Project

General

Profile

Bug #3405

Updated by Szilárd Páll 9 months ago

For at least a year year+ there have been intermittent failures in OpenCL, typically observed on non-PME offloaded runs, often only a single x/f component being off, regularly already at step 0 (see example uploaded case).

It's been reproduced mostly with nbnxn_vsite and dd121 regressiontests on the AMD slave, but it is possibly likely related to position-restraints failures on Intel reported in #2974.

Back