ASUS Ascent GX10 Compact AI Supercomputer - 1TB
PRODUCT DESCRIPTION
ASUS Ascent GX10 Compact AI Supercomputer - 1TB (GX10-GG0007BN) - 1 Year Onsite Warranty
Petaflop-Class AI Compute in a 1.48 kg Desktop
The ASUS Ascent GX10 (GX10-GG0007BN) is built around the NVIDIA GB10 Grace Blackwell Superchip — a single package pairing a 20-core Arm CPU with an integrated NVIDIA Blackwell GPU and 128 GB of coherent unified memory shared between the two. It is the same GB10 platform behind NVIDIA's own DGX Spark, housed in ASUS's Stellar Grey chassis, and it delivers up to 1 petaFLOP of FP4 tensor performance from a box that sits on a desk rather than in a rack. For a Singapore team evaluating an on-premise AI development workstation — data science, local LLM inference, or a low-noise alternative to booking cloud GPU time — this is the class of hardware that used to need a server room, scaled down to something you plug into a wall socket.
Unified Memory Means the Model Fits, Not Just the Weights
Because CPU and GPU share the same 128 GB LPDDR5x memory pool instead of copying data across a PCIe bus to a separate VRAM pool, the GX10 can hold language models up to roughly 200 billion parameters entirely in memory on a single unit. That is the practical difference between a workstation GPU with 16–24 GB of dedicated VRAM and this architecture: model size is bounded by the unified pool, not by a much smaller discrete graphics buffer, which is what makes local inference and fine-tuning of genuinely large models realistic on a desktop footprint.
1 TB of Local NVMe Storage
1 TB M.2 NVMe PCIe 4.0 SSD.
ConnectX-7 Networking for Two-Unit Clusters
A built-in NVIDIA ConnectX-7 SmartNIC gives the GX10 a dedicated high-speed interconnect, separate from the general-purpose 10G Ethernet port used for normal network access. Two GX10 units connected over ConnectX-7 can pool their memory to work on models beyond the roughly 200-billion-parameter ceiling of a single unit — a path to scale up without moving the workload off-premise. Wi-Fi 7 and Bluetooth 5.4 cover general wireless connectivity for a workstation that is not tied to a fixed desk position.
Quiet Enough for a Desk, Not Just a Data Hall
A triple-fan, dual vapor-chamber cooling system (ASUS calls it QuietFlow) is built to sustain continuous AI workloads without the fan noise typical of rack-mounted GPU servers, which matters when the unit is going to sit next to a developer rather than in a dedicated machine room. At 150 x 150 x 51 mm and 1.48 kg, it occupies less desk space than a typical mini-tower PSU.
Also Available in 2TB and 4TB
The GX10 range at SourceIT covers three storage tiers on the same GB10 Grace Blackwell platform, compute, memory and networking — only the SSD capacity changes. 1TB suits development and single-model work where storage is not the constraint. Talk to us about which tier fits your model library before ordering — the compute and memory are identical across all three, so the decision comes down to how much you need resident locally.
ASUS Ascent GX10 GX10-GG0007BN Datasheet — Technical Specifications
General
| Model | ASUS Ascent GX10 — 1TB |
| SKU / Part Number | GX10-GG0007BN / 90MS0371-M00070 |
| Product Type | Ultra-Compact Desktop AI Supercomputer, NVIDIA GB10 Grace Blackwell platform |
| Operating System | NVIDIA DGX OS (Ubuntu Linux based) |
| Colour | Stellar Grey, metal chassis |
| Manufacturer Warranty | 1-Year Onsite Warranty (ASUS) |
| Local Warranty (Singapore) | Included with every unit supplied by SourceIT, together with a GST invoice |
Processor & AI Accelerator — NVIDIA GB10 Grace Blackwell Superchip
| CPU | 20-core Arm v9.2-A — 10 x Cortex-X925 performance cores + 10 x Cortex-A725 efficiency cores |
| GPU | NVIDIA Blackwell architecture (integrated), 5th-generation Tensor Cores, 4th-generation RT Cores |
| AI Performance | Up to 1 PFLOP FP4 tensor performance, per NVIDIA/ASUS |
| Local Model Support | Up to approximately 200-billion-parameter language models on a single unit |
| Multi-Unit Scaling | Two units linked via ConnectX-7 can address larger models than a single unit supports |
Memory & Storage
| System Memory | 128 GB LPDDR5x coherent unified memory, shared between CPU and GPU |
| Memory Architecture | Single unified pool rather than separate system RAM and discrete GPU VRAM |
| Storage | 1 TB M.2 NVMe PCIe 4.0 SSD. |
Networking
| SmartNIC | NVIDIA ConnectX-7 — one port, used for dual-unit clustering and high-speed data movement |
| Ethernet | 1 x 10 GbE RJ-45 for general network connectivity |
| Wireless | Wi-Fi 7 (2x2), Bluetooth 5.4 with LE |
Ports & I/O
| USB Type-C (data) | 3 x USB 3.2 Gen 2x2 Type-C, DisplayPort 2.1 Alt Mode on each |
| USB Type-C (power) | 1 x USB-C with Power Delivery, USB PD 3.1 EPR |
| Video Output | 1 x HDMI 2.1a, plus DisplayPort 2.1 over the Type-C ports above |
| Security | Kensington lock slot |
Physical & Cooling
| Cooling | QuietFlow — triple-fan, dual vapor-chamber cooling system for sustained AI workloads |
| Dimensions | 150 x 150 x 51 mm |
| Weight | Approximately 1.48 kg |
Power
| Power Input | USB-C PD 3.1 (EPR, up to 180 W negotiated) |
| Power Adapter | 240 W external adapter, supplied in the box |
Software & Supported Frameworks
| Operating System | NVIDIA DGX OS, an Ubuntu Linux-based distribution tuned for the GB10 platform |
| AI / ML Stack | CUDA, TensorRT, NVIDIA NIM microservices, PyTorch, TensorFlow, Jupyter, and the wider NVIDIA AI software stack |
| Typical Workloads | Local LLM inference and fine-tuning, agentic AI development, retrieval-augmented generation (RAG), vector database workloads, deep learning training and experimentation |
Connectivity, Clustering & Setup
What "Clustering" Actually Means Here
The ConnectX-7 SmartNIC is a dedicated high-speed link, separate from the 10 GbE port used for ordinary network access. Connecting two GX10 units over ConnectX-7 lets them pool memory and compute to work on models larger than roughly 200 billion parameters, which a single unit cannot hold in its 128 GB unified pool. A direct-attach copper cable rated for the platform's networking speed is the normal way to make that link — ask us for the correct cable when you order a second unit rather than guessing at compatibility.
Display and Peripheral Connectivity
Video output covers HDMI 2.1a plus DisplayPort 2.1 delivered over any of the three USB 3.2 Gen 2x2 Type-C ports, so a single monitor cable handles both video and, where the monitor supports it, data. The fourth Type-C port is dedicated to power delivery from the 240 W adapter. There is no internal expansion — memory, storage type and networking are fixed at the factory, which is standard for this class of unified-memory AI accelerator rather than a limitation specific to this unit.
Software Setup
The unit ships with NVIDIA DGX OS preinstalled, an Ubuntu Linux-based distribution with CUDA, TensorRT, NVIDIA NIM and the common Python ML frameworks already positioned to install or already present. Standard Linux system administration applies for further software installation — this is a Linux workstation running NVIDIA's own AI stack, not a Windows PC.
Wireless and Wired Networking
Wi-Fi 7 and Bluetooth 5.4 cover general wireless use; the 10 GbE RJ-45 port is the recommended connection for anything involving sustained large file transfer, such as pulling model weights or datasets onto local storage, since it is materially faster than a typical office 1 GbE drop.
Package Contents & Ordering Information
In the Box
ASUS Ascent GX10 AI supercomputer (1TB), 240 W power adapter, power cord, user manual, warranty card.
ASUS Ascent GX10 Range — Ordering Information
| ASUS Ascent GX10 1TB | GX10-GG0007BN / 90MS0371-M00070 (this listing) |
| ASUS Ascent GX10 2TB | GX10-GG0030BN / 90MS0371-M000Y0 |
| ASUS Ascent GX10 4TB | GX10-GG0031BN / 90MS0371-M000Z0 |
All three tiers share the same NVIDIA GB10 Grace Blackwell Superchip, 128 GB unified memory and ConnectX-7 networking; only the SSD capacity and, on the 4TB unit, the NVMe generation, differ. Email sales@sourceit.com.sg quoting the part number to check stock and lead time, or to ask about linking two units for larger models.
A Note on Networking Accessories
A direct-attach copper cable is the normal way to link two GX10 units over ConnectX-7 for dual-unit clustering. Confirm the exact cable part number with us at the time of order, since the correct cable depends on the run length and connector type you need.
Datasheet & Downloads
ASUS hosts the official product and specification pages for the Ascent GX10 platform.
| ASUS Singapore product page | ASUS Ascent GX10 — Overview |
| ASUS Singapore specifications page | ASUS Ascent GX10 — Full Specifications |
For the PDF datasheet, user manual or NVIDIA DGX OS documentation, email sales@sourceit.com.sg quoting GX10-GG0007BN and we will send you the current files directly.
Frequently Asked Questions — ASUS Ascent GX10 GX10-GG0007BN
What is the difference between the ASUS Ascent GX10 and the NVIDIA DGX Spark?
Both run the same NVIDIA GB10 Grace Blackwell Superchip — 128 GB unified memory, up to 1 PFLOP FP4, ConnectX-7 networking — since the GX10 is built on the DGX Spark platform. ASUS offers the GX10 across three storage tiers in its own Stellar Grey chassis.
What AI models can it actually run?
Local inference and fine-tuning of models up to roughly 200 billion parameters on a single unit, covering current open-weight model families in that size class. Linking two units over ConnectX-7 raises the ceiling for models that do not fit in one unit's 128 GB unified pool.
Which storage tier should I choose?
The 1TB unit suits development work and running a small number of models at a time, where the local dataset is not large. If you keep multiple fine-tuned checkpoints, larger datasets or vector database indexes resident on the same machine, the 2TB or 4TB tiers give more headroom without changing the compute or memory.
Is this a general-purpose Windows desktop as well?
No. It ships with NVIDIA DGX OS, a Linux-based operating system built for the GB10 platform's AI software stack. It is purpose-built for AI development work rather than as a Windows office PC.
Do I need special power or cooling for it?
No — it runs from a standard socket through the supplied 240 W adapter and uses a triple-fan, dual vapor-chamber cooling system designed to stay quiet under sustained AI workloads, so it can sit on a desk rather than needing a server room.
Can I link two units together, and what for?
Yes, over the built-in ConnectX-7 SmartNIC. Two linked GX10 units can pool memory and compute to work on models beyond what a single unit's 128 GB can hold. Ask us about the correct interconnect cable when ordering a second unit.
What is included in the box?
The GX10 unit itself, the 240 W power adapter, a power cord, a user manual and a warranty card. It does not include a monitor, keyboard or mouse.
What warranty do I get in Singapore, and is there a GST invoice?
ASUS backs the Ascent GX10 with a 1-year onsite warranty. Every unit supplied by SourceIT comes with local Singapore support and a GST invoice; email sales@sourceit.com.sg for enterprise or volume procurement enquiries.
