Linux (Ubuntu/HiveOS) low PPD - 3080

Moderators: Site Moderators, FAHC Science Team

Post Reply
mew905
Posts: 5
Joined: Thu Jun 19, 2008 12:58 pm

Linux (Ubuntu/HiveOS) low PPD - 3080

Post by mew905 »

Hello, I decided to try out Linux with my RTX 3080 as folding.lar reports it was averaging ~7M+ PPD on there. However upon switching I've been hovering around 4.5-5m. Windows sits at 5-5.5m. Is there something I'm missing? I thought Linux was the better OS for raw folding, or is ~5M expected on both?
toTOW
Site Moderator
Posts: 6296
Joined: Sun Dec 02, 2007 10:38 am
Location: Bordeaux, France
Contact:

Re: Linux (Ubuntu/HiveOS) low PPD - 3080

Post by toTOW »

It's probably just the mix of projects that is not the same ...
Image

Folding@Home beta tester since 2002. Folding Forum moderator since July 2008.
BobWilliams757
Posts: 493
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: Linux (Ubuntu/HiveOS) low PPD - 3080

Post by BobWilliams757 »

The mix of projects can quickly skew the numbers, especially with GPUs with a lot of shader cores. If you check on the LAR site, you can look at PPD per project, and then scroll down to your GPU. That will give you the numbers for that specific project vs all around averages.
Fold them if you get them!
mew905
Posts: 5
Joined: Thu Jun 19, 2008 12:58 pm

Re: Linux (Ubuntu/HiveOS) low PPD - 3080

Post by mew905 »

yes but the averaging should remove the variance between WUs -- thats why we typically use averages after all. This would also suggest that there should be roughly an equivalent amount of WUs that are 30-50% higher than the average.

I just find it strange that folding.lar uses automated reporting via reading the html reporting page, it's not user reported (as far as I'm aware) which would mean that 7M ppd should be pretty realistic.

That said I just saw that there are two "base" 3080s listed: just a 3080, and a 3080 10/20gb. Mine is 10GB, however the "3080" should also be 10GB as 12GB is listed separately. FAHControl doesn't say "10/20 GB" so I'm guessing mine is that base 3080.
toTOW
Site Moderator
Posts: 6296
Joined: Sun Dec 02, 2007 10:38 am
Location: Bordeaux, France
Contact:

Re: Linux (Ubuntu/HiveOS) low PPD - 3080

Post by toTOW »

I don't fully trust folding.lar since you never know the environment of the measurements ...
Image

Folding@Home beta tester since 2002. Folding Forum moderator since July 2008.
BobWilliams757
Posts: 493
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: Linux (Ubuntu/HiveOS) low PPD - 3080

Post by BobWilliams757 »

mew905 wrote: Mon Dec 26, 2022 7:57 pm yes but the averaging should remove the variance between WUs -- thats why we typically use averages after all. This would also suggest that there should be roughly an equivalent amount of WUs that are 30-50% higher than the average.

I just find it strange that folding.lar uses automated reporting via reading the html reporting page, it's not user reported (as far as I'm aware) which would mean that 7M ppd should be pretty realistic.

That said I just saw that there are two "base" 3080s listed: just a 3080, and a 3080 10/20gb. Mine is 10GB, however the "3080" should also be 10GB as 12GB is listed separately. FAHControl doesn't say "10/20 GB" so I'm guessing mine is that base 3080.
Your statement concerning averages is correct, but only if your baseline you are comparing is coming from the same average that took place on the LAR website. The averages they show are stated as being based on the latest 500 PPD samples. They also have charts that go back in history and show the ups and downs of various available projects and the rolling PPD averages.

I think LAR designates one plain language name for each specific device ID used for F@H. I see at least 3 different listings for the RTX3080, and that is excluding the ti, Lite Hash Rate, and mobile versions. Since not all GPU's are the same, they get broken down by device ID and to some extent basic characteristics that create that device ID. But even beyond that, not all manufacturers will run GPU's at the same clocks, they vary quite a bit. Based on https://www.techpowerup.com/gpu-specs/ ... 80.c3621 boost clocks can vary from the stock 1710 up to 1905 MHz. Add to this cooling variables (which might impact boost clocks you can reach) and any/all user overclocks, memory tweaks, OS types, lack of system use or frequent (other than folding) system use, etc.... and it all adds up to a fairly wide path of variables.

I think LAR is great for giving people a ballpark idea of what they might expect, but you have to take it with a grain of salt. If you have a lesser GPU, cooling solution, bulky OS, and frequent use of the computer you aren't going to get the returns of those running optimized and sometimes overclocked systems doing nothing but folding using a higher strung GPU of the same basic configuration. And since we can never really know where the last 500 samples have come from, we have to take that into consideration.



The only real way to compare one OS to another on your system is to compare like project averages under similar if not the same conditions.
Fold them if you get them!
Post Reply