Scientific computing has been pushing the boundaries of traditional architectures, driving the exploration of novel architectures for improved efficiency and acceleration. While GPUs have revolutionized many scientific workloads, their general-purpose design poses limitations. Understanding the performance characteristics of these novel architectures and their software stacks is crucial. The vast diversity of hardware designs and evolving software stacks complicate comparing these accelerators. It's essential to grasp their fundamental differences, capabilities, programming approaches, and performance scaling, especially for large-scale scientific applications both from scientific machine learning and traditional HPC domains.
This seminar will discuss the comparative performance evaluation of a broad range of applications across a variety of novel architectures. Early insights into the challenges and best practices for porting, optimizing, scaling, and benchmarking AI accelerators will be discussed. In addition, I will shed some light on future directions and ongoing work to understand the efficacy of Novel architectures for Accelerated Scientific Computing.
To add to calendar:
Enter your credentials.
Search for your seminar
Click “Add to calendar”