Large AI models can be hard to run. This small desktop makes it easy, fast, and safe to train and test AI right on your desk.
Many researchers, developers, and data scientists often face a common challenge like training and running large AI models. Large models require massive computing power, but high-performance systems are usually bulky, expensive, or require cloud resources that raise security concerns. ASUS’s Ascent GX10 addresses this gap by offering a compact desktop AI supercomputer that delivers enterprise-level AI performance without taking up an entire room.
At its core, the Ascent GX10 combines a 20-core NVIDIA Grace CPU with an NVIDIA Blackwell GPU, reaching up to 1 petaflop of AI performance. This allows users to run large-scale models and fine-tune them directly on a desktop, handling models with up to 200 billion parameters thanks to 128GB of unified memory. Its small 150 x 150 x 51mm footprint makes it suitable for any workspace, offering high performance without compromising space.
The system comes preloaded with the full NVIDIA AI software stack, giving developers everything needed for prototyping, model tuning, and inference. It supports AI applications in robotics, computer vision, and Vision-Language Models (VLMs), making it easier for teams to move from research to real-world testing without complex setups.
For teams needing more power, the AI supercomputer can scale. Two units connected via high-speed ConnectX-7 interfaces can double performance to 2 petaflops, expand memory to 256GB, and provide up to 8TB storage. This enables local, secure AI training without relying on cloud infrastructure, keeping sensitive data on-premises while reducing costs.
Storage options are flexible to match different workflows. A 1TB NVMe SSD suits beginners or small datasets, while upcoming 2TB and future 4TB models will support larger datasets, multiple models, and concurrent experiments. Users can access all their data instantly without bottlenecks.
Finally, the Ascent GX10 tackles the common problem of overheating in compact high-performance systems. Its advanced thermal design including 7-level fan control, ultrawide fins, five heat pipes, and dual 140 x 80mm fans ensures stable performance even under heavy workloads, keeping operations smooth and reliable.
