Hi folks;
I'm here to plea, probably not for the forums first time, for standardized points per hour for any given type of video card. Right now I have 2 identical cards folding 2 different tasks. One earning 3.23 Mill ppd and the other 5.07 mill ppd. To put it nicely ...nonsense. The same card. The same opportunity costs for you economics types. Very similar heat outputs. The reward you give me costs you nothing. So why are the rewards per hour different? Different rewards encourage cherry picking of tasks. You have target rates given to you here..https://folding.lar.systems/gpu_ppd/overall_ranks. Please consider standardizing rates for video cards. Thanks for listening.
Badsinger
Edited for spelling
A plea thats probably been made before.
Moderators: Site Moderators, FAHC Science Team
-
- Posts: 100
- Joined: Wed Jan 05, 2022 1:06 am
- Hardware configuration: 4080 / 12700F, 3090Ti/12900KS, 3090/12900K, 3090/10940X, 3080Ti/12700K, 3080Ti/9900X, 3080Ti/9900X
Re: A plea thats probably been made before.
What other factors should be at play?
What if I am using the machine to play games on at same time?
What if one card is in a more air cooling effective case and the other isn’t? What if one has had every single wire moved out of the airflow path and the other is a tangled mess of air disrupting wires?
What if I never blow the dust out of the heat sinks?
What if I replace my thermal paste and tape every year?
What if I change to highly effective cooling for the card?
What if I buy the golden binned version of a GPU, and you buy just the average bin version?
As you say they are only pretend points anyway. It is the loosest of competitions with made up points, and I think the same made up points arguments hold here too. Reward folks who know how and who do (if it is important to them) maximize their cards performance against those of us (like me) who just buy more cards.
I have a few “duplicate” cards… I should look at them over time and see how much they vary. Figure out if anything I said above would make a difference.
What if I am using the machine to play games on at same time?
What if one card is in a more air cooling effective case and the other isn’t? What if one has had every single wire moved out of the airflow path and the other is a tangled mess of air disrupting wires?
What if I never blow the dust out of the heat sinks?
What if I replace my thermal paste and tape every year?
What if I change to highly effective cooling for the card?
What if I buy the golden binned version of a GPU, and you buy just the average bin version?
As you say they are only pretend points anyway. It is the loosest of competitions with made up points, and I think the same made up points arguments hold here too. Reward folks who know how and who do (if it is important to them) maximize their cards performance against those of us (like me) who just buy more cards.
I have a few “duplicate” cards… I should look at them over time and see how much they vary. Figure out if anything I said above would make a difference.
-
- Site Admin
- Posts: 7922
- Joined: Tue Apr 21, 2009 4:41 pm
- Hardware configuration: Mac Pro 2.8 quad 12 GB smp4
MacBook Pro 2.9 i7 8 GB smp2 - Location: W. MA
Re: A plea thats probably been made before.
If the projects were all very similar, then it might be possible to keep the points with bonuses that close. But they aren't. They could pick a specific card and have that be the standard, that would just move the variance in points to other cards.
This is one of the times I wish a blog post by Dr. Pande had not been lost. He posted the graphical results of points with bonuses included for a range of different projects over differing processing times, i.e. more to less powerful GPUs, that was done by one of the grad students in his group. The curves would tend to coincide for some range that would be mid-range cards, but went all over the place for the the highest and lowest power cards.
Now you can compare cards by base points. There you should see similar results for PPD between projects for the same cards. But the exponential nature of the bonus points exaggerate any speed differences.
As for the Lar.system's numbers there have been many problems with that database. As a coarse indicator of GPU performance it works, but there is no consistent requirements for submissions that would make its numbers useful as benchmarks.
This is one of the times I wish a blog post by Dr. Pande had not been lost. He posted the graphical results of points with bonuses included for a range of different projects over differing processing times, i.e. more to less powerful GPUs, that was done by one of the grad students in his group. The curves would tend to coincide for some range that would be mid-range cards, but went all over the place for the the highest and lowest power cards.
Now you can compare cards by base points. There you should see similar results for PPD between projects for the same cards. But the exponential nature of the bonus points exaggerate any speed differences.
As for the Lar.system's numbers there have been many problems with that database. As a coarse indicator of GPU performance it works, but there is no consistent requirements for submissions that would make its numbers useful as benchmarks.
iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
-
- Posts: 513
- Joined: Fri Apr 03, 2020 2:22 pm
- Hardware configuration: ASRock X370M PRO4
Ryzen 2400G APU
16 GB DDR4-3200
MSI GTX 1660 Super Gaming X
Re: A plea thats probably been made before.
Great points above by both Lazvon and Joe_H. In theory if everyone used the same GPU's and had folding dedicated rigs, the system could give the same PPD returns for all people and projects. But both hardware and projects vary in huge amounts, and it gets complicated quickly.
I saw over a 400% variance in PPD with my iGPU when I folded with it. Why? It would be bogged down and barely meet deadlines with most of the larger work units it picked up with higher atom counts. But those little tiny atom count projects that most people complained about.... it ate them for lunch. Once the atom count fell below about 5000, it was twice as fast as the higher atom count projects.
But over time, things average out. You win some, you lose some. In my opinion anyone cherry picking is just cheating themselves, as they hurt the science for their own self serving reasons. Liars lie, cheaters cheat. We might not ever change that, but the majority of folders hopefully aren't doing any such thing.
And the LARS numbers.... yeah what Joe said. Though I haven't used it, I've seen how the flaws in it are exposed. I got a return of over 12M PPD on a 1660 Super.
I saw over a 400% variance in PPD with my iGPU when I folded with it. Why? It would be bogged down and barely meet deadlines with most of the larger work units it picked up with higher atom counts. But those little tiny atom count projects that most people complained about.... it ate them for lunch. Once the atom count fell below about 5000, it was twice as fast as the higher atom count projects.
But over time, things average out. You win some, you lose some. In my opinion anyone cherry picking is just cheating themselves, as they hurt the science for their own self serving reasons. Liars lie, cheaters cheat. We might not ever change that, but the majority of folders hopefully aren't doing any such thing.
And the LARS numbers.... yeah what Joe said. Though I haven't used it, I've seen how the flaws in it are exposed. I got a return of over 12M PPD on a 1660 Super.
Fold them if you get them!
-
- Posts: 70
- Joined: Wed Apr 08, 2020 8:55 pm
- Hardware configuration: Lots... Look at my website: www.mxyzptlk.us
- Location: California
- Contact:
Re: A plea thats probably been made before.
"You have target rates given to you here..https://folding.lar.systems/gpu_ppd/overall_ranks."
These are not target numbers for GPU's, they are average results from many GPU WU's. Lars will be the first to tell you the flaws of his own database. If you dig deeper then the overall average the front page gives it will go into detail of how each card does on average with the many different projects out there. But Lars has no way of knowing what power limits people are running or the conditions the rigs are being run in or the CPU's running the GPU.
I run all my rigs using his extension to help out with the database. But at the same time I probable skew the averages downward because I run my cards on power limits (temp control) and I choose Cancer as my priority which has the the worst PPD for my 40XX cards.
I am curious, as you didn't say, but are the GPU's in the same rig or do they have different CPU's running them?
These are not target numbers for GPU's, they are average results from many GPU WU's. Lars will be the first to tell you the flaws of his own database. If you dig deeper then the overall average the front page gives it will go into detail of how each card does on average with the many different projects out there. But Lars has no way of knowing what power limits people are running or the conditions the rigs are being run in or the CPU's running the GPU.
I run all my rigs using his extension to help out with the database. But at the same time I probable skew the averages downward because I run my cards on power limits (temp control) and I choose Cancer as my priority which has the the worst PPD for my 40XX cards.
I am curious, as you didn't say, but are the GPU's in the same rig or do they have different CPU's running them?
I fold..... look at my folding setups here: https://mxyzptlk.us/about/
-
- Site Moderator
- Posts: 6349
- Joined: Sun Dec 02, 2007 10:38 am
- Location: Bordeaux, France
- Contact:
Re: A plea thats probably been made before.
Which GPUs do you use ? Which projects are you comparing ? Do you have any idea about what influence performance on FAH ?Badsinger wrote: ↑Fri Apr 14, 2023 3:47 am Hi folks;
I'm here to plea, probably not for the forums first time, for standardized points per hour for any given type of video card. Right now I have 2 identical cards folding 2 different tasks. One earning 3.23 Mill ppd and the other 5.07 mill ppd. To put it nicely ...nonsense. The same card. The same opportunity costs for you economics types. Very similar heat outputs. The reward you give me costs you nothing. So why are the rewards per hour different? Different rewards encourage cherry picking of tasks. You have target rates given to you here..https://folding.lar.systems/gpu_ppd/overall_ranks. Please consider standardizing rates for video cards. Thanks for listening.
Badsinger
Edited for spelling