The ASUS Ascent GX10 is a compact personal AI supercomputer built around the NVIDIA GB10 Grace Blackwell Superchip. Aimed at developers, researchers, and fast-moving startups, it squeezes serious acceleration into a 150 x 150 x 51 mm chassis without the usual server-room drama.
Grace Blackwell power in a tiny box
At its core sits a 20-core Grace CPU paired with a Blackwell GPU in one package, pushing up to 1 petaflop of AI performance for fine-tuning and high-throughput inference. The platform runs the full NVIDIA AI software stack, so you can jump from prototyping to deployment with fewer detours and less yak-shaving.
Unified memory and fast I/O for modern workflows
The GX10 ships with 128GB unified LPDDR5x memory, enabling work on models up to ~200-billion parameters right on your desk. Storage is modular: 1TB and 2TB M.2 2242 NVMe PCIe 4.0 x4 options, plus a forthcoming 4TB PCIe 5.0 x4 SKU for heavier datasets and multi-run experiments.
Scale out with low-latency links
Need more grunt tomorrow? Two GX10 units can be clustered via NVIDIA ConnectX-7 networking to double compute to 2 petaflops, while unified memory scales to 256GB and local storage to 8TB. The Superchip’s NVLink-C2C keeps CPU-GPU memory traffic moving with ultra-low latency, ideal for robotics, vision, and VLM pipelines.
Thermals tuned for sustained loads
The chassis is precision-engineered for quiet efficiency: 7-level fan control, ultrawide fins, five heat pipes, and twin 140 x 80 mm fans drawing fresh air from discreet bottom vents. The goal is simple: hold clocks under prolonged heat without throttling, even when every tensor core is busy.
I/O covers creator essentials without clutter. Rear ports include three USB 3.2 Gen 2×2 Type-C (20 Gbps with DP Alt Mode), one USB 3.2 Gen 2×2 Type-C with PD-in (EPR 180 W), HDMI 2.1, a 10G LAN, and a Kensington Lock, plus the ConnectX-7 NIC. Up front you get a clean power button for quick access.
Connectivity is equally modern thanks to WiFi 7 (2×2) and Bluetooth 5, backed by NVIDIA DGX OS and the NVIDIA AI Software Stack for a consistent dev environment. Power delivery supports up to 240 W over USB-C PD 3.1 EPR, while typical device input is rated at 180 W, keeping cables and bricks minimal.
Availability starts October 15, 2025, with configurations varying by storage tier. Pricing will depend on the chosen SSD option; the small-box compute value is the point here. For teams building locally with privacy in mind, the Ascent GX10 keeps datasets on-prem while offering a clean path to scale. Learn more at ASUS. ASUS positions it as the friendly on-ramp to serious AI work without dedicating a rack.
Technical Specifications
| ASUS Ascent GX10 | |
|---|---|
| Processor | ARM v9.2-A CPU (GB10) |
| Graphics | NVIDIA Blackwell GPU (GB10, integrated) |
| Memory | 128GB LPDDR5x Coherent Unified System Memory |
| Storage | 1TB M.2 2242 NVMe PCIe 4.0 x4 SSD; or 2TB M.2 2242 NVMe PCIe 4.0 x4 SSD; or 4TB M.2 2242 NVMe PCIe 5.0 x4 SSD |
| OS and software | NVIDIA DGX OS / NVIDIA AI Software Stack |
| Wireless | AW-EM637 WiFi 7 (Gig+) 2×2 and Bluetooth 5 |
| LAN | 1 x 10G LAN |
| I/O (rear) | 3 x USB 3.2 Gen 2×2 Type-C (20 Gbps, DP Alt Mode); 1 x USB 3.2 Gen 2×2 Type-C with PD-in (EPR 180 W, PD3.1); 1 x HDMI 2.1; 1 x NVIDIA ConnectX-7 NIC; 1 x 10G LAN; 1 x Kensington Lock |
| I/O (front) | Power button |
| Power adaptor output | Up to 240 W (USB-C PD 3.1 EPR), up to 48 V/5 A |
| Device power input (max) | 180 W via USB-C |
| Certifications | BSMI / CB / CE / FCC / UL / CCC / C-Tick / WiFi / RF / VCCI |
| Size | 150 x 150 x 51 mm (5.91 x 5.91 x 2.01 inches) |
| Weight | 1.48 kg (3.26 lb) |
