Asus Ascent GX10: 64GB compact PC launched for local AI workloads

Asus has expanded its lineup of compact Ascent GX10 computers by introducing a new configuration featuring 64 GB of unified memory. Built on the Nvidia DGX Spark platform, the system is designed for local artificial intelligence development, testing, and deployment. Previously, the company offered the GX10 exclusively with 128 GB of memory, and the new version aims to make working with open-source AI models more accessible.

• Asus launched the Ascent GX10 variant with 64 GB of LPDDR5x memory and a 1 TB SSD.
• The system is powered by the Nvidia GB10 Grace-Blackwell superchip, delivering up to 1 petaflop of FP4 performance.
• The computer is already available for pre-order in Europe at an estimated price of around 6,000 euros.
• Users can cluster two such systems together using Nvidia ConnectX-7 network technology for doubled resources.

Hardware Specs and Grace-Blackwell Superchip

At the heart of the Ascent GX10 is the Nvidia GB10 Grace-Blackwell superchip, which combines a Blackwell GPU featuring fifth-generation tensor cores with a 20-core CPU based on the Grace Arm architecture. The processor and graphics accelerator are linked via the NVLink-C2C interface, providing a shared memory space and bandwidth up to five times higher than PCIe 5.0, eliminating the need to constantly copy data between system and video RAM.

The 64GB version utilizes LPDDR5x unified memory with a bandwidth of 273 GB/s, accessible simultaneously by both the CPU and GPU. Asus targets this configuration at open-source models with 30 to 35 billion parameters, such as Qwen3.8-27B. While peak performance reaches 1 petaflop under low-precision FP4 computing with sparsity, actual speed depends on the specific model, software, and workload.

Scalability and Cluster Configuration

To handle higher demands, Asus allows users to link two GX10 systems using built-in Nvidia ConnectX-7 networking technology. According to the manufacturer, this cluster configuration provides up to 128 GB of combined memory, up to 2 petaflops of AI performance, and a total memory bandwidth of 546 GB/s. Setting up the cluster takes just a few minutes using the Nvidia Sync App with its built-in Cluster Assistant.

The cooling system for the 64GB version remains identical to the 128GB model, featuring seven fan control modes, wide cooling fins, five heat pipes, and two 140x80mm fans. Intake air flows through openings at the bottom of the chassis, though independent noise and power consumption measurements were not officially disclosed by the manufacturer.

Software Ecosystem and Enterprise Value

The pre-installed software bundle includes Nvidia DGX OS, PyTorch, Jupyter, and Ollama, alongside support for Nvidia OpenShell and Nvidia NemoClaw for security policy configuration. IT teams can remotely manage multiple devices simultaneously via SSH, making deployment across corporate environments straightforward and secure.

The shift toward local AI infrastructure is driven by rising cloud computing costs and the critical need to protect intellectual property and sensitive data. Running models locally ensures that user data and generated tokens do not need to pass through public APIs, giving businesses complete control over their operational environment and security standards.