Modular Chiplet Opportunities for High-Performance Computing

Learn about the evolution and future of High Performance Computing (HPC) in this 17-minute conference talk from Lawrence Berkeley Lab's John Shalf at the Open Compute Project. Explore how HPC performance growth has significantly slowed from historical 1000x improvements over three decades to just 10x every 10 years. Discover why the traditional reliance on Moore's Law and the broader microprocessor market economics is no longer sufficient for advancing HPC capabilities. Examine the shift towards architectural specialization and the potential of chiplets containing specialized scientific accelerators, such as FFTs, as a solution for delivering future performance gains in HPC systems.