UberCloud will be at ISC 2022 presenting an interesting Industrial HPC use case!
Mitral regurgitation, a condition where the heart’s mitral valve leaflets do not close tightly, causes excessive blood leakage into the heart’s left atrium which can lead to advanced heart failure. Recently, a minimally invasive alternative to open-heart surgery comes from a device called MitraClip, a clip that allows cardiologists to use catheter-based technology to repair the mitral valve.
3DTHoldings in San Diego, contributes to the discovery of devices and therapeutics in the treatment of human disease conditions. In two recent papers they used machine learning to predict important parameters such as left ventricle pressure, volume, and stresses, in a matter of seconds, with high prediction accuracy. They used feedforward and recurrent DL algorithms, TensorFlow and Pytorch.
For this project, a multi-cloud engineering simulation infrastructure has been implemented on Google Cloud Platform (GCP) and Microsoft Azure. To train the neural network, 1000 simulations with Dassault Systèmes' Abaqus have been executed on GCP's preemptible instances within HPC containers on managed Kubernetes clusters (GKE), resulting in 80% cost savings. All management, monitoring, and health checks have been orchestrated from UberCloud’s account on the Azure Cloud performed through SUSE’s Kubernetes management platform Rancher. This project reduced simulation times from 6 – 10 hours per simulation on the end-user's high-end workstation to just a couple of seconds. The entire workflow used more than a million vCPU hours of Google Cloud Spot instances and cost only $20,000.
This work was honored with the HPCwire 2021 Editors’ Choice Award for “Best Use of HPC in the Cloud”.
https://www.isc-hpc.com/agenda.html