fah-client 8.4.9: HTTP_SERVICE_UNAVAILABLE ("No appropriate assignment")?

[malkie]

...

I do not know why CPU folding is so unpopular, never really thought about it. Now that you mention it, a viable solution would be to increase the points awarded for CPU projects, that would get people folding on them (assuming those same WUs cannot be processed on GPUs).

...

I have to admit that I'm curious about the idea that increasing the points for CPU WUs would make people more likely to permit them on their machines. Is this a really competitive program? Are people really "driven" by points?

btw, the first time I allocated CPUs to FAH I found that the response time of my machine dropped noticeably (it has only a Ryzen 5 2400), but when I tried again after finding little or no GPU-only work, it no longer seems to have that issue.

I'm not saying that increased points would be a bad thing - who am I to judge - only that it surprises me that it might be effective. I started off in machine sharing back in the early days of SETI@Home, leaving my computer on permanently (well, as permanently as anything could be running Windows95) to give the project my idle time. I don't remember if there were "points", but if there were, I didn't care.

With FAH I care only to the extent that every so often I amuse my wife by telling her where I am on the leader board - currently 132,160. (I wanted to put an exclamation at the end of that sentence, but I know that some wag would read it as "factorial", and I hate to think of how low on the list that would put me )

So just how cutthroat is the competition here? Should I bump my machine up to all but one of its 8 CPUs to try to get ahead of whomever is at 132,159?

[enroscado]

I imagine points is by no means the main driving factor, but I would certainly not discard the possibility that it is at least and incentive / motivator. It is the only explanation I can find as to why there are WAY more CPU projects in the queue than GPU projects, after all, every single computer with a GPU must have a CPU to function, but not the other way around. Still, most people choose to just fold with GPUs due to (I think) more bang for the buck.

[malkie]

I imagine points is by no means the main driving factor, but I would certainly not discard the possibility that it is at least and incentive / motivator. It is the only explanation I can find as to why there are WAY more CPU projects in the queue than GPU projects, after all, every single computer with a GPU must have a CPU to function, but not the other way around. Still, most people choose to just fold with GPUs due to (I think) more bang for the buck.

Thanks. My wife just told me that she thought my question was naïve - of course points are an incentive; and competition, if short of fierce, is a notable factor (subtext: we're talking about geeks here )

[muziqaz]

The biggest reason why there are so many more CPU WUs, than GPU:
Because GPUs are so much faster than CPUs, their WUs can be built much longer (bigger). So let's say a researcher wants to get back 10fs worth of simulations. They create a single WU, which can be run on a GPU, which does 10fs worth of simulations. GPU will finish that WU in 1 day.
Now, their WU cannot be run on a GPU due to the type of simulation, which is not supported by GPU arch. If researcher keeps that WU same size (10fs), a CPU would be able to finish it in let's say 100 days. That is impractical. So researcher has to split that 10fs WU into 100 smaller WUs, each of them would take 1 day to finish with a CPU.
So what GPU can do in a day through one WU, CPU would need 100 WUs, each taking one day.
Researcher can get needed data with 1000 WUs run on GPUs
Researcher can get that same data with 100 000 WUs if they could be only run on a CPUs

Hope this make sense

[appepi]

My own perspective on the CPU/GPU business is that if there were types of modelling that actually NEEDED the fancier capabilities of CPU's and could not be done at all with GPUs, then I'd be happy to run those CPU WUs alongside the GPU ones. But, since it seems that they are just alternatives to do the same work, then it is a matter of efficient use of resources.

It happens that my newer workstations have GPUs and CPUs that use about the same power: eg Xeon E-1650 v3 (140W) / RTX 2060 (160W); Xeon W-2133 or W-2145 (140W) /GTX1080 (180W) or 1070 (150W). To run all of them 24/365 would cost far more than I can afford to spend on a hobby activity. So we are talking return on investment here.

150 watts of electricity costs the same whether it is powering a CPU or GPU, and whether it is being used in what would otherwise be idle time (but also idle power use) or in a device dedicated to folding and not much else. So, since the GPU does a lot more modelling per hour than a CPU, and the allocated points reflect this, it is simply a better return on investment to run GPUs and relegate the CPU (and the rest of the box) to a support role. And this is true whether you consider the return as points or science, and the investment as electricity, money, or contribution to global warming.

[calxalot]

I believe most or all CPU WUs cannot be done on GPU. Some definitely can’t.

muziqaz’s post might have been confusing on that.

[muziqaz]

I believe most or all CPU WUs cannot be done on GPU. Some definitely can’t.

muziqaz’s post might have been confusing on that.

Yes, whatever is created for CPUs, cannot be run on GPUs. I didn't mention it, because that wasn't the topic of conversation
Researcher, with limited grant money and deadlines to achieve, wouldn't just forgo order of magnitude faster methods of computation just to have a crack at CPU folding

[appepi]

Interesting. First a quick check with https://www.etymonline.com/word/fungible , which tells me that a fungible things is "capable of being used in place of another ...".

So it seems from the advice above that I can safely conclude that CPU and GPU WUs are non-fungible these days.

Yet it would seem that the points associated with them are treated as fungible, as they are used, for example, in the ATTO ecosystem.

And it also seems from another thread ( viewtopic.php?t=43170 ) that the account of the way points are determined (in the FAQ section of the on the FaH website) is now outdated and wrong. Certainly the FAQ answer states explicitly that the same WU could be run on a "benchmark" CPU as well as on GPUs, and that the aim was to ensure "equal pay for equal work". Presumably the more complex processes currently used have the same aim, but (to me at least) they are a black box. I think the key question here is whether or not the current ways of allocating points leads to "equal pay for equal work".

After some thought, I am inclined to think that the kilowatt-hours per WU when run on suitably modern benchmark devices might be the most fungible version of "equal work" for points allocation. The word "modern" above presumes "more efficient" and thus helps the planet.

[enroscado]

I believe most or all CPU WUs cannot be done on GPU. Some definitely can’t.

muziqaz’s post might have been confusing on that.

Yes, whatever is created for CPUs, cannot be run on GPUs. I didn't mention it, because that wasn't the topic of conversation
Researcher, with limited grant money and deadlines to achieve, wouldn't just forgo order of magnitude faster methods of computation just to have a crack at CPU folding

Thanks for that clarification.

Then with the following in mind:
- CPU WUs cannot run on GPUs
- There are far more CPU WUs in the queue than GPU WUs

I still reach the same conclusion: To process / fold CPU WUs there should be an "incentive" for donors / contributors. Otherwise, the imbalance on CPUs/GPUs cycles being donated will remain as is: CPU WUs don't get folded as much or as needed.

[muziqaz]

Incentive is helping science and researchers who are stuck with simulations which cannot be done with GPUs. That's all there is to it.
If you need other incentives, FAH will not provide them for you, and it is OK for you to chose not to fold on your CPU

[enroscado]

Incentive is helping science and researchers who are stuck with simulations which cannot be done with GPUs. That's all there is to it.
If you need other incentives, FAH will not provide them for you, and it is OK for you to chose not to fold on your CPU

I completely get that, and agree with you. But I also think you are missing my point.

Evidently, "competition" (so to speak) is used as an incentive to fold, hence the points rewarded to each donor. If this wasn't the case, there would be no points at all in the stats, there would just be Work Units folded.

Example: I fold a WU (project 18255) on one of my 5090s in ~17 minutes (~58,000,000 PPD). The same WU (project 18255) on one of my 4070Ti takes ~60 minutes (~9,000,000 PPD). They are both 1 WU, and they are both very similar WUs. However, the points rewarded are very different, because "science" benefits from getting it back sooner, therefore I am rewarded (in points) accordingly.

Elaborating from that, if "science" needs CPU WUs to be folded, but are not being folded, clearly there needs to be an incentive to do so. One possibility would be to increase the points awarded to CPU WUs folded/completed.

In any case, it is just a thought exercise...

[muziqaz]

It's hard to miss the point when I was the one who suggested new point system for CPUs and got refused that plan
All of us here completely understand everyone's issues

[Joe_H]

Yes, updating the points system has been suggested for many years. It often has come down to that they know the current one has problems but haven't come up with a plan that wouldn't make things worse by ticking off one group or another.

Personally I fold almost exclusively CPU WUs as I mostly use macOS systems. My son occasionally folds on his Windows system for my username, etc. and may contribute some GPU work. But the last time was when his room was cold for a couple days back in November until I could get someone to fix the furnace.

[enroscado]

the last time was when his room was cold for a couple days back in November until I could get someone to fix the furnace.