CPU PPD Ranks

[dcrand2]

https://folding.lar.systems/cpu_ppd/overall_ranks

could someone explain why an intel I5 8365U and I3 3225 ranked 1 and 3, right next to a threadripper 3970x32-core and a 7900x 12-core? What am I missing? The 8365U PPD ALL LC almost doubles the 7900x. My 5600x, which is a decent mid range CPU is down at like 82nd place. Is there something about those older CPUs that are tailored for folding?

[Mxyzptlk]

Depends upon how you sort it. You can sort it a multitude of ways on that page. You should go over to the Linus Tech Tip Forums and ask Lars there, he is on there pretty regularly.

But part of the answer is on the individual CPU Page:

"The following data for the Intel(R) Core(TM) i5-8365U CPU @ 1.60GHz is estmated based on the average PPD per each logical CPU processor mulitplied by various counts of those logical processors.

PPD data for CPUs is presented this way as folders can optimize their setup by folding on any number of the CPUs available logical processors on a project work unit.

How many logical processors a folder should use for best results is suggested based on the number of GPUs they may also be folding on to allow CPU bandwidth for GPU folding.

The max logical processors a CPU has available should only be used on a system that is dedicated solely to CPU folding with very little other overhead."

[dcrand2]

Well, the way its sorted is by Points Per Day, but even PDP-PLP and PPD ALL, those two take the top spots. Ill try to go over to the LTT forum and poke around there, but I was trying to find any other way to read the charts besides, "4 year old mobile processor is faster than anything else according to the records".

[BobWilliams757]

The CPU pages are filled with freak scores for some work units, and often the returns are so huge that they skew the results.

12 Million plus points on a Celeron processor.... sweet!

https://folding.lar.systems/projects/fo ... file/14188


Scrolling down the CPU list had obvious errors, but when you click around some you find out how the averages were skewed so much. I have no idea what caused it, but it does make you wonder if due to some glitch every now and then someone gets a massive points payout just by random chance.


ETA - we know how many points our hardware can pull, and that the LAR site has flaws. It seems to work ok for GPU figures, but obviously something is wrong on the GPU stuff.

[Mxyzptlk]

ETA - we know how many points our hardware can pull, and that the LAR site has flaws. It seems to work ok for GPU figures, but obviously something is wrong on the GPU stuff.

GPU is pretty accurate. CPU has some flaws....

[BobWilliams757]

ETA - we know how many points our hardware can pull, and that the LAR site has flaws. It seems to work ok for GPU figures, but obviously something is wrong on the GPU stuff.

GPU is pretty accurate. CPU has some flaws....

Yeah.

What he said!


It does make me wonder how it got messed up. I'm not sure if LAR data shows specifics as to particulars, so it could be a problem on either end. It seems there are only really a handful of people that post on the forums, and many people might not even notice if they did get a freaky high PPD payout on a single work unit or something.