Neither worked.
Having a go with unverified success - but in an hour I'll know more.

This has been cold start tested
and it works for me -

app_config.xml for Einstein@home:

<app_config>

  <app_version>
    <app_name>einstein_O3AS</app_name>
    <plan_class>GW-opencl-nvidia-2</plan_class>
    <avg_ncpus>1</avg_ncpus>
  </app_version>

</app_config>

6 CPUs

Einstein@home   Running                                 Binary Radio Pulsar Search (Arecibo,GBT,long) 1.33 (BRP4X64)
Einstein@home   Running                                 Binary Radio Pulsar Search (Arecibo,GBT,long) 1.33 (BRP4X64)
Einstein@home   Running                                 Binary Radio Pulsar Search (Arecibo,GBT,long) 1.33 (BRP4X64)
Einstein@home   Running                                 Binary Radio Pulsar Search (Arecibo,GBT,long) 1.33 (BRP4X64)
Einstein@home   Running                                 Gamma-ray pulsar search #5 1.08 (FGRPSSE)
Einstein@home   Running (1 CPUs + 1 NVIDIA GPU)         All-Sky Gravitational Wave search on O3 1.07 (GW-opencl-nvidia-2)

I'm certain that setup is more efficient now. Thank you very much Richard. I could never have done it without your help and encouragement.

Now, because I've reduced the number of days' work I request in order to stop missing a deadline that in turn causes an override of my CPUs preferences by the scheduler to try to get it done in time, I doubt I'll get any long LHC jobs now.
Unless I manually stop Einstein from getting new tasks now and again. But that's OK.

Thank you. Thank you.

Thank you. Glad it all worked out in the end.