ASUS recently launched the Ascent GX10 mini workstation, equipped with NVIDIA GB10 Grace Blackwell Superchip chip. The 1TB version is priced at 32,999 yuan.The overall size of Ascent GX10 is 150 x 150×51 mm, which is less than one-tenth the size of a traditional workstation. It uses a customized cooling system internally, combining ultra-wide fins, five heat pipes and dual 140×80mm fans to ensure that the device can maintain stable operation under sustained high loads.
The NVIDIA GB10 Grace Blackwell Superchip equipped on this device is composed of Blackwell GPU and Grace CPU and is optimized for a compact body. It can run large language models with more than 200 billion parameters.Among them, Grace CPU has 20 Arm cores, which can enhance data preprocessing and orchestration efficiency, accelerate model tuning and real-time inference.
In terms of hardware configuration, Ascent GX10 is equipped with 128GB LPDDR5X unified memory and provides a variety of SSD storage options from 1TB to 4TB. The GPU is based on the Blackwell architecture, has 6144 CUDA cores, integrates the fifth-generation Tensor Core, and supports FP4 data format. Through NVIDIA NVLink-C2C chip interconnection technology, unified interconnection of CPU and GPU memory models is achieved, with bandwidth reaching five times that of PCIe 5.0.
Thanks to the integrated NVIDIA ConnectX-7 network port, users can connect two Ascent GX10s locally to handle more complex AI tasks, such as large models such as Llama 3.1 with up to 405 billion parameters.
In addition, users can complete the development and testing of AI models in a local environment, and then seamlessly migrate to NVIDIA DGX Cloud or other accelerated cloud and data center infrastructure. The entire process requires almost no code adjustment, thus significantly simplifying the prototyping, model fine-tuning and iteration process.
