Arts & Sciences Events

Location

Abstract: 

The memory system has been evolving at a fast pace recently, driven by the emergence of large-scale applications and the advance of hardware technology. This trend calls for the birth of memory systems with rich heterogeneity, which combines multiple memory technologies with different latency, bandwidth, and capacity to construct main memory. The heterogeneity of memory systems brings a substantial disparity in the performance and efficiency, making the decision of which technology to use at what times intricate.

In this talk, I will discuss how to enable big memory applications with memory heterogeneity. First, I will present my recent work on enabling multi-billion parameter models with limited GPU resources by offloading both data and computation to CPU. Second, I will present our recent work on using heterogeneous persistent memory/DRAM to enable a large-scale plasma simulation code (i.e., WarpX) at unprecedented scales on a single machine. As a closing remark, I will share my view on memory management for future generation heterogeneous computing and how they handle emerging workloads more efficiently.

Bio:

Jie Ren is a PhD candidate at Electrical Engineering and Computer Science, the University of California, Merced. Her research focuses on developing practical techniques to solve memory management issues in parallel computing systems, particularly in developing runtime and system supports for heterogeneous memory systems. Jie obtains Graduate Dean's Dissertation Fellowship at UC Merced. Her publication appears in top-tier conferences in high-performance computing, machine learning, and computer systems, such as ICS, PPoPP, NeurIPS, USENIX ATC, HPCA, and FAST.

Contact

Jiajia Li

This Event Appears On

• Arts & Sciences Events
• Computer Science Events