I started running Einstein@Home jobs on an old Dell computer about a month ago. The machine has been running Gamma-Ray Pulsar Search tasks in the CPUs and Binary Radio Pulsar Searches in the Intel GPU. I have noticed that typically about 4 to 6 of the latter tasks result in validate errors each day, while most of these tasks complete normally and are validated. Is this normal?
Copyright © 2024 Einstein@Home. All rights reserved.
It's probably driver related,
)
It's probably driver related, and depending on your CPU there may or may not be a driver that works reliably. I've been told the issue is that a lot of Intel's GPU compute drivers accumulate rounding errors at a higher rate than AMD or NVidia's drivers do, with the result that their results don't reliably validate with anything else. The only intel GPU I compute on is an i5-6200 laptop; and it runs SETI because the last time I'd asked around they were the only project to have an app that didn't have a significant percentage of validation errors. (SETI apparently had to write a different compute path for the intel GPU that is slower but reduces the error accumulation rate.)