Photo credit: www.techradar.com
Asus Unveils Ascent GX10: Bringing AI Supercomputing to Developers
AI processing capabilities that rival high-end systems
The landscape of AI development is evolving rapidly, prompting companies like Asus to introduce innovative solutions. The new Ascent GX10 is a compact AI supercomputer designed to serve developers, researchers, and data scientists. It leverages Nvidia’s cutting-edge Grace Blackwell GB10 Superchip to deliver exceptional performance in AI workloads.
Positioned as a direct competitor to Nvidia’s DGX Spark, the Ascent GX10 streamlines the process of developing and testing AI models locally. This focus on in-house computing capability allows users to prototype, modify, and execute extensive models with greater ease, minimizing their dependence on external cloud or data center infrastructure.
At the heart of the Ascent GX10 lies an impressive architecture featuring 128GB of unified memory. Equipped with a Blackwell GPU that boasts fifth-generation Tensor Cores and support for FP4 precision, this system can achieve a stunning 1000 TOPS of AI processing prowess. Additionally, the inclusion of a 20-core Grace Arm CPU enhances data processing speeds, optimizing AI inferencing and model tuning. According to Asus, the GX10 can comfortably manage AI models up to 200 billion parameters, addressing potential performance bottlenecks often encountered in large-scale AI tasks.
A Compact Solution for AI Development
As KuoWei Chao, General Manager of Asus IoT and NUC Business Group, highlighted, “AI is transforming every industry, and the Asus Ascent GX10 is designed to bring this transformative power to every developer’s fingertips.” He emphasized that by incorporating the Nvidia Grace Blackwell Superchip, Asus has created a robust yet space-efficient tool that empowers developers, data scientists, and AI researchers to innovate directly from their workspaces.
To further enhance its performance, the GX10 utilizes NVLink-C2C technology, which substantially increases bandwidth — more than five times that provided by PCIe 5.0. This feature allows for efficient memory sharing between the CPU and GPU, resulting in improved performance across various AI workloads.
Moreover, the system is equipped with an integrated ConnectX network interface, enabling multiple Ascent GX10 units to be connected. This feature is particularly beneficial for handling even larger AI models, including Llama 3.1, which contains an astonishing 405 billion parameters.
Asus plans to offer the Ascent GX10 for pre-order in the second quarter of 2025. Although pricing details are yet to be finalized, Nvidia has indicated a starting price of $2999, which includes 1TB of storage. In contrast, Nvidia’s DGX Spark — a more expensive offering at $3999 with 4TB of storage — highlights the competitive positioning of Asus in this rapidly growing market.
Related Topics
Source
www.techradar.com