This paper represents the first investigation of the suitability and performance of Graphcore Intelligence Processing Units (IPUs) for deep learning applications in cosmology. It presents the benchmark between a Nvidia V100 GPU and a Graphcore MK1 (GC2) IPU on three cosmological use cases: a classical deep neural network and a Bayesian neural network (BNN) for galaxy shape estimation, and a generative network for galaxy images production. The results suggest that IPUs could be a potential avenue to address the increasing computation needs in cosmology.
The paper can be found here: https://arxiv.org/abs/2106.02465.
This work was done with the help of people from Graphcore and they wrote a blog post (click here) and edited the video on the left to summarize and present this work.
I aso presented this work at the IN2P3/IRFU Machine Learning workshop. The slides are available here and the video on the left is the recording of the presentation.