NVIDIA's GH200 Grace Hopper Superchip offers strong potential for accelerating large-scale AI and HPC workflows through its tightly integrated CPU-GPU architecture. In this talk, we share CU Boulder Research Computing’s first-hand experience providing GH200 nodes to users. We'll cover the GH200 architecture, our approach to the software stack, an overview of our beta testing phase, and successful use cases run on the GH200s. We'll conclude with potential future directions for CURC’s GH200 resources and describe how RMACC members can access the nodes for their research or educational purposes.
