Virtual Workshop 2026 - AI Meets CI - Day 3 - Navigating the Heterogeneous Computing Landscape, with Prasanna Balaprakash, PrimaLabs
Talk VI: Navigating the Heterogeneous Computing Landscape: Optimization Strategies for Modern AI Workloads, Prasanna Balaprakash, Co-founder and CEO of PrimaLabs
Abstract: Navigating the Heterogenous Computing Landscape: Optimization Strategies for Modern AI Workloads
The proliferation of diverse accelerator architectures—GPUs from multiple vendors, custom ASICs, and emerging neuromorphic and quantum-classical hybrid systems—presents both unprecedented opportunities and significant challenges for AI practitioners. While hardware diversity promises performance gains and cost efficiencies, realizing these benefits requires fundamentally rethinking how we design, deploy, and optimize AI workflows.
This talk examines the optimization landscape for heterogeneous computing environments, addressing three interconnected dimensions: hardware-aware optimization that accounts for architectural differences in memory hierarchies, interconnects, and compute characteristics; cost-aware scheduling that balances performance against operational expenditure in multi-cloud and on-premise deployments; and application-aware tuning that matches workload characteristics to optimal hardware configurations.
Drawing on experience deploying large-scale AI systems across leadership computing facilities, I present practical strategies for achieving efficient utilization on mixed-vendor GPU clusters, discuss automated approaches to hyperparameter and architecture search in heterogeneous environments, and outline emerging techniques for portable performance across diverse backends. The talk concludes with observations on where the field is headed as hardware fragmentation accelerates and the economic stakes of efficient AI infrastructure continue to rise.
The NSF CI Compass virtual workshop - AI Meets CI: Intelligent Infrastructure for Major & Midscale Facilities featured themes to host speakers, presentations, and discussions surrounding artificial intelligence (AI) and cyberinfrastructure (CI) within the data lifecycle in facilities, cloud computing, and about societal and human intersections with AI.
Originally hosted January 12, 13, and 14, 2026
This project is supported by the U.S. National Science Foundation Office of Advanced Cyberinfrastructure in the Directorate for Computer Information Science under Grant #2127548.