Powered by Indico
v3.2.8
Title: Practical HPC
Abstract: High performance computing (HPC) involves maximising the efficiency of system resources such as CPU, GPU, and FPGA. As a developer, it's important to have practical skills that can achieve high performance. This talk will give practical advice on the following topics:
* Reliable micro-benchmarking techniques for identifying sources of inefficiency in the code
* Strategies for solving common bottlenecks such as branches
* Techniques for generating portable and fast vector code on any platform
* Maximising compile-time execution to minimise runtime performance costs.
The aim of the talk is to provide insights and tools for achieving optimal performance in HPC development, with a focus on C++. It should be accessible to everyone across the Institute.
Bio:
Marco Barbone is a graduate student in the Department of Computing at Imperial College, London. His research areas include MRI imaging, high performance computing, FPGAs, performance modeling, and Monte Carlo algorithms. He is a member of the CMS experiment at CERN, and a former software engineer at Maxeler Tecnologies and Intel. He is also a contributor to the FINUFFT code based at the Flatiron Institute.
https://github.com/DiamonDinoia