Why do WU's take longer on many cores, than multiple WU's on a few cores??

[Ichai]

I have a Ryzen 2700X and a RX 580. In both scenarios i was also running one WU on my GPU as well
My setup before was 4 WU's with 3 cores each, these took between 4-8 hours to complete.
My setup now is 1 WU with 12 cores, and these take anywhere from 12-20 hours to complete.

How come that the time it takes to complete INCREASES with additional cores???
shouldn't it be the other way around?
I'm quite confused, can someone please clarify for me?

edit: by cores i meant threads, 3 and 12 Threads respectively

[Joe_H]

WUs are not all the same size, you can only compare run times between WUs from the same project. You will get WUs from various projects, and which projects will also depend on the number of CPU threads you have set when the client requests a new WU. There are larger projects that only get assigned to systems offering use of 4 or more CPU threads including some that require 12 or more.

Within a specific project you may see some variation in processing time, usually less than +/-5%, between WUs. A WU from that project will take a bit over half the time processed on twice the number threads as another WU processed at the lower number of threads. There is some inter-thread processing overhead involved as the thread count increases. With projects that are simulating a smaller number of atoms this does result in the overhead eventually negating the benefit of adding more CPU threads. These projects may have a maximum thread count to which they will be assigned.

As for your Ryzen 2700X, it has 8 cores, 8 FPUs, and 16 threads supported. As CPU folding is floating point intensive, using more than 8 threads will not see the same decreases in processing time as you would see going from say 6 to 8. If it was my system I would just set it for one CPU slot for 8 threads and a GPU slot for the RX 580.

[BobWilliams757]

Ichai,

If you monitor your GPU work units you will find quite a bit of variation in time as well. Though the assignments work differently than with CPU's, there are still impacted by overall work unit size as well as your hardware capabilities within their assigned "species". I have had work units less than two hours to complete on my GPU, and some over 8 hours. Though the variations are not huge, at the extreme ends of the assigned work units they can be several hundred percent.

Over time the work units available to your hardware change, and the trends in time to process vary. Overall it seems to me that more CPU work units are getting larger with most settings I've tried with my 8 thread processor. But that might only be short term, and is always subject to change.