Extreme GPU Cluster Coming to Stanford
The Stanford Research Computing Facility (SRCF) will soon be the home of an exciting new computational cluster, thanks to a National Science Foundation Major Research Instrumentation grant awarded to Stanford principal investigator (PI) Todd Martinez (Chemistry/PULSE Institute) and co-PIs Tom Abel (Physics/KIPAC), Margot Gerritsen (ICME/Earth, Energy & Environmental Sciences) and Vijay Pande (Chemistry). Additional faculty participating on the grant were from the Schools of Earth, Energy & Environmental Sciences; Engineering; Humanities & Sciences; and Medicine.
The multimillion dollar cluster will feature 65 Cray CS-Storm servers (20 core, 256 GB RAM), each populated with NVIDIA Graphics Processing Units (GPUs). What makes this cluster stand out above other GPU-based systems is the extreme scale: Each server will have eight NVIDIA Tesla K80 accelerator boards, and each K80 board has 2 GPUs. So each server ends up with 16 GPUs, with more than 1,000 across the entire cluster. In addition, the system will have 1.9 petabytes of raw storage, an essential feature in this era of data-intensive computing.
GPU-accelerated computing is transforming computational science across many disciplines. Stanford researchers will use this unique computational resource to focus on four primary areas: astrophysics/cosmology, structural biology and bioinformatics, materials modeling, and climate/geophysics. These are areas where Stanford researchers have already put significant effort into optimizing their applications to take advantage of GPUs, but they have not had any systems with this many GPUs to run them on. In addition, the researchers will also focus on developing tools to help other researchers redesign their algorithms for GPUs.
Scheduled for delivery at the SRCF in mid-March, the cluster is expected to be ready for production use by June. Research Computing will provide system and storage administration as well as computational end-user support and training.