Ignoring request for no GWO2 GPU work units

You might be wondering why you needed to do this in addition to having deselected that GW search.  You would be excused for thinking that not having the search selected should be sufficient on its own.

The explanation (given years ago) is something like the following.  The available tasks for the FGRPB1G search exist in two separate groups.  There is a small 'rapid response' cache of work that the scheduler can immediately access.  There is a second, much larger (and slower) group that the workunit generator keeps topped up when necessary.  If the small cache gets depleted quickly (lots of requests), there is a small time window until another background process wakes up and transfers a new batch from the large second group into the rapid response group.  If you happen to make a work request during that small window, the scheduler will (momentarily) have no available 'preferred' work and will therefore look at your setting for 'non-preferred' work.

By setting that to 'NO', your client will get a 'no work available' message but will then (60 secs later) repeat the request, by which time the rapid cache will have been refilled.  So, if you only ever want work for preferred apps, you really need to have that setting on 'NO'.  At the moment, with the surge in demand, it's probably more common for people to 'hit' that small 'window of opportunity' and be allocated non-preferred work as a result.