I have never seen that. It would seem that the betas are leaving something in memory. Maybe if you disable "leave applications in memory" it would help.

I think this might have nailed it if the issue was unwillingness to run tasks for other projects (although without more details about the "tinkering" we can't be sure!)

I usually have a mix of MilkyWay@home and Einstein@home (hello Tom M!) and these Betas came in, tended to dominate for the couple of hours it took to get them out of the way on my NVIDIA GPUs (neither of them "top of the range") then normality was restored until another batch of Betas came along. Admittedly, I'm not using AMD, and I'm not doing anything exotic such as trying to run multiple tasks on a GPU at the same time (which may or may not have been the case here - when I see "tinkering" I start to worry )

@Tom M - See the above! I presume you were only trying to run one at a time, though -- if you have been playing with multiple tasks per GPU, that's a different matter...
By the way, if/when you have issues, I'd suggest you locate the latest Beta thread that isn't newer than when you ran the jobs and post in there! (Of course, if another Beta has started since you had the problems, the best thing to do is to simply "bite the bullet", re-test, and see if the problem still exists!) For the OPN GPU betas the WCG tech (Uplinger) has been quite willing to keep us in the loop...

Cheers - Al.

P.S. It'll be fun in here if/when some of the Seti Orphans decide to bring their big GPUs in here -- I wonder if WCG will be able to fill the work queues fast enough for the demand?!?

[Edited a typo that I spotted...]

Was this in Windows or Linux?

Way back when WCG was running HCC on GPUs, the GPU tasks did not oversubscribe CPU cores/virtual cores. They just occupied a normal CPU thread slot. Furthermore, multiple GPU tasks would automatically run simultaneously, starting in the normal order of tasks displayed by BOINC Manager as "Ready to start". I ran 4 tasks in a 2GB ATI HD7870, fed by a Q9650 C2Q CPU @ 3.8GHz, which was a suitable match for the GPU.
There was another phenomenon though: If multiple GPU tasks started simultaneously, the GPU really bogged down. Every few hours I had to stop starting any new tasks and then restart them one at a time, spreading the startup times evenly though 1 WU run time.

This soon became very tedious, and if it turns out to happen with OPN1-GPU, WCG should try to make the startups spread out automatically.

So summarizing if this takes off it seems best to lower the allowed CPU usage, like in my case set it to 68% to actually still use 75%, also I can see the GPU tasks uses little over 300 MB of memory (the others about 100MB), the upside is if what was claimed is true and the jobs are the same, it can process them in about 30 minutes compared to over 2 hours on the CPU cores [i9-9900K]. Also since the Intel GPU in my case sat unused it's the first time any processing is actually going on there.