• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

F@H project 18490 points suck

N

NixZiZ

Limp Gawd
2FA
Joined
Dec 30, 2013
Messages
163
Was wondering if it was anyone else who gets terrible points on CPU project 18490 vs others.
12477 is 2.5 to 3.1 million PPD (on a certain hardware class) 12476 is the same.
18490 is ~300-500k ppd (on the same hardware class)
Multiple identical pieces of hardware experience it, at first I thought it was a problem with a specific server.

12477 on hardware ~ 1/4-1/3 as powerful is getting around 700k PPD.

"Good" projects point wise seem to be:
18240
12486
12475
12479

"Bad" projects seem to be:
12485
16969

Honestly though neither stand out as much as 18490 does.

Of course ultimately points don't matter as much as the actual goal to provide scientific computing resources, just wonder if this is an AMD thing? All of my systems are AMD.
Do intel CPUs fare differently with this project?

Anyone have any knowledge of why this project doesn't generate the same quantity of PPD?
It does seem to involve development of new techniques.
 
NixZiZ said:
Was wondering if it was anyone else who gets terrible points on CPU project 18490 vs others.
12477 is 2.5 to 3.1 million PPD (on a certain hardware class) 12476 is the same.
18490 is ~300-500k ppd (on the same hardware class)
Multiple identical pieces of hardware experience it, at first I thought it was a problem with a specific server.

12477 on hardware ~ 1/4-1/3 as powerful is getting around 700k PPD.

"Good" projects point wise seem to be:
18240
12486
12475
12479

"Bad" projects seem to be:
12485
16969

Honestly though neither stand out as much as 18490 does.

Of course ultimately points don't matter as much as the actual goal to provide scientific computing resources, just wonder if this is an AMD thing? All of my systems are AMD.
Do intel CPUs fare differently with this project?

Anyone have any knowledge of why this project doesn't generate the same quantity of PPD?
It does seem to involve development of new techniques.
Click to expand...
There are at least two key things:
- number of atoms, that will determine good or bad scaling on a high count of threads (like 32). Some people will prefer to run 4 resources groups of 8 threads to maximize WU output (not good for PPD)
- intra-project variability : some initial conditions vary widely and impact the trajectories, it is especially true for GPU

P18490 is only 21100 atoms so not scaling good.

Below my average PPDs on a 9950x running GPU and CPU (28):
18490: 1,1 M PPD - 21 k atoms
12475: 2 M PPD -180 k atoms
12476: 2,3 M PPD
12477: 2 M PPD
12478: 2,1 M PPD
12479: 2 M PPD
12485: 1,3 M PPD - 113 k atoms
12486: 1,3 M PPD
16959: 1,3 M PPD - 56 k atoms
16969: 1,3 M PPD
 
I get 105-110K PPD on CPU projects 18240, 16959, 16969 and 18490 running 6 physical cores on a i5-10400 Dell box.

I can't speak to the 124xx projects. I don't get them with 6 cores enabled. I suspect 18490 and 16969 don't load all of your cores.

I get 20% lower PPD on the 1922x projects. I reported this recently in the FAH support forum. I managed to get the base credit for 19229 increased. Not much improvement after the change. 1922x work units are just slow. Fortunately, they recently disappeared.

https://forum.foldingathome.org/viewtopic.php?t=43473
 
Last edited:
Nicolas_orleans said:
There are at least two key things:
- number of atoms, that will determine good or bad scaling on a high count of threads (like 32). Some people will prefer to run 4 resources groups of 8 threads to maximize WU output (not good for PPD)
- intra-project variability : some initial conditions vary widely and impact the trajectories, it is especially true for GPU

P18490 is only 21100 atoms so not scaling good.

Below my average PPDs on a 9950x running GPU and CPU (28):
18490: 1,1 M PPD - 21 k atoms
12475: 2 M PPD -180 k atoms
12476: 2,3 M PPD
12477: 2 M PPD
12478: 2,1 M PPD
12479: 2 M PPD
12485: 1,3 M PPD - 113 k atoms
12486: 1,3 M PPD
16959: 1,3 M PPD - 56 k atoms
16969: 1,3 M PPD
Click to expand...
Wow that's a great PPD on 18490. I assume that's because the 9950X has quite a bit of single-thread performance compared to 2x32C EPYC which is what I'm running for the bigger PPD numbers. These are fairly low clocked too.
Guess the key takeaway there is PPD on smaller #s of atoms scales more on single-thread performance, bigger numbers increasingly scale by thread.

bigadv returns when???? :LOL:
 
You must log in or register to reply here.
Back
Top