When running the WW subproject on AMD GPUs, you will notice the GPU utilization (GPU-Z) is spike-idle-spike-idle-etc. Utilization stinks.

Add something like this to an app_config.xml:

Code:
<app_version>
  <app_name>ww</app_name>
  <plan_class>opencl_ati_WW</plan_class>
  <cmdline>-ws 50</cmdline>
  <avg_ncpus>0.0001</avg_ncpus>
  <ngpus>1</ngpus>
</app_version>
The key bit is the cmdline value. Default without an app_config.xml is 11, I think. The range is 1-100.

Changing it from the app default to 50 with my AMD VII, task run times dropped from 1200 to 985 seconds. For my RX580, 4200 seconds to 3700 seconds.

Fiddle with the values until the utilization graph is maxed out solid. Going higher than the value that maxes out utilization does not seem to improve performance any further. The higher the value, the more lag you may experience if you are using your machines for other stuff. Mine are dedicated crunchers, so I really don't care about lag.

Apparently increasing this value for an Nvidia GPU makes no difference. But I haven't tried that out myself.