System suddenly not reserving CPU for GPU tasks

It's really nothing to do with that setting (NNT).  This problem is just a continuation of your previous one.  BOINC is now in 'Panic Mode' because it thinks tasks are going to miss deadlines.

This was created initially when you were trying to encourage new tasks after you had hit the 5GB disk space limit.  You had been increasing the work cache size to no effect.  When you eventually increased the allowed disk space from 5GB to 50GB, you forgot to reduce your excessive work cache first.  You got a very large slab of work, as I mentioned in the previous thread.  At that time you had the full 14 day deadline so no panic mode initially.

Now, several days later, BOINC is starting to think that tasks are going to miss deadlines.  The main thing that has gone wrong is that using all 16 threads on an 8 true core host and having the GPU requirements to contend with as well, has meant that odd CPU tasks have taken way longer than usual.  Take a look at the completed times for GW CPU tasks.  Values range from around 90Ksecs to a high of ~140Ksecs.  You only need one task doing that high value and BOINC will revise the estimate for every task up to that value.  That's probably why BOINC is in panic mode all of a sudden.

The fix doesn't involve NNT.  You've done the first bit by setting the work cache to 0.1 days.  Make sure the second (extra days) setting is 0.0 days.  You currently have (when I checked) 107 GW CPU tasks in progress.  You also have another 47 Gamma-ray CPU tasks.  You just need to suspend enough tasks to allow BOINC to drop out of panic mode.  Pick tasks with the longest deadlines, say 10 GRP tasks and 20 GW tasks and suspend those 30 by highlighting them in BOINC Manager (tasks tab) and clicking the suspend button.  BOINC should immediately drop out of panic mode and everything should settle down.

The next step is to allow BOINC to start reducing the high task estimates that have created the panic.  You will best achieve that by making sure that there are NEVER more than 14 max concurrent CPU tasks.  You should keep running 2 GPU tasks on the RX 580.  One thread to support each one should be quite enough.  If you need to stop more than 14 CPU tasks from running, just start reducing the CPU usage limits from 100% to whatever value (perhaps in the 90-98% region) until there are no more than 14 CPU tasks running.  If you are using website prefs, each time you make a setting change, click update in BOINC Manager for the client to be advised of the change immediately.  If you are using local prefs (in the manager), the change would be immediate.  Just use whatever you normally use.

Each time a task finishes, BOINC should be able to start reducing all the remaining estimates.  In a day or two, if the estimates are lower, you may very well be able to release all the suspended tasks without causing BOINC to panic.

You only have around 180 GPU tasks in progress.  You are going to run out of those long before CPU tasks.  You wont be able to get new GPU tasks whilst any tasks are suspended.  The obvious time to try releasing all remaining CPU tasks from suspension will be when you are down to the last few GPU tasks.  With any luck at all you probably will be able to do that without causing a panic :-).

Ask if anything is not clear.

I didn't say to suspend the "older CPU WUs".  I said specifically, "Pick tasks with the longest deadlines".  They would be the very newest tasks - the ones with the longest remaining time before they expire.  The reason for that is you really want the "older CPU WUs" (which have the shortest deadlines) to be done done first because they are the ones that will be creating BOINCs panic mode.  You need them finished and returned as soon as possible.

Check carefully!!  If you have any short deadline (ie 'older') tasks suspended, then suspend the same number of longest deadline tasks so that you can then safely release the short deadline ones to be crunched next.

Please understand what I wrote.  I said, "If you need to stop more than 14 CPU tasks from running ...".  All I was doing was giving you a quick way to get the number of CPU tasks to NOT exceed 14 - if you found more than 14 still running for some unexplained reason.  Of course, you're not expected to make that particular change if it's not needed.

Please realise that this isn't over yet.  It all depends on BOINC not going back into panic mode.  Only you can see the critical information - the current estimated crunch time (as seen in BOINC Manager) for CPU tasks (both GW and GRP).  If you report those values now, and again at the very end of your day, it will allow me to tell you if there is likely to be a future issue.  There are other things that can be done if necessary.