Processor microarchitecture research is experiencing a renaissance, as evidenced with recent striking advances in designs from Apple, Arm, NVIDIA, AMD and Intel, amongst others. Current generation cycle-level processor simulators do not provide the level of accuracy, performance or ease of use to enable this level of fundamental microarchitecture research. The problem is compounded when considering applying machine-learning based design space exploration, in order to automate the microarchitecture optimisation process.
In this talk we will explain the challenges currently facing researchers in this area, and introduce the Simulation Engine, or SimEng. SimEng is a new processor simulation framework which enables fast, accurate, easy to use cycle-level modelling of contemporary microarchitectures. It is being used to understand existing advanced processors, such as Apple’s M1 and Fujitsu’s A64fx, while also being used to explore hypothetical future designs.
Speaker bio:
Simon McIntosh-Smith is Professor of HPC at the University of Bristol, UK. He began his career in industry as a microprocessor architect, first at Inmos and STMicro in the 1990s, before co-designing the world’s first fully programmable GPU at Pixelfusion in 1999. In 2002 he co-founded ClearSpeed Technology where, as Director of Architecture and Applications, he co-developed the first modern many-core HPC accelerators. He now leads the HPC Research Group in Bristol, where his research focuses on advanced computer architectures and performance portability. He leads the Isambard supercomputer service which combines Arm-based CPUs with a diverse range of CPUs and GPUs from all the main vendors.
Processor microarchitecture research is experiencing a renaissance, as evidenced with recent striking advances in designs from Apple, Arm, NVIDIA, AMD and Intel, amongst others. Current generation cycle-level processor simulators do not provide the level of accuracy, performance or ease of use to enable this level of fundamental microarchitecture research. The problem is compounded when considering applying machine-learning based design space exploration, in order to automate the microarchitecture optimisation process.