Moderator: Dan Olds (Olds Research)
Panelists: Phil Pokorny (Penguin Solutions), Kurt Keville (SemiconDx), Anil Godbole (Intel Corporation)
Abstract: As memory-bound workloads begin to dominate HPC and AI, traditional architectures face severe limitations in memory bandwidth and capacity. Compute Express Link® (CXL®) is providing the means to create disaggregated and composable memory tiers to accelerate HPC and AI performance.
This panel, moderated by Dan Olds, features experts from Penguin Solutions, SemiconDx, and Intel, who will explore the evolving role of CXL in next-generation HPC/AI systems. Panelists will discuss emerging memory architectures, performance benchmarks, cost and scale trade-offs, alongside lessons learned from early implementations and incorporation of emerging technologies, such as silicon photonics, that should be on everyone’s roadmap.
Through interactive discussion and audience engagement, the session will highlight real-world benchmarks discussing CXL-enabled systems, as well as practical insights into memory-constrained workloads. Attendees will understand current limitations, gain actionable guidance, and leave better equipped to evaluate and resolve memory constraints in HPC and AI environments.