Cognitive simulation (CogSim) is an important and emerging workflow for HPC scientific exploration and scientific machine learning (SciML). This presentation will discuss recent tests with hybrid workflows that intertwine data-driven, learning models with traditional scientific simulation.  These workflows include complex physical simulation with a surrogate model “inside” the computational loop.  Our expert speakers will highlight the learnings of applying large, dedicated Deep Learning / AI accelerators that are disaggregated from compute nodes for CogSim workloads. 

Attendees will gain insights into:

• Learnings from the first of its kind Disaggregated CogSim system and how it was built

• Hybrid system results in accelerating multiple workloads

• Benefits of a highly programable architecture allowing multiple models to be laid out on a chip 

• The ability of multiple model deployments to be programmed to handle each incoming request with production-ready efficiency