The increasing size and complexity of today’s generative AI models has presented developers with a significant bottleneck. Prototyping, testing, and fine-tuning these massive models, with billions of parameters, required massive amounts of memory and processing power typically only available in large data centers or cloud infrastructures. This both increased costs and slowed down development processes.

This is where the ASUS Ascent GX10 steps in, addressing this obstacle. Designed as a powerful and affordable desktop solution, the Ascent GX10 promises AI professionals flexibility and speed previously unimaginable.

What does the ASUS Ascent GX1 offer?

At the heart of the Ascent GX10 lies NVIDIA’s groundbreaking GB10 Grace Blackwell Superchip architecture. This technology combines a high-performance Blackwell GPU and a 20-core Grace Arm CPU on a single chip. The NVIDIA NVLink-C2C connection, which facilitates data flow between the two units, offers five times the bandwidth of traditional PCIe 5.0 technology. This architecture unlocks incredible AI processing power of up to 1,000 TOPS.

The technical features offered by the ASUS Ascent GX10 translate into tangible benefits in AI development:

• Massive Model Support: Thanks to its 128GB of unified memory, it easily handles massive AI models with up to 200 billion parameters.
• Scalable Architecture: A single Ascent GX10 can handle models with 70 billion parameters, while the two systems are connected via integrated ConnectX-7 NICs, providing sufficient power for massive models with 405 billion parameters, such as Llama 3.1.
• Local and Flexible Development: Developers can prototype, test, and optimize their models directly at their desks without tying up expensive cloud resources. Thanks to the NVIDIA AI software ecosystem, a model developed on the desktop can be easily ported to cloud or data center infrastructures with virtually no code changes.