Key Highlights
Up to 1 PFLOP FP4
NVIDIA GB10 Grace Blackwell architecture delivers high-throughput AI computing for supported inference and development workloads.
128GB LPDDR5x
A coherent unified memory architecture allows the CPU and GPU to work with a shared memory pool for supported AI workloads.
ConnectX-7 SmartNIC
High-performance networking supports data-intensive workflows and linking compatible GX10 systems for larger workloads.
150 × 150 × 51mm
A small desktop footprint provides a dedicated AI development platform for labs, offices and research environments.
Introduction
The ASUS Ascent GX10 is an ultra-compact desktop AI supercomputer based on the NVIDIA DGX Spark platform. At its core is the NVIDIA GB10 Grace Blackwell Superchip, which combines an Arm-based CPU and an integrated NVIDIA Blackwell GPU in a unified computing architecture.
Unlike a conventional desktop PC with a separate graphics card and upgradeable DIMM memory, the GX10 uses 128GB of LPDDR5x unified memory shared across the CPU and GPU architecture. This design is intended for local AI development, model inference, fine-tuning experiments, data science and AI agent workflows that benefit from substantial shared memory in a compact system.
Features and Technologies
NVIDIA GB10 Grace Blackwell Superchip
The GB10 integrates a 20-core Arm CPU and an NVIDIA Blackwell GPU with fifth-generation Tensor Cores and fourth-generation RT Cores. This integrated design targets AI development and inference in a compact system.
128GB Coherent Unified Memory
The 128GB LPDDR5x memory pool is shared across CPU and GPU resources through the unified architecture. This can accommodate workloads that benefit from a larger shared memory space, subject to model, framework and runtime requirements.
NVIDIA AI Software Stack
The platform is designed around NVIDIA DGX OS and its AI software ecosystem, including CUDA, PyTorch, TensorFlow, Jupyter, TensorRT, NVIDIA NIM and AI Blueprints. Availability and versions can vary with software updates.
ConnectX-7 High-Speed Networking
The integrated NVIDIA ConnectX-7 SmartNIC provides high-performance networking for supported data-intensive applications and enables compatible systems to be linked for distributed AI workflows.
NVLink-C2C Technology
NVLink-C2C provides high-bandwidth communication between CPU and GPU components, supporting the unified-memory architecture and reducing reliance on conventional discrete CPU-to-GPU transfer paths.
Ultra-Compact AI Computing
The 150 × 150 × 51mm chassis and approximately 1.48kg listed weight make the GX10 suitable for space-constrained labs, development desks and edge-computing environments.
A Compact Platform for Large-Model Experimentation
ASUS positions the Ascent GX10 for prototyping, inference and fine-tuning of large AI models, with support for models up to 200 billion parameters cited in its product material. Model feasibility depends on precision, quantization, context length, runtime overhead and workload configuration. Larger distributed workloads may require linking compatible GX10 systems and configuring suitable software.
Technical Specifications
Specifications are based on the ASUS manufacturer specifications and the Prime ABGB 4TB listing. Confirm the exact storage variant, bundled accessories, warranty coverage and software configuration with the seller before purchase.
Ideal Use Cases
Local AI Inference
Run compatible open-weight models locally and experiment with inference workflows without requiring every task to be processed by a remote service.
AI Model Development
Prototype, fine-tune and evaluate supported AI models using the NVIDIA software stack and compatible development tools.
Agentic AI Applications
Develop and test AI agents and automated workflows using compatible frameworks, local models and appropriate security controls.
Research and Data Science
Support data processing, experimentation and model evaluation in laboratories, development teams and research environments.
Edge AI Deployment
Use the compact form factor for supported AI applications in space-constrained environments, subject to deployment and environmental requirements.
Multi-System AI Workflows
Link compatible GX10 systems over supported networking to explore larger distributed workloads. Software setup and model partitioning are required.
Before You Buy
- Confirm the included SSD capacity. ASUS lists 1TB, 2TB and 4TB configurations; the retailer listing associated with this product identifies the 4TB version.
- Remember that the 128GB LPDDR5x memory is unified system memory, not conventional upgradeable desktop RAM. Confirm your model and memory requirements before purchasing.
- Check that your intended AI frameworks, model formats and development tools support the Arm-based system and NVIDIA DGX software environment.
- The 1 PFLOP figure refers to peak AI performance using FP4. Performance varies by precision, workload, software and model configuration.
- For larger workloads, verify the networking, software and configuration requirements for connecting multiple systems.
- Confirm the bundled power adapter, accessories, warranty period and after-sales support with the seller.
Frequently Asked Questions
Explore AI Computing and Workstation Solutions
Discover computing hardware at Clarion Computers for AI development, GPU-accelerated workflows and high-performance workstation builds.