ASUS made the Ascent GX10 AI supercomputer for experts who want to build and check AI models right at home. You can use it by itself on your desk or hook it up with others to make your private cloud. It comes with NVIDIA ConnectX network cards inside that let you link several GX10 machines together. When you connect just two of these systems, they handle huge AI models with as many as 405 billion moving parts.
Experts love this setup because they can try out ideas, improve things, and test how models work without needing to borrow time on big shared computers. The machine runs NVIDIA AI Software, which means you can create something on your desktop and move it to cloud systems later without changing much of your code. This saves tons of time when you're ready to scale up.
The brain of the GX10 is an NVIDIA GB10 Grace Blackwell super chip that combines a strong Blackwell GPU with a 20-core Arm CPU. This team-up speeds up AI work thanks to the 128 GB of memory that both parts share. The chip has brand new fifth-generation Tensor Cores plus FP4 support, letting it pump out 1,000 TOPS of AI number-crunching power.
ASUS built the GX10 with NVIDIA NVLink-C2C technology, which creates a smooth memory system that works much better than regular PCIe 5.0 connections. These features help the small box handle tasks that usually require big machines, like making split-second decisions with AI or fine-tuning massive models. Everything fits in a compact case that sits right on your desk but works like something you'd find in a data center.
Experts love this setup because they can try out ideas, improve things, and test how models work without needing to borrow time on big shared computers. The machine runs NVIDIA AI Software, which means you can create something on your desktop and move it to cloud systems later without changing much of your code. This saves tons of time when you're ready to scale up.
The brain of the GX10 is an NVIDIA GB10 Grace Blackwell super chip that combines a strong Blackwell GPU with a 20-core Arm CPU. This team-up speeds up AI work thanks to the 128 GB of memory that both parts share. The chip has brand new fifth-generation Tensor Cores plus FP4 support, letting it pump out 1,000 TOPS of AI number-crunching power.
ASUS built the GX10 with NVIDIA NVLink-C2C technology, which creates a smooth memory system that works much better than regular PCIe 5.0 connections. These features help the small box handle tasks that usually require big machines, like making split-second decisions with AI or fine-tuning massive models. Everything fits in a compact case that sits right on your desk but works like something you'd find in a data center.
