Desktop AI Supercomputer

Premium. Accessible.

1 petaFLOP of AI performance. 128 GB unified memory. On your desk.

Scroll to explore

Overview

Supercomputer-class AI.
Desktop-class footprint.

The ASUS Ascent GX10 packs NVIDIA Grace Blackwell into a 1.6-litre chassis. Run, fine-tune, and deploy models up to 200B parameters — without the cloud.

1
petaFLOP
FP4 AI Performance
128
GB
Unified LPDDR5x
200B
parameters
Single-Unit Fine-tuning
1.6L
volume
Ultra-Compact Chassis

Features

Built for serious AI work.

GB10
Grace Blackwell
Superchip

NVIDIA Grace Blackwell Superchip

20-core ARM CPU paired with a Blackwell GPU featuring 5th-generation Tensor Cores. NVLink-C2C delivers 5x the bandwidth of PCIe Gen 5 for seamless CPU-GPU coherency.

  • FP4 precision for maximum throughput
  • 128 GB LPDDR5x coherent unified memory
  • Dual-unit stacking for 405B parameter models
Dual Vapor Chamber
7-Level Fan Control

Advanced thermal engineering

Dual vapor chamber design delivers 1.6x more efficient thermal coverage than comparable systems. 7-level fan control keeps acoustics in check — from silent desktop use to full compute loads.

  • MIL-STD 810H certified durability
  • 282.4 x 187.7 x 56.5 mm — fits anywhere
  • 24/7 continuous AI execution
ConnectX-7
10 GbE + 4x USB-C 20Gbps

Enterprise-grade connectivity

NVIDIA ConnectX-7 NIC with hardware TLS/IPsec acceleration. Link two GX10 units for 405B parameter workloads. 4x USB-C 20Gbps ports with DisplayPort 2.1 alt mode and 180W power delivery.

  • HDMI 2.1b + DisplayPort 2.1 via USB-C
  • IEEE 1588v2 PTP microsecond sync
  • Kensington lock for physical security

Technical Specifications

Every detail, at a glance.

Processor
NVIDIA GB10 Grace Blackwell Superchip
20-core ARM CPU + Blackwell GPU
GPU
Blackwell Architecture
5th-gen Tensor Cores, FP4 support
Memory
128 GB LPDDR5x
Coherent unified, NVLink-C2C (5x PCIe Gen 5)
AI Performance
1 petaFLOP (FP4)
Up to 200B parameters single-unit
Networking
NVIDIA ConnectX-7 + 10 GbE LAN
Hardware TLS, IPsec, MACsec acceleration
Display Output
1x HDMI 2.1b
+ DisplayPort 2.1 via USB-C alt mode
USB Ports
4x USB 3.2 Gen 2x2 Type-C
20 Gbps each, 180W EPR PD3.1 support
Dimensions
282.4 x 187.7 x 56.5 mm
~1.6L volume, dual vapor chamber cooling
Cooling
Dual Vapor Chamber + Dual Fan
7-level fan control, 1.6x thermal efficiency
Operating System
NVIDIA DGX OS
Ubuntu-based, preloaded AI stack
Certifications
MIL-STD 810H
Military-grade durability testing
Security
On-premises data sovereignty
Kensington lock, NemoClaw guardrails

Software

Ready to run. Out of the box.

Preloaded with the full NVIDIA AI software stack. Start training and inferencing within minutes of unboxing.

PyTorch
Deep learning
TensorFlow
ML framework
TensorRT
Inference engine
CUDA
GPU compute
Jupyter
Notebooks
Ollama
Local LLM inference
NVIDIA NIM
Model deployment
NemoClaw
AI guardrails

Supported Models

Llama 3.1 (405B dual) DeepSeek R1 (70B) Gemma Mistral Qwen Stable Diffusion Custom fine-tuned models

Use Cases

From prototype to production.

LLM Development

Fine-tune and run large language models locally. Prototype before scaling to cloud — without per-token API costs.

Computer Vision

Real-time object detection, tracking, and image generation. Process video feeds locally with zero latency to the cloud.

Data Sovereignty

Keep sensitive data on-premises. Run inference and training without sending a single byte to the cloud. Full control, always.

Get Started

Ready to bring AI in-house?

Download the full datasheet for complete specifications, benchmarks, and deployment guides.

Download Datasheet (PDF)
Shopping cart

Your cart is empty.

Return to shop
close