Computer Science

Shifting sands of hardware and software in exascale simulations

·
1 min read

Our Perspective in Nature Reviews Physics on how GPUs and heterogeneous architectures are changing software strategy for quantum mechanical simulations.

Scatter plot of the compute used to train notable AI models from 1980 to 2025, on a logarithmic scale and coloured by domain. Training compute climbs by about seventeen orders of magnitude from the CPU computing era to the deep learning era that began around 2012; AlexNet, AlphaFold, DALL-E, GNoME and Gemini 1.0 Ultra are labelled.

Our latest Perspective in Nature Reviews Physics explores the evolving landscape of exascale computing for quantum mechanical simulations.

We examine the challenges and opportunities presented by heterogeneous architectures, particularly the rise of GPUs, and the necessary shift in software strategies.

Read the full Perspective without a subscription, or see the article on Nature Reviews Physics.

About the Author

Dr. Ravindra Shinde
Dr. Ravindra Shinde

I'm Dr. Ravindra Shinde, a computational physicist and quantum chemist. My work spans quantum chemistry, condensed matter physics, AI for materials, and scientific computing: I use and develop electronic-structure methods to understand how electrons shape the properties of molecules and materials, and I build the software that makes those calculations possible.

Continue reading

New writing, by email.

An occasional note when I publish something new: research updates, essays on scientific computing, and thoughts on recent papers.

Your address is used only to send these notes.

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.