Star Plenary Debate: Prepping for Dense Data: What Breaks When Cognitive Workloads Dominate Compute?
The march towards intelligent, software-writing data sets is evolving and taking the operation of mission-critical facilities along for the ride. From 2015 to 2017, the amount of compute required by deep learning algorithms jumped by 15x while the compute delivered by GPUs jumped by 10x. With the consolidation of hardware coupled with the virtualization of applications driving up average density per rack to 15-16kW, it is difficult to forecast how the industry's standard rack design, temperature frameworks and power generation will withstand the pressure of next-generation workloads. Our panellists will join together to outline the future of data center compute and provide a path to infrastructure optimization for operators.