ASUS Announces New 64GB Configuration of ASUS Ascent GX10

Extending local agentic AI development to more builders alongside the existing 128GB model 

• Built for agentic AI: Purpose-built to develop and run autonomous AI agents locally; keeping data, prompts and inference on-device
• NVIDIA GB10 Grace Blackwell Superchip: Delivers up to 1 petaflop of AI performance in a compact chassis
• 64GB coherent unified memory: More accessible entry point for builders; sized to run today’s most capable 30–35-billion-parameter open-source models and agents entirely on-device
• Cluster-ready: Connect two units via built-in NVIDIA ConnectX-7 networking for 128GB of pooled memory and up to 2 petaflops of AI compute

TAIWAN - ASUS announced a new 64GB configuration of ASUS Ascent GX10, expanding the lineup alongside the existing 128GB model. This compact AI supercomputer is designed to bring local agentic AI development within reach of more developers, researchers and data scientists. Based on the NVIDIA DGX Spark™ platform and powered by the full NVIDIA AI software stack, ASUS Ascent GX10 64GB delivers the same petaflop-scale architecture found on ASUS Ascent GX10 128GB in a memory configuration built for today’s efficient open-source models — enough to build and run today’s most capable 30–35B class models and autonomous agents. When workloads grow, two systems can be clustered with the NVIDIA Sync Cluster Assistant for 128GB of pooled memory and around 1.7X the throughput, according to the official website of ASUS. 

Ubiquitous AI, engineered for more builders

Guided by its Ubiquitous AI. Incredible Possibilities. strategy, ASUS is working to put agentic AI within reach of every builder, empowering not only the early professional AI teams already running frontier-scale models locally, but the much larger early-majority audience of developers, researchers and AI-curious teams who are ready to start experimenting hands-on.

ASUS Ascent GX10 64GB is designed for exactly that audience and gives them a more accessible entry point into the same NVIDIA DGX Spark platform so more people can move from curiosity to proper agentic AI development. Meanwhile, ASUS Ascent GX10 128GB continues to serve teams pushing the largest local models and most demanding workloads.

Local infrastructure for the age of AI agents

Agentic AI has changed how organizations use AI — from a tool people query into a system that autonomously plans, uses tools and executes complex, multi-step tasks. As frameworks across the industry move agents from prototypes into daily production workloads, token consumption has surged, driving up cloud inference costs and raising new questions about how confidential business data and intellectual property move through public APIs.

Running open-source models on local hardware lets teams keep token generation and data in-house, without giving up the performance agentic workflows demand. As agentic AI expands to more users and use cases, capable local systems are becoming essential infrastructure, with ASUS Ascent GX10 64GB ready to meet that need.

Unleash the potential of AI

ASUS Ascent GX10 64GB delivers uncompromised local AI. Its unified architecture eliminates bottlenecks, providing a robust memory pool where the latest and most powerful LLMs are engineered to thrive, ensuring peak performance for agentic AI workloads.

 
Petaflop-scale performance, right-sized for today’s models

At the heart of ASUS Ascent GX10 64GB, based on the NVIDIA DGX Spark platform is the same NVIDIA GB10 Grace Blackwell Superchip that powers the original Ascent GX10, pairing a Blackwell GPU with fifth-generation Tensor Cores and FP4 support with a 20-core NVIDIA Grace Arm CPU. The GB10 Superchip uses NVIDIA NVLink-C2C to deliver a coherent CPU+GPU memory model with 5X the bandwidth of PCIe® Gen 5, enabling up to 1 petaflop of AI performance[i] from a compact, power-efficient device that runs from a standard wall socket — ideal for always-on agents.

64GB unified memory

ASUS Ascent GX10 64GB pairs 64GB of LPDDR5x coherent unified memory with 273GB/s of memory bandwidth to give developers, researchers and AI enthusiasts a platform sized for the latest generation of efficient open-source models. AI models such as Qwen3.8-27B now deliver performance comparable to frontier models from just months ago, within a memory footprint that fits comfortably on ASUS Ascent GX10 64GB. ASUS Ascent GX10 64GB offers a more flexible, lower cost of entry to the same NVIDIA DGX Spark platform that powers the 128GB configuration, enabling local AI experimentation, model development and inference, while the 128GB variant offers greater memory capacity for demanding AI workloads.

Scale on demand

ASUS Ascent GX10 64GB is built to grow with a developer’s workload. Using the system’s built-in NVIDIA ConnectX-7 networking, two Ascent GX10 64GB units can be clustered for up to 128GB of unified memory, up to 2 petaflops of AI compute and up to 546GB/s of combined memory bandwidth — delivering up to 1.7X the performance of a single ASUS Ascent GX10 128GB system. With clustering, and support of NVIDIA’s NVFP4 format users can run compressed versions of larger models, including Qwen3.8-Flash-Next. Setup takes just minutes with the NVIDIA Sync App and its Cluster Assistant, with no changes to the software environment or workflow.

This scale-out design gives teams a clear upgrade path: start with a single ASUS Ascent GX10 64GB for prototyping and inference, then cluster additional units as models, context windows and concurrent agents grow — all while keeping models, prompts and data on-device.

Precision-crafted for ultimate thermal efficiency

ASUS Ascent GX10 64GB shares the meticulously engineered thermal design of ASUS Ascent GX10 128GB and is optimized to handle demanding AI workloads and sustained high performance over long-running agent sessions. The advanced thermal system features 7-level fan control, ultrawide fins, five heat pipes and two 140x80mm fans that draw air through discreet bottom vents for powerful, efficient airflow.

ASUS Ascent GX10 64GB ships with the full NVIDIA AI software stack on NVIDIA DGX OS — PyTorch, Jupyter and Ollama — plus NVIDIA OpenShell and NVIDIA NemoClaw for policy-based agent security, and built-in SSH-based remote management so IT teams can provision and monitor system fleets alongside existing infrastructure.

From personal agent computers to enterprise AI infrastructure

ASUS is extending agentic AI across a comprehensive portfolio of local systems. ProArt laptops and mini PCs powered by NVIDIA RTX Spark serve as personal, front-end AI workspaces for content creation, agent interaction, application control, and workflow development. ASUS Ascent GX10, powered by NVIDIA DGX Spark, extends local AI development and model prototyping capabilities, while ASUS ExpertCenter Pro ET900, powered by NVIDIA DGX Station, address larger-scale professional and enterprise workloads.

[i] Petaflop of FP4 AI performance using sparsity