ASUS has expanded its compact AI computing lineup with a new 64GB variant of the Ascent GX10, giving developers and AI enthusiasts access to powerful local AI computing in a desktop-sized system. The announcement was made on October 5, 2026, as demand for running increasingly capable AI models locally continues to grow.
At the heart of the ASUS Ascent GX10 is NVIDIA’s GB10 Grace Blackwell Superchip, which combines a 20-core NVIDIA Grace CPU with a Blackwell GPU architecture. The system features 64GB of LPDDR5X unified memory and offers memory bandwidth of up to 273GB/s, allowing the CPU and GPU to efficiently access the same memory pool.
ASUS says the Ascent GX10 can deliver up to 1 PFLOP of AI performance, making it suitable for local development, inference, generative AI workloads and experimentation with large AI models. The system is designed to handle AI models with tens of billions of parameters, reducing the need to send workloads to cloud-based platforms for every task.
The GX10 also supports NVIDIA’s AI software ecosystem, including tools such as PyTorch, Jupyter and Ollama, giving developers multiple options for building and running AI applications locally. Its compact form factor makes it significantly smaller than traditional workstation-class AI systems.
For users who need additional memory and performance, two Ascent GX10 systems can be connected, allowing up to 128GB of pooled memory for larger AI workloads. ASUS also uses a dedicated cooling system with heat pipes and dual fans to manage the system’s performance under sustained workloads.
With its combination of Blackwell architecture, unified memory and compact desktop design, the ASUS Ascent GX10 64GB variant targets developers who want serious AI computing without relying entirely on cloud infrastructure.
Comments