About
AI/HPC applications, which are becoming an issue these days, use large amounts of data and have very high computational complexity, necessitating the development of a system structure that improves computing/memory access performance.
Not only CPUs and GPUs but also accelerators are continuously being developed. On the memory side, new memory/computer complex structures and interconnectors such as PIM, PNM, and CXL are emerging to resolve the bottlenecks.
It is important to explore how these various hardware components can be combined to maximize performance. Instead of using actual hardware, it is necessary to develop a simulation methodology that predicts the application's performance using a hardware simulation model.
We expand our laboratory's trace-based parallel simulator (HSIM) and build a system simulation environment including HW accelerator and HW simulation models such as PIM, PNM, and CXL. We develop optimal structural design space exploration technology through this.
Through the simulator, design space exploration is performed to find a better-performing system structure by identifying bottlenecks that occur when performing AI/HPC applications in the system and repeating the structural change process to eliminate them.
Demo Videos
|
DAC 2011 |
| HSIM, parallel s... |
| 2014-10-02 12:30 |