An OPNG workunit contains multiple sub-jobs with each the equivalent of an OPN1 workunit.
When each sub-job completes the GPU has to tidy up the current workunit and prepare for the next one. The GPU processors have little to do during the changeover so the temperature.drops.
I noticed this behaviour on my lowly GT1030, so I amended my app_config.xml file to run two OPNG workunits in parallel. The relevant part of the file is:
<app_config>
<app>
<name>opng</name>
<fraction_done_exact/>
<report_results_immediately/>
<gpu_versions>
<gpu_usage>0.50</gpu_usage>
<cpu_usage>0.89</cpu_usage>
</gpu_versions>
</app>
</app_config>

I believe a RTX 2070 is much more powerful so consider gpu_usage of 0.25 to run 4 workunits in parallel.

The cpu_usage number of 0.89 is taken from boincmgr and is the amount of a CPU which is required to support each running GPU workunit.
I assume this changes for each combination GPU model and CPU model.

None for me currently and few in last 18+ hours.
https://wuprop.boinc-af.org/active_projects.py
Shows half the number in progress than yesterday.

Paul.

Windows 10, Intel HD 4600 & Nvidia GT 740M,
crunching OPN

Intel is fine, Nvidia has errors, often immediately, sometimes only at 50%.

Set DWORD TdrLevel 0 in registry.
I'm open to further registry mangling etc. until it works, but am not so experienced on WIndows.

Anymore suggestions to make that NVIDIA work?