Yes, set flags you don't know what they do.
You cannot use a GPU application on a project that doesn't have a GPU application. It's not you who decides to use the GPU, but it's the project who decides if they release an application that can do the science on the GPU. So setting use_all_pus won't work. You can't magically make things work through the GPU.

<ncpus> is a test flag with which you set a hypothetical number of CPUs or cores and run that on the one CPU you have. If your CPU doesn't have 8 cores, but a lot less, you're now overworking the CPU.
<use_all_gpus> is only used when you have two or more same brand but different model GPUs and you want both/all of them to do work... if there are projects that have work for that brand GPU, that is.

---

<ncpus> is a test flag with which you set a hypothetical number of CPUs or cores and run that on the one CPU you have. If your CPU doesn't have 8 cores, but a lot less, you're now overworking the CPU

And having tried this flag out of interest a couple of years ago, at least on CPDN tasks it tends to lead to crashes. I didn't spend enough time on it to work out whether that was because I didn't have enough memory or for some other reason.There certainly wasn't enough information in the task error message to determine exactly what the problem was.

running 32 tasks and they are working at 20% per hour. [/img]

World Community Grid (application)
Computer: Windows 10.
CPU: Pentium G3420 (2 Core).
tasks on World Community Grid (application):
Core 1
Task 1 (cpu0.25)
Task 2 (cpu0.25)
Task 3 (cpu0.25)
Task 4 (cpu0.25)
Core 2
Task 5 (cpu0.25)
Task 6 (cpu0.25)
Task 7 (cpu0.25)
Task 8 (cpu0.25)

By the look of this you're wanting to run each core at 25%, rather than running on 25% of the available cores?

Under Linux, you would use cgroups*. Windows 10 runs Linux so that may work, but I haven't tested it because I run Windows 10 in Linux, and won't be changing that soon.

There's probably some awful third-party app you can use in Windows, but that's for you to discover.

* Article is fairly old and refers to cgroups v1, however the principle for cgroups v2 is exactly the same.