No Valid Results in over a week

That would be alright. Cancelled tasks will disappear relatively quickly from your task list. In the future it would always be a good thing to set a somewhat lower work cache before trying out a new type of work (or for new type of hardware). But no harm done. It happens. Nice November for you I wish from this side of the planet.

My two cents (you probably know all the following, but I'll try anyway):

You have nerver mentioned the size of your caches.

I mean the settings in BOINC Manager under "Options",

then under "Computing prefernces ...",

then check what is set under section "Other" on the line "Store at least x" and on the line "Store up to ... x".

Try setting 0.01 for both and then do

"Save" and then under

same "Options" tab click on "Read local prefs file".

Maybe this will stop the wild over-downloadig of tasks.

You can then slowly increase the cache size till you're happy with the amount of downloading.

Have a nice Sunday.

No, it didn't.

Either you didn't have a cache size of 1.0/0.5 at the time the work request was made, Or the estimate was NOT 3 hrs at the time of the work request (but much, much smaller), Or perhaps there is some sort of impact from the fact that you have 2 very dissimilar GPUs installed and BOINC seems confused about which is the better performer.

Actually, you probably could if you properly configured your machine, but that would probably involve putting other projects on hold and I'm not suggesting you do that.

Please understand that the scheduler just responds to the request it is given and (if it can) delivers exactly what was asked for and nothing more.  If you consider 800 tasks to be manifestly excessive, it's up to you to figure out why your client made that excessive request.

Here are a couple more things for you to consider.  Your machine shows as having both an nvidia GT 1030 and an AMD 7800 series - perhaps 7850 or 7870.  Whatever it is, it would be more powerful than a GT 1030.  If you had bothered to check a random task that took 3 hours or if you had even just looked at BOINC Manager to see what app was running, you could have seen thisGamma-ray pulsar binary search #1 on GPUs v1.22 (FGRPopencl-nvidia)which shows that BOINC was using just the much poorer performing nvidia GPU and NOT your much better AMD GPU.

BOINC is supposed to detect and use the most capable GPU when you have a mixture like this.  For some reason it seems to be identifying the wrong one.  This is nothing to do with the Einstein scheduler.  It believes what BOINC tells it.  If you check BOINC configuration options for <use_all_gpus> you will see how to configure BOINC to use both.  There is also an <exclude_gpu> option where you could exclude the weaker GPU if you wished.  You do need to read the documentation before making complaints that are incorrect.

To give you some idea of the performance difference, a HD 7850 GPU can crunch a GRP task in about 20 mins.  That's a hell of a lot better than 3 hrs.  At 20 mins, a 14 day deadline actually represents a total of 1008 tasks which is why (tongue in cheek) I mentioned that it would be possible to get rid of 800 without aborting them.

A final comment about your 1.0/0.5 day combination.  This means that initially BOINC will 'fill up' with 1.5 days worth and then wont ask again until the remainder is less than 1 day.  Then it will take a big drink - back to 1.5 days total.  If something is way too low with task estimates, that big drink could be an excessive number of tasks for a later time when the estimate is corrected. I always think it is safer to pick a suitable size for just the first setting (with the second at zero) which helps avoid any possibility of a 'big drink' event.

Unfortunately, these two of the alternatives I suggested are correct descriptions of possible ways you were sent 800 tasks.  There is nothing "incorrect" about them.  It's up to you to work out if either were applicable to your problem.

What you don't seem to understand is that the scheduler cannot send you work you didn't ask for.  Your client must have asked for that amount of work for some reason.  Period.

It's really in your best interests to work out why this happened so that you don't get caught again in the future.  It could even be something really trivial like missing a decimal point - eg. instead of 1.0/0.5 you might have had 10/0.5.  I'm NOT saying this was the case but I am saying you should stop blaming the scheduler and try to find the real reason.

Ultimately, a full cache setting of 1.5 days should be fine if that's what you'd like to have.  Just set it at 1.5/0 rather than 1.0/0.5 if you want to avoid sudden bursts of downloading new work at odd times.  If you're fiddling with cache sizes, don't make changes in big steps.  Much safer to make a partial change and observe what happens.

The project doesn't cancel tasks - you should abort the excess so that you don't force your 'wingmen' to wait till possibly the new year for replacement tasks (issued after current deadlines expire) to eventually be returned.  If you are allowing the GT 1030 to crunch at ~3hrs a task, you could finish around 8 tasks per day for the next 12 days so say 100 tasks max.  If you have ~800 left, just choose the earliest deadline ~700 tasks and abort them so that they can be reissued ASAP.  If your 'Chrissie Pressie' new GPU is nvidia, and you could install it soonish, you might get through the whole bundle, if you're quick :-).

What sort of GPU were you thinking of getting?  If you quote the model, people would quickly tell you the expected output in tasks per day.