There are enough tasks.

29-Apr-2021 09:26:03 [World Community Grid] This computer has reached a limit on tasks in progress
29-Apr-2021 09:26:03 [World Community Grid] Project has no tasks available

I woke up a bit early this morning, and I'm seeing that my relatively modest rig, with app_config.xml set up for two tasks, is just crushing it. It's just about 8 AM UTC, and it's already over 115,000 BOINC credits on the day. So yes, if your GPU can hack it and you have CPU slots to spare, running extra GPU tasks is a big win. It does play silly buggers with time estimates, but filling in those CPU-mode gaps is well worth it.

Of course, if you've broken the bank for an RTX3080 and a Threadripper, you might want to run more than two tasks.

@nanoprobe - How many consecutive wu's are you running with your AMD cards?
I'm currently only running 1 wu at a time with my R9 280x.

I've done a fair bit of trolling through the stats and about 98% of those stats I looked at were NVIDIA cards.
Curious to hear what other Radeon cards are being used and the results seen.

There seems to be continual confusion over the CPU parameter in app_config.xml. It is NOT a limit on how much CPU a task can use. Each OpenCL task will use up to 1 CPU regardless.

The CPU parameter influences the BOINC scheduler, when determining how many tasks of which type to start concurrently. So for the case where your OS/driver/GPU always uses 100% CPU (e.g. linux/NVidia) you will want to set CPU to 1 or even higher.

As an example: On a 4c/8t Intel system with an RTX2060, I find that 3 concurrent GPU tasks seems optimal (more gives higher overall GPU utilisation but lower overall throughput - obviously this depends on many things such as CPU model, bandwidth on the motherboard, disk bandwidth, etc). And I get better overall GPU throughput when limiting to the number of cores rather than threads. But for CPU-only tasks I get better total throughput by using all the threads, even though each individual task runs slower. So, I set overall compute capacity to 90%, and for OPNG GPU=0.33, CPU=2. When there are no GPU tasks 7 threads run CPU tasks. When there are GPU tasks available, it runs 3 GPU + 1 CPU task concurrently. (If you don't have hyperthreading you need a slightly different strategy).

But this illustrates how you need to set the CPU value according to how your system behaves. Sometimes less than 1 is appropriate, and sometimes greater than 1 is. Setting it too low merely introduces undesirable CPU contention as you are telling BOINC to overcommit the CPUs.

thanks for clarification