Disaggregated Memory for High Performance Computing; Architectures and Opportunities for CXL

Tue Sep 17 | 9:45am
Location:
Santa Clara Ballroom
Abstract

HPE pioneered the concept of "memory-driven computing," spurred by its invention of memristor ReRAM, practical mid-board optics technology, and the GenZ interconnect. This presentation looks at some of that research and the direction it has taken since the consolidation of OpenCAPI, CCIX, and GenZ into the CXL specification and the failure of storage-class memory to gain a sustainable commercial foothold. We will examine current HPC architectures, including current HPE Cray EX systems such as the Frontier system at Oakridge National Laboratory and the Aurora system at Argonne National Laboratory--presently the #1 and #2 systems on the Top500 list--and look at future opportunities for emerging technologies, such as CXL memories and computational storage, in next-generation HPC architectures.

Learning Objectives

Upon completion, the participant will understand some of the design challenges faced by HPC system designers in the area of memory system design and storage.
Upon completion, the participant will understand some of the research that has been performed in the area.
Upon completion, the participant will understand understand the opportunities presented by new technology, such as CXL, HBM4, and 800Gb fabric interconnects.
Upon completion, the participant will be able to see the role that future computational storage systems may be able to play in high performance computing.
Upon completion, the participant will be able to understand the implications to HPC architecture posed by the end of Moore's Law and Dennard scaling.


---

David Emberson
Hewlett Packard Enterprise
Related Sessions