — NVIDIA GB10 Grace Blackwell Superchip
| Architecture | NVIDIA Grace Blackwell — CPU + GPU on one superchip, NVLink-C2C interconnect |
| CPU | 20-core Arm: 10 × Cortex-X925 performance cores + 10 × Cortex-A725 efficiency cores |
| GPU | NVIDIA Blackwell architecture GPU (integrated) |
| Tensor Cores | 5th generation, FP4 / FP8 |
| RT Cores | 4th generation |
| AI Performance | Up to 1 PFLOP (1,000 TOPS) FP4 with sparsity |
| Model Support | Model size is bounded by the 64 GB unified memory; NVIDIA's 200B-inference / 70B-fine-tuning platform figures apply to 128 GB units |
| Multi-Unit Scaling | Units link over the ConnectX-7 200 GbE SmartNIC |
| Graphics Features | NVIDIA DLSS 4 |
Memory
| System Memory | 64 GB LPDDR5X-8533 coherent unified system memory, shared by CPU and GPU |
| Upgradeability | Soldered — not user-upgradeable |
Storage
| SSD | 1 TB M.2 NVMe PCIe 4.0 SSD |
| Slots | 1 × M.2 NVMe, up to 4 TB |
| Range | 1TB (PCIe 4.0), 2TB (PCIe 4.0), 4TB (PCIe 5.0) |
Networking
| High-Speed NIC | 1 × NVIDIA ConnectX-7 SmartNIC, 200 GbE, RDMA; TLS / IPsec / MACsec hardware acceleration; IEEE 1588v2 PTP |
| Ethernet | 1 × 10G LAN RJ-45 |
| Wi-Fi | Wi-Fi 7, 2x2 |
| Bluetooth | Bluetooth 5.4 with LE |
I/O Ports
| USB (data / display) | 3 × USB 3.2 Gen 2x2 Type-C, 20 Gbps, DisplayPort 2.1 Alt Mode on each |
| USB (power in) | 1 × USB 3.2 Gen 2x2 Type-C with Power Delivery input, USB PD 3.1 EPR |
| Video | 1 × HDMI 2.1a; up to 3 × DisplayPort 2.1 over USB-C |
| Front Panel | Power button |
| Security | Kensington lock slot |
Software
| Operating System | NVIDIA DGX OS (Ubuntu Linux based), arm64 |
| AI Stack (preinstalled) | NVIDIA AI software stack — CUDA, TensorRT, NVIDIA NIM and Blueprints, PyTorch, TensorFlow, Jupyter, Ollama |
| Agentic AI | NVIDIA NemoClaw + OpenShell; compatible with OpenClaw and Hermes Agent; NVIDIA PAIR; NVIDIA Sync cluster management |
Cooling, Power, Physical & Compliance
| Cooling | ASUS QuietFlow — three fans and dual vapor chambers; 1.6× thermal coverage vs comparable compact systems, per ASUS |
| Power Adapter | 240 W AC adapter with power cord (USB-C PD 3.1 EPR input) |
| Dimensions (W × D × H) | 150 × 150 × 51 mm (5.91 × 5.91 × 2.01 in) |
| Weight | 1.48 kg (3.26 lb) |
| Certifications | BSMI, CB, CE, FCC, UL, CCC, C-Tick, Wi-Fi, RF, VCCI |
| In the Box | GX10 unit, AC adapter, power cord, user manual, warranty card |
Compatibility, Clustering & Software Support
What Runs on the Ascent GX10
Anything in the NVIDIA and open-source AI ecosystem published for arm64: Hugging Face Transformers and Diffusers, vLLM, Ollama, llama.cpp, LangChain and LlamaIndex RAG pipelines, NVIDIA NIM microservices and Blueprints, NeMo for fine-tuning and guardrails, TensorRT-LLM for optimised serving, ComfyUI and FLUX-class image generation, RAPIDS for GPU data science, and the NVIDIA Isaac, Metropolis and Holoscan frameworks for robotics, vision and edge work. NVIDIA's DGX Spark Playbooks apply directly to the GX10 because it is the same GB10 platform.
Arm64 Linux — Plan Your Software Accordingly
The GB10's CPU is Arm-based and the GX10 ships with NVIDIA DGX OS, so drivers and containers are arm64 (aarch64) builds. Every major AI framework and NVIDIA NGC container is published for arm64, but x86-only proprietary desktop software will not run natively and Windows is not supported. Most teams drive the GX10 from a laptop over SSH, Jupyter or VS Code Remote and keep Windows applications there. For a Windows 11 Arm desktop in the same family, see the ASUS Ascent QN10.
Linking Two or Four GX10 Units
Two GX10 units connect back-to-back over their ConnectX-7 ports with a single QSFP direct-attach cable — SourceIT stocks the ASUS QSFP cable for Ascent GX10, 40 cm (ASU-90MA0000-P01000) — and ASUS's published netplan configuration brings the pair up as one system (256 GB with two 128 GB units). Scaling to four units needs an external 200 GbE switch and uses NVIDIA Sync for cluster management. The 10G RJ-45 port stays free for ordinary site networking throughout.
Where the GX10 Sits in the GB10 Family
The Ascent GX10 is one of eight GB10 desktops SourceIT stocks alongside the NVIDIA DGX Spark itself: Dell Pro Max with GB10, Lenovo ThinkStation PGX, HP ZGX Nano, GIGABYTE AI TOP ATOM, MSI EdgeXpert, Acer Altos BrainSphere GB10 and xFusion FusionXpark GB10. All share the GB10 Superchip and 1 PFLOP FP4; the GX10 is distinguished by ASUS's QuietFlow cooling, MIL-STD 810H toolless chassis, 1TB, 2TB and 4TB storage tiers, a choice of 64 GB or 128 GB unified memory and ASUS onsite warranty.
Package Contents & Ordering Information
What Ships in the Box
- ASUS Ascent GX10 (GX10-GG0058BN) with NVIDIA GB10 Grace Blackwell Superchip, 64 GB unified memory and 1 TB M.2 NVMe PCIe 4.0 SSD
- 240 W AC adapter and power cord
- NVIDIA DGX OS preinstalled with the NVIDIA AI software stack
- User manual and warranty card
Monitor, keyboard and mouse are not included. The QSFP cable for linking two units is sold separately — see the ASUS QSFP cable for Ascent GX10 (40 cm).
ASUS Ascent GX10 Range at SourceIT Singapore
- ASUS Ascent GX10 64GB / 1TB — GX10-GG0058BN / 90MS0371-M001U0 (this listing), 64 GB unified memory, PCIe 4.0 NVMe, power cord included
- ASUS Ascent GX10 128GB / 1TB — GX10-GG0010BN / 90MS0371-M000A0, PCIe 4.0 NVMe, AC power cord not included
- ASUS Ascent GX10 128GB / 1TB — GX10-GG0007BN / 90MS0371-M00070, PCIe 4.0 NVMe, power cord included
- ASUS Ascent GX10 128GB / 1TB JP-variant SKU — GX10-GG0006BN / 90MS0371-M00060, PCIe 4.0 NVMe, AC power cord not included
- ASUS Ascent GX10 128GB / 2TB — GX10-GG0030BN / 90MS0371-M000Y0, PCIe 4.0 NVMe, power cord included
- ASUS Ascent GX10 128GB / 4TB — GX10-GG0031BN / 90MS0371-M000Z0, PCIe 5.0 NVMe, power cord included
- ASUS QSFP Cable for Ascent GX10, 40 cm — ASU-90MA0000-P01000 (links two units)
Every GX10 in the range uses the same NVIDIA GB10 Superchip, ConnectX-7 networking and chassis. Memory is 128 GB on every model except GX10-GG0058BN (64 GB); the 4TB GX10-GG0031BN is the only model on a PCIe 5.0 SSD; and GX10-GG0010BN and GX10-GG0006BN are supplied without the AC power cord.
Other NVIDIA GB10 and DGX Station Systems at SourceIT
- NVIDIA DGX Spark 4TB — NVIDIA's own first-party GB10 unit
- Dell Pro Max with GB10 — 2TB / 4TB, 3-year onsite warranty
- Lenovo ThinkStation PGX — 1TB / 4TB
- HP ZGX Nano G1n — 4TB
- GIGABYTE AI TOP ATOM — 1TB / 4TB
- MSI EdgeXpert MS-C931 — 4TB
- Acer Altos BrainSphere GB10 F1 — 4TB
- xFusion FusionXpark GB10 — 1TB / 2TB / 4TB with technical support
- ASUS ExpertCenter Pro ET900N G3 (NVIDIA DGX Station GB300) — step up to 748 GB coherent memory and 20 PFLOPS for trillion-parameter models
Browse the ASUS Ascent GX10 collection and the full NVIDIA DGX Spark & DGX Station collection. For volume, project, research-grant or education pricing, GST invoicing and current lead times, email sales@sourceit.com.sg quoting GX10-GG0058BN.
Datasheet, Downloads & Setup Guide
Official ASUS and NVIDIA documentation for the Ascent GX10 and the GB10 platform:
- ASUS Ascent GX10 — product page (Singapore)
- ASUS Ascent GX10 — full technical specifications
- ASUS Ascent GX10 — official datasheet (PDF)
- NVIDIA DGX Spark (GB10) — official platform page (Singapore)
- NVIDIA DGX Spark / DGX OS user documentation
- NVIDIA DGX Spark Playbooks — step-by-step recipes for LLMs, fine-tuning, RAG and agents
- NVIDIA DGX Spark / GB10 developer forum
- NVIDIA DGX Spark Setup and Enterprise Deployment Guide — SourceIT (applies to the GX10)
For the ASUS user manual, email sales@sourceit.com.sg quoting GX10-GG0058BN.
Frequently Asked Questions — ASUS Ascent GX10 64GB / 1TB
What is the ASUS Ascent GX10?
ASUS's ultra-small AI supercomputer built on the NVIDIA DGX Spark GB10 platform: a 150 × 150 mm, 1.48 kg desktop with the NVIDIA GB10 Grace Blackwell Superchip, up to 1 PFLOP of FP4 AI performance and, on this model, 64 GB of unified memory, running NVIDIA DGX OS for local LLM inference, fine-tuning, data science and agentic AI.
What is the difference between the ASUS Ascent GX10 and the NVIDIA DGX Spark?
Same NVIDIA GB10 Superchip, 1 PFLOP FP4 and ConnectX-7 networking. This GX10 model carries 64 GB of unified memory, half the 128 GB of the DGX Spark and the 128 GB GX10 models, in ASUS's Stellar Grey MIL-STD 810H toolless chassis with QuietFlow cooling and ASUS onsite warranty; the DGX Spark is NVIDIA's first-party unit in 4TB only.
Which storage tier should I choose — 1TB, 2TB or 4TB?
Across the 128 GB models, compute, memory and networking are identical; only storage differs. Choose by how much you need resident locally: a 70B-parameter model at 4-bit is roughly 40 GB and at 16-bit roughly 140 GB, so 1TB holds a working library of quantised models, 2TB adds room for several full-precision checkpoints and RAG indexes, and 4TB keeps an entire model zoo plus datasets on one unit — and is the only tier on a PCIe 5.0 drive.
What AI models can the GX10 run?
Any model that fits in its 64 GB of unified memory, using the same NVIDIA AI software stack, frameworks and DGX Spark Playbooks as the 128 GB models. NVIDIA's figures of up to 200 billion parameters for inference and 70 billion for fine-tuning refer to the 128 GB configuration — for models at that scale, choose a 128 GB GX10.
Can I link two GX10 units together?
Yes, back-to-back over the ConnectX-7 200 GbE ports with the ASUS 40 cm QSFP cable and ASUS's published netplan configuration. Four units need an external 200 GbE switch and NVIDIA Sync.
Does the GX10 run Windows or x86 software?
No. It ships with NVIDIA DGX OS (Ubuntu-based, arm64) and is a Linux AI development system. Mainstream AI frameworks and NVIDIA NGC containers are published for arm64; proprietary x86-only desktop applications will not run natively.
What is NemoClaw and does the GX10 support it?
NemoClaw is NVIDIA's open-source reference stack that adds OpenShell policy-based security and privacy guardrails around OpenClaw agents and runs open models such as Nemotron locally. ASUS lists NemoClaw, OpenClaw and Hermes Agent compatibility for the GX10, which is why it positions the unit for 24/7 agentic AI without per-token cloud costs.
Does it need special power or cooling?
No. It runs from a standard socket through the supplied 240 W USB-C PD 3.1 adapter, and ASUS's QuietFlow three-fan, dual-vapor-chamber cooling is designed for sustained AI load on a desk rather than in a server room.
Can I upgrade the memory or storage?
Memory is soldered and fixed at 64 GB. The single M.2 NVMe slot supports up to 4 TB; the chassis is toolless, but choose the tier you need at purchase so the ASUS warranty is unaffected.
Who is the ASUS Ascent GX10 for?
AI developers and ML engineers prototyping and fine-tuning before scaling to DGX Cloud; university labs and research groups (ASUS runs higher-education and NVIDIA Inception offers); data scientists running large-memory analytics; robotics and computer-vision teams; and Singapore enterprises in finance, healthcare, legal and government that must keep models and data on-premise.
What warranty does the GX10 come with in Singapore?
ASUS backs this GX10 model with a 1-year onsite warranty. Every unit from SourceIT comes with a GST invoice and local support.
How do I buy the ASUS Ascent GX10 in Singapore?
Order on this page or email sales@sourceit.com.sg quoting GX10-GG0058BN. SourceIT is an authorised ASUS reseller in Singapore and supplies the GX10 with GST invoice, onsite warranty, the QSFP clustering cable and the rest of the NVIDIA GB10 and DGX Station range.
