GigaIO
10 Case Studies
A GigaIO Case Study
San Diego Supercomputer Center (SDSC) needed a better way to run large-scale GPU simulation workloads than its converged Comet environment, where GPU resources were locked inside individual servers and could not be easily shared. The challenge was to improve utilization, performance, scale, and management while controlling cost.
GigaIO addressed this with its FabreX composable PCIe fabric, using a 1U server and 12 NVIDIA 1080 Ti GPUs pooled across two Accelerator Pooling Appliances. The result was higher efficiency and better performance: utilization rose from 62% to 98% with four GPUs, and at 12 GPUs performance increased to 82% with GigaIO versus 39% on SDSC’s converged setup, delivering faster time to results and lower CapEx/OpEx.