There are plenty subprojects where GPUs can't do them, so why aren't the CPUs left to do those? Why allow CPUs to very slowly do something that a GPU could do in a fraction of the time?

I don't think that anyone answered the actual question. I read the question as "if both GPU and CPU can participate in a project and the GPU is xxx faster than the CPU, why support a CPU worker for that project?".

I see two reasons:

1) A user wants to participate in a project because they are interested in it.
2) A user wants a complete set of badges.

PrimeGrid supports because "if someone wants to walk from Atlanta to Boston, we won't stop them, but it will them a lot longer time to do that than to take a plane".

Well, we*re talking about a project with a scientific goal.
There are people like me who support just about any project that pops up in their known Boinc universe because they want to further science, start to like challenges here and there and throw one machine or a lot of machines at them. Others support one project because it's just their thing, full stop!
If you want your support to be as efficient as possible, you research the projects and their applications, tweak some config files, set different resources for the projects and do whatever you think helps the efficiency.
For me it's a few machines, and my everyday machine is older and slower than those taking part in Boinc.
For people who run Boinc on their personal worker, media player, surfing machine, or what else they are doing with their PET (Personal Electronic Thing, nice labelling by Douglas Adams), the above-mentioned efficiency might not be the best - because it tampers with the PET's responsiveness.
Tweaking everything possible is fine with me, as I use another machine for myself, but imagine someone running Boinc on their PET:
It's nice to have an i7-8700 that runs LLR on all CPU threads (best efficiency for me) and Genefer 21 on the GPU just to push PrimeGrid, while everything else is awfully slow. Oh, wait - that's not nice. Instead it's nice to run Genefer 22 on the GPU (less impact than 21) plus something that does not kill the overall responsiveness as much as LLR. And this something might be a sieve. PrimeGrid had CPU-only sieves, but at the very moment there are none, as there are no sieving goals that can only be achieved without GPU.
Btw.: Sieving is not only better regarding responsiveness on high-end machines - it's also less taxing for the system. So, another use case might be someone who administers a data center or cluster that they want to be doing work so the computers and their cooling run at around the same level all the time to reduce stress on the hardware. Nice job for Boinc and a PrimeGrid sieve, even if it might be done faster by GPUs.

So, there are different users that might want to support PrimeGrid with their own, totally different goals in mind.
____________
Greetings, Jens

92914140^65536+1

I ran GFN-22 on my CPU. It took 29.4 days but my computer loved every second of it. I only wish I had talked to Michael earlier and learned that my mutithreader wasn't multithreading....
____________
Werinbert is not prime... or PRPnet keeps telling me so.
Badge score: 1x1 + 12x3 + 1x4 + 1x5 + 1x6 + 2x7 + 1x8 + 1x10 = 84

I ran GFN-22 on my CPU. It took 29.4 days but my computer loved every second of it. I only wish I had talked to Michael earlier and learned that my mutithreader wasn't multithreading....

IIUC, you are suggesting that PrimeGrid force end-users to only participate in projects that their computer is best suited for.

It doesn't hurt PrimeGrid in any way if a CPU works on a task that is better suited to a GPU. I think of it this way. If the work done by a project is done 90% GPU and 10% CPU, then you have removed 10% of the computing power from that project. That actually hurts that project instead of helping it.

Conversely one could say that the 10% of CPU could be of better use on a CPU only project. I would agree with that, but what if the participant doesn't care about the CPU only project? What if the participant is only participating in a PrimeGrid project because that project is the only one hosted by PrimeGrid that is of interest to them. Weiferich and Wall-Sun-Sun is a good example of that. It isn't a prime searching project like the others. I would not be surprised if there are participants that are only interested in those projects and none of the others and those participants only have a CPU at their disposal. Why would we want to exclude them from working on the project?

Primegrid has so many subprojects that surely it's better for all the Primegrid users with CPUs to work on the stuff that GPUs cannot.

You're correct -- or at least you would be correct if PrimeGrid owned all of those CPUs.

But we don't own them; you do.

What's best for PrimeGrid is to keep PrimeGrid's users -- who are donating their hardware, their electricity, their personal effort, and even their money -- happy and motivated. We do that in lots of ways, and a part of that is giving people the option to run whatever they want. I'd rather have computers running less than optimal tasks (from out perspective) than have them not running anything at all.

It seems that you are assuming that CPUs excluded from GPU sub-projects would naturally be used for CPU-sub-projects at PrimeGrid. Perhaps they would just move to other projects all-together. CPUs on GPU sub-projects at PrimeGrid is better than no CPUs at PrimeGrid at all.

Personally, I want to be able to *also* run CPUs on GPU sub-projects, in order to achieve hour-based goals on wuprop.

Many people set and forget. They do not read all the information, they just think it is neat to run, then let it roll. They do not care about setting stuff up. This would deplete the project if things were not set balanced.
Those who want to do better on the project work it out and do what they feel their goal is. The project does not want to discourage personal goals.
For example, myself. I at one point had many systems running with some good hardware, life changes required me to slow down. I now have old hardware, want my hardware for certain abilities, but also want to help the project. So I am choosing to run things slower, and gaining my personal goals, without locking me out of using my hardware as I want.
As with life, inclusion makes it easier for most.
____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1

Many people set and forget. They do not read all the information, they just think it is neat to run, then let it roll. They do not care about setting stuff up. This would deplete the project if things were not set balanced.

Those who want to do better on the project work it out and do what they feel their goal is. The project does not want to discourage personal goals.
For example, myself. I at one point had many systems running with some good hardware, life changes required me to slow down. I now have old hardware, want my hardware for certain abilities, but also want to help the project. So I am choosing to run things slower, and gaining my personal goals, without locking me out of using my hardware as I want.
As with life, inclusion makes it easier for most.

Half the subprojects on here are CPU-only, surely there's at least one you want to do?

I have two GPUs, currently running on other projects that don't do CPU apps. I have many CPUs, I use them here on the GPU/CPU apps because I don't have a GPU available to run those apps faster.

My priorities are not necessarily the same as other people's. I am also into WU_Prop.....I want hours. So for me it is actually better to run the GPU/CPU apps on my CPUs. Yes, it is not as "efficient" for the project but at least it is tasks being crunched.
____________
Werinbert is not prime... or PRPnet keeps telling me so.
Badge score: 1x1 + 12x3 + 1x4 + 1x5 + 1x6 + 2x7 + 1x8 + 1x10 = 84

But surely all users think like I do, they want to do as much useful work as possible. I see no point in wasting my CPU time on something that my GPU (or someone else's GPU if I didn't have one) could do in a fraction of the time. Perhaps what you could do is have the default for new users as all the GPU projects unticked for CPU, but let the user re-tick them if they really want to. There could even be a note saying "this subproject runs more efficiently on a GPU, please consider running your CPU on the subprojects that need it."

But surely all users think like I do...

(Trying this again without joking about politics. My bad, sorry...)

I can't even agree with myself most of the time, let alone alone anyone else! :)

But surely all users think like I do, they want to do as much useful work as possible...

Actually, there is literally decades of social and behavioral scientific work (i.e., sociology, psychology, economics, etc.) that describes in detail why human behavior frequently follows patterns that are far from optimal. And perhaps more importantly, an abundance of literature that notes how humans often "don't want to" operate optimally even when they are told how to do so.

PG has long held to a principle of having the most people have the most access to the most projects possible. I think we should continue to do so.

Actually, there is literally decades of social and behavioral scientific work (i.e., sociology, psychology, economics, etc.) that describes in detail why human behavior frequently follows patterns that are far from optimal. And perhaps more importantly, an abundance of literature that notes how humans often "don't want to" operate optimally even when they are told how to do so.

Because we are living beings. If you operate "optimally" then everyone would operate in the same way. With such behaviour, mankind would die in facing an unforeseen event.
Doing anything is a basis for the survival of the group.
You can define an optimal process if all assumptions are known. But this doesn't exist in real life.
... this is not off-topic: that's why the largest number of capabilities for PrimeGrid is better for its future Q.E.D. :o)

What's considered "optimal" may differ depending on the user's goals, even if the hardware is otherwise same.

For example, if you want boinc credit/time, you will run whatever gives the best in that metric. If someone is badge hunting, they might want to put all possible resources on getting there faster, especially if it is a project that is about to end. I'd still consider myself primarily interested in getting the most possible prime number discoveries. For projects still using LLR (not LLR2) then I will trade off maximum throughput for a less efficient but shorter runtime where it makes sense to do so.

I think the balance here is right. Give the user the all the possible options that aren't insane (work will finish before universe ends). Then it is up to the user to do what they want. What is best for one person is not best for everyone. Also I'm not sure what the new user default project selection is now, but I do recall they were chosen to be the "safer" option in terms of hardware stress. So that's another factor.

My personal resources are heavily biased towards CPU running. I wont often put them on GPU projects, but I like to have that option open to me in case something happens to make that something I want to do.

A basic question behind the premise is "how many users use the default configuration and never change".

I don't know the answer to that question. If any user changes the default configuration, I don't think it is too much to ask them to determine which subprojects they should participate in so that they benefit PrimeGrid the most.

I would certainly not kick users off subprojects because their CPU/GPU cycles would be better off on a different subproject.

Half the subprojects on here are CPU-only, surely there's at least one you want to do?

Actually NO!
When you have been around this project as long as I have, I want to do what I want, if I can. Since I can, I do. I know I am not always the first, but at this juncture I do not care. I work on things like badges and then a separate "status" on a different project that ties into all projects. Old farts like me, who have worked with computers since the 1970s, have some idea of what they want to do.

But surely all users think like I do, they want to do as much useful work as possible. I see no point in wasting my CPU time on something that my GPU (or someone else's GPU if I didn't have one) could do in a fraction of the time. Perhaps what you could do is have the default for new users as all the GPU projects unticked for CPU, but let the user re-tick them if they really want to. There could even be a note saying "this subproject runs more efficiently on a GPU, please consider running your CPU on the subprojects that need it."

I think you are missing something in the big picture...sometimes a particular pc for whatever reason just won't crunch a Project without giving continous errors while using the cpu's and being able to come to PrimeGrid and see that everything just runs great is a huge advantage to the user. ie my pc can crunch somewhere doing something needed albeit a bit slower than is ideal but at least the pc is contributing instead of sitting idle AND I know it's NOT the pc that's the problem.

Another example is the RPi people running Einstein, it's one of the few projects where you can use them so they do, it's not very efficient or fast but the work gets done.

But surely all users think like I do...

(Trying this again without joking about politics. My bad, sorry...)

I can't even agree with myself most of the time, let alone alone anyone else! :)

Since you asked politely unlike some moderators I could mention (not this forum), I'll re-write mine too.

Do you find that you disagree with your opinion of last year so much that you call people with that opinion you used to have a fool? I know I do.

If you're asking if I change my mind, sure. I don't believe I know everything there is to know, and I'm always looking to educate myself further. So of course I change my opinion.

As for thinking people are fools for not agreeing with me, I've been around the block a few times. Usually when I think someone else is a moron because their opinion is different than mine, it turns out they know something I don't. :)

Actually, could all Primegrid subprojects be made to work on GPUs? (But it's only been done for those really needing more power). If so, then my original question is irrelevant.

But surely all users think like I do, they want to do as much useful work as possible. I see no point in wasting my CPU time on something that my GPU (or someone else's GPU if I didn't have one) could do in a fraction of the time.

Good morning.

If you can't follow this then I'm at a loss on how to educate you further.