Description
The ASUS Ascent GX10 represents a paradigm shift in localized AI development, packing the NVIDIA GB10 Grace Blackwell Superchip into a 1.48kg chassis. Delivering 1 petaFLOP of AI performance, this "DGX Spark" unit enables researchers to run complex simulations and fine-tune large-scale models without the latency or cost of cloud-based instances. Its high-density design bridges the gap between desktop workstations and data center infrastructure.
This product is for:
- AI Researchers: Ideal for local prototyping of neural networks and deep learning models without requiring constant cloud connectivity.
- Data Scientists: Optimized for fine-tuning models with up to 200 billion parameters using 128GB of high-speed LPDDR5X memory.
- Edge AI Engineers: Perfect for deploying high-performance inference at the edge in industrial, medical, or laboratory settings.
- Enterprise Developers: Provides a seamless transition from local development to NVIDIA-powered data centers via a unified software stack.
- Academic Institutions: Offers supercomputing-grade performance in a silent, space-saving form factor suitable for shared office environments.
Detailed specifications:
- Processor: NVIDIA GB10 Grace Blackwell Superchip (ARM v9.2-A CPU architecture).
- Memory: 128GB LPDDR5X integrated memory for high-bandwidth AI workloads.
- Storage: 1TB M.2 NVMe PCIe 4.0 SSD (available configurations up to 4TB).
- Networking: NVIDIA ConnectX-7 SmartNIC, 10G LAN, Wi-Fi 7, and Bluetooth 5.4.
- I/O Ports: 3x USB 3.2 Gen 2x2 Type-C (DP 2.1), 1x HDMI 2.1, and 180W PD 3.1 support.
- Physical: 150 x 150 x 51 mm dimensions; 1.48kg total weight; Stellar Grey finish.
Operating on Ubuntu Linux, the GX10 utilizes NVIDIA NVLink-C2C technology for ultra-high-speed communication between the CPU and GPU. The advanced thermal design ensures the 240W power envelope is managed efficiently, maintaining peak performance during sustained AI workloads. Its dual-system stacking capability via ConnectX-7 allows for linear scaling of compute resources, making it a versatile building block for modular AI clusters.
Unlike traditional rack-mounted servers that require dedicated cooling and specialized infrastructure, the Ascent GX10 provides data-center-class Blackwell architecture in a desktop footprint. It outperforms standard consumer workstations in AI-specific tasks by leveraging the integrated Grace Blackwell chip, which offers significantly higher memory bandwidth and specialized AI tensor cores compared to discrete consumer-grade GPUs.
Q&A:
- What is the maximum AI compute performance? The GX10 delivers up to 1 petaFLOP of AI performance for specialized workloads.
- Can this unit fine-tune Large Language Models? Yes, it is specifically designed to support the fine-tuning of models with up to 200 billion (200B) parameters.
- What networking standards are supported? It features 10G Gigabit Ethernet, Wi-Fi 7, and an NVIDIA ConnectX-7 SmartNIC for high-speed data transfer and low-latency networking.
- Is the storage expandable? While the base model includes a 1TB M.2 NVMe SSD, the platform supports configurations up to 4TB.
- What operating system is pre-installed? The system comes with Ubuntu Linux, ensuring compatibility with the full NVIDIA AI software stack.
