Alex Pankratov :
Jun 07, 2019
If they are going to be run one at a time, there's really no limit except perhaps with things getting hard to manage in the UI, because of the list getting too long and some such.
If they are to be run concurrently, then the rule of thumb is that it's best to arrange them so that no two jobs are using the same device at the same time. This can be done with custom queues (see FAQ). That said, NVMe devices can generally be shared because they _are_ designed for massive parallelism.
Once jobs start sharing devices, the performance of each job will degrade, and I suspect that you will see significant _cumulative_ performance drop with as little as 3-4 competing jobs for HDDs and perhaps a bit more than that for SSDs.
The only way to tell for sure is to test.