ASUS recently unveiled the Ascent GX10, a mini AI supercomputer equipped with NVIDIA’s GB10 Grace Blackwell Superchip. This powerful yet “affordable” desktop computing solution packs up to 1000 TOPS (trillion operations per second) of AI performance into a compact chassis. It aims to meet the growing demand for AI model development while maintaining cost-effectiveness and portability.
At the heart of the Ascent GX10 is the NVIDIA GB10 chip, a platform that integrates a 20-core Arm-based Grace CPU with a Blackwell-architecture GPU. Using NVLink-C2C high-speed interconnect technology, it enables unified memory access between the CPU and GPU, delivering bandwidth five times that of PCIe 5.0. In contrast, NVIDIA’s higher-end GB300 platform offers up to 20,000 TOPS for massive-scale AI tasks, while the GB10 focuses on cost-performance balance, making it suitable for most small-to-medium AI workloads. The Ascent GX10 comes with 128GB of LPDDR5X unified memory, supporting developers in training, testing, and inferring AI models with up to 200 billion parameters, such as open-source large language models like DeepSeek, Grok, or Llama 2.
In terms of hardware design, the Ascent GX10 leverages the fifth-generation Tensor Cores and FP4 precision support of the Blackwell GPU. FP4, a low-precision floating-point format, significantly boosts computational speed and reduces memory usage while maintaining accuracy, making it ideal for real-time inference and model fine-tuning. Additionally, the device features 4TB of NVMe storage, sufficient to house large datasets and model files. ASUS has also integrated NVIDIA’s ConnectX-7 network interface, enabling users to scale computing power through dual-machine interconnectivity to handle models with doubled parameter sizes, such as Meta’s recently released Llama 3.1, which boasts 405 billion parameters.
The mini supercomputer’s exterior design is another highlight. The Ascent GX10 sports a sleek white chassis with a power button on the front and a distinctive, practical, and aesthetically pleasing top design. While specific port details have yet to be fully disclosed, based on NVIDIA’s similar Project DIGITS, it’s expected to offer common connectivity options like USB4 Type-C, Wi-Fi, Ethernet, and HDMI 2.1, allowing users to easily connect monitors, keyboards, and mice to set up a workstation quickly.
ASUS IoT and NUC Business Unit General Manager Kuo Wei-chao stated: “AI is reshaping industries, and the Ascent GX10 is designed to put this transformative power within reach of developers. By incorporating NVIDIA’s Grace Blackwell chip, we’ve created a compact yet powerful tool to help data scientists and AI researchers innovate in a desktop environment.” This positioning clearly reflects the Ascent GX10’s target audience: individuals or small teams needing local high-performance computing resources on a limited budget.
Meanwhile, competition in the AI supercomputer space is heating up. South Korea’s planned national AI supercomputer has been delayed to 2026 due to chip supply issues, while NVIDIA unveiled Project DIGITS at CES 2025 in January, also based on the GB10 chip, with a starting price of around $3,000. By comparison, the Ascent GX10 is expected to open pre-orders in Q2 2025, with pricing likely in a similar range, though exact details await ASUS’s official announcement. Market analysts note that as generative AI models grow increasingly complex, the demand for local computing power and memory is surging, and devices like the Ascent GX10 could fill the gap between cloud services and expensive data center solutions.
From a technical perspective, the GB10’s Grace CPU is based on the Arm Neoverse V2 architecture, featuring 10 Cortex-X925 high-performance cores and 10 Cortex-A725 efficiency cores, balancing multitasking and energy efficiency. The Blackwell GPU supports CUDA cores and ray tracing, delivering 1 PFLOPS of FP4 compute power, providing robust support for AI inference and lightweight training tasks. Notably, the Ascent GX10’s energy efficiency stands out—it runs on standard household power with a single plug, significantly lowering the entry barrier compared to traditional servers.
The launch of the Ascent GX10 is well-timed. Early 2025 marks a new wave of excitement in the AI hardware market, with AMD and Intel rolling out AI workstation processors, while NVIDIA continues to dominate with its GPU ecosystem. By partnering with NVIDIA to bring the GB10 chip to a desktop-grade product, ASUS not only expands its AIoT portfolio but also offers developers a flexible experimentation platform. Whether for prototyping, model tuning, or direct inference tasks, the Ascent GX10 delivers unexpectedly high performance in a compact form factor.
For tech enthusiasts, the Ascent GX10 is more than just hardware—it’s a window into the trend of AI technology moving from data centers to desktops. More details, including full specifications and release plans, are expected to be revealed in the coming months, making it a product worth keeping an eye on.