Seminar Abstract:

With the rapid development of artificial intelligence and scientific computing, the demand for computational resources continues to grow. Modern hardware performance improvements are increasingly driven by massively parallel processors, including GPUs, multi-core CPUs, and heterogeneous architectures. To enable the design of next-generation systems, it is essential to understand commercial microarchitectures and develop efficient simulators for microarchitectural analysis and exploration.

This talk will introduce methodologies for uncovering and simulating modern parallel processors, such as GPUs and many-core CPUs. Building on these simulation frameworks, it will also present novel microarchitectural designs that improve performance for large-scale workloads, including large language models (LLMs).

About the Speaker:

Liu Changxi is a Research Fellow at the National University of Singapore (NUS). He received his Ph.D. degree from the School of Computing, NUS. His research focuses on CPU/GPU architecture and simulation. His work has been published in top-tier computer architecture conferences, including ISCA and MICRO.

Several of his research contributions have been integrated into widely used simulation frameworks. For example, PacSim has been merged into Sniper, a well-known multi-core CPU simulator, and Photon has been incorporated into MGPUSim, a high-performance GPU simulator that supports multi-GPU simulation.