Big Compute Podcast – New Architectures for HPC

In the Big Compute podcast episode, host Gabriel Broner engages with Mike Woodacre, an HPE Fellow, to discuss the transition from traditional CPU architectures to a diversified range of systems incorporating CPUs, GPUs, FPGAs, and other technologies. This shift has been prompted by the plateauing performance of CPUs due to miniaturization and power constraints, leading to a new era of specialized architectures that align with the exponential growth of data. Woodacre highlights how GPUs, initially designed for graphics, have become pivotal in AI and machine learning due to their data parallelism capabilities. The conversation also explores memory-driven architectures, which hold all data in memory to eliminate storage bottlenecks and accelerate data analysis processes. Additionally, the cloud is recognized as a facilitator of this architectural diversity, allowing for the deployment of different applications on suitable architectures, thereby optimizing efficiency. Security concerns regarding cloud usage have diminished over time, making it a viable entry point for experimenting with various architectural options.