ASUS Announces New 64GB Configuration of ASUS Ascent GX10 AI Supercomputer
The market for localized artificial intelligence hardware is expanding beyond traditional data centers. ASUS has introduced a new 64GB configuration of the Ascent GX10, a compact supercomputer built on the NVIDIA DGX Spark platform. This release targets developers who require high-density compute power without the infrastructure of a server room.
High-Performance Compute in a Compact Chassis
As press.asus.com reports, the ASUS Ascent GX10 64GB utilizes the NVIDIA GB10 Grace Blackwell Superchip. This hardware enables the system to reach 1 petaflop of FP4 AI performance. The architecture links a Blackwell GPU with a 20-core NVIDIA Grace Arm CPU through an NVLink-C2C connection, ensuring high-speed communication between the processor and graphics unit.
The technical specifications for the new model include:
- 64GB of LPDDR5x coherent unified memory.
- Memory bandwidth reaching 273GB/s.
- Advanced cooling via five heat pipes and two 140x80mm fans.
- A 7-level fan control system for thermal management.
- Compatibility with standard wall sockets for power.
Scalability and Software Integration
What this means for you is the ability to run complex 30–35-billion-parameter open-source models and autonomous agents locally. For workloads exceeding the base capacity, ASUS allows two units to be clustered using NVIDIA ConnectX-7 networking. This configuration doubles the resources to 128GB of pooled memory and 2 petaflops of compute power.
Sized to run today’s most capable 30–35-billion-parameter open-source models and agents entirely on-device
— press.asus.com
The system ships with a pre-configured software environment based on NVIDIA DGX OS. The stack includes PyTorch, Jupyter, and Ollama. It also features NVIDIA OpenShell and NVIDIA NemoClaw to facilitate immediate development of AI agents.
On Paper vs. In Practice
On paper, the Ascent GX10 solves the problem of needing massive cloud instances for mid-sized model inference. In practice, its reliance on a standard wall socket makes it a pragmatic choice for office environments or small labs. While it competes with high-end workstations, the specific Grace Blackwell integration provides a specialized path for those already utilizing the NVIDIA AI stack.
Whether the ASUS Ascent GX10 64GB justifies its specialized hardware over a traditional multi-GPU desktop remains to be seen in long-term deployment tests. For developers focused on 35-billion-parameter models who value a quiet, plug-and-play supercomputer, this is an interesting option to monitor as shipping dates are confirmed.