Updated
Updated · USC Viterbi School of Engineering · Sep 3
4 USC Viterbi Researchers Complete Extreme-Scale Computing Training for GPU-Heavy Science
Updated
Updated · USC Viterbi School of Engineering · Sep 3

4 USC Viterbi Researchers Complete Extreme-Scale Computing Training for GPU-Heavy Science

2 articles · Updated · USC Viterbi School of Engineering · Sep 3

Summary

  • Four USC Viterbi researchers were among 75 people selected worldwide for Argonne’s two-week ATPESC program, then returned with training aimed at large-scale scientific computing.
  • The curriculum focused on GPU architectures, parallel programming, performance profiling, numerical methods and AI-enabled workflows—skills the group said are increasingly necessary as hardware shifts toward AI-oriented designs.
  • Benran Zhang and Nitish Baradwaj apply that training to materials simulations, from first-principles calculations that can use tens of thousands of GPUs to machine-learned molecular dynamics spanning millions of atoms.
  • Aishwarya Krishnan and Ryan Zapp use similar tools for hypersonic reentry research, where simulations can involve hundreds of millions of cells or run for up to one month on supercomputers.
  • Across those projects, the researchers said the main challenge is not just scaling code but verifying accuracy, uncertainty and physical validity as AI-generated software and machine-learned models spread through scientific workflows.

Insights

As supercomputers pivot to AI workloads, how will traditional scientific codes survive the hardware shift without losing physical accuracy?
Could the rise of AI-centric supercomputing hardware actually slow down scientific discovery by creating massive new bottlenecks in code verification?
If AI accelerates extreme-scale simulations, who ensures these powerful black-box models are not hallucinating the fundamental laws of physics?