Hardware configuration: 2 x GTX 460 (825/1600/1650) AMD Athlon II X2 250 3.0Ghz Kingston 2Gb DDR2 1066 Mhz MSI K9A2 Platinum Western Digital 500Gb Sata II LiteOn DVD Coolermaster 900W UCP Antec 902 Windows XP SP3
diwakar wrote:... I will change the settings so that p80xx is consistent with other SMP WU's or vice versa but it would only happen on Tuesday (first working day next week).
Please correct me if I'm wrong, but projects 8001, 8004 and 8011 are not SMP WU's. I folded all with the v6.23 Uni-processor clients.
Thanks everyone for your comments about the PPD of 80xx series. I have adjusted the bonus factors for all 80xx projects and base points for 8011. There would still be 10% deviation in the PPD for individual WU's especially for 8011 due to the slight differences in the number of water molecules in the system. Let me know if there is still a huge difference between 80xx and other SMP projects.
it might be a little low now a 8004 is getting about 1/2 of what it was getting before which puts it on the low end of the SMP wu's 20% to 30% below the middle of the road so to speak.
Thanks Grandpa_01. Let me know how 8011 and 8001 are behaving as well. I am experimenting with these values right now and the current bonus factor is same as p7611. I will have a quick meeting with everyone in Pande lab on Tuesday about SMP WU's and then readjust the bonus factors on 80xx projects.
My results qualitatively agree with what Grandpa_01 posted. On my MacBook Pro (i7 2720, OSX.6.8), the ppd on the last 2 8001s has dropped by 50% (consistent with reducing k by a factor of 4) and is now roughly 10% lower than I usually see on SMP units with this machine.
0:34 tpf
client reports 25k ppd (compared with 55k ppd using previous k factor)
To be fair, that doesn't seem unreasonable compared with the core_a3 units under windows.
However, it would become an impossible task to try make a4 and a3 WUs match across different OS's. I've seen numbers like 20% improvement in a3 under linux, is the same true for a4?
I think the cut in PPD when a little to far.
PPD was cut by more than 50%
On my i7-980x, PDD got from a little more than 40K to about 16K-17K. I have not got an a3 in a long time, but I mostly get p70xx that fold at about 40K
On my i7-875k, PPD drop from 20K to 10K, with a3 on this machine I get about 16K.
This subject of PPD has been going on for years, and PPD variability is the most irritating thing about f@h. There are "good" WUs and "bad" WUs. It drives me mad. All I can say it's not getting better, it's not getting worse either. Same-old, same-old.
On a 2600K @ 4.8 GHz running v7 in a VM (Ubuntu 10.10), performance on p8004 went from 59,500 ppd to 29,400 ppd. THe average a3 SMP WU on version 6.34 produces about 36,500 ppd. So we've gone from about 63% more than the average a3 WU to 20% less, at least on a 2600K.
The real question isn't what happens on your machine ... it's what happens on the benchmark machine, and at the moment, I don't recall what that is (& I'm too lazy to search the FAQ).
Although nobody likes WUs that are below whatever THEY consider normal, 20% is a lot closer than 63%. Clearly 63% was an error and 20% is closer. It's still up to the Pande Group to benchmark the WUs in accordance with the documented procedures on the specified hardware.
1) Maybe projects 80xx are now too low.
2) Maybe the projects you consider "normal" are too high.
3) Maybe your machine is enough different from the benchmark machine that both groups of projects are correct and you're just looking at a +/- 10% variation based on hardware differences. (A pretty small variation in TPF can make a +/- 10% variation in bonus points.)
diwakar wrote:Thanks Grandpa_01. Let me know how 8011 and 8001 are behaving as well. I am experimenting with these values right now and the current bonus factor is same as p7611. I will have a quick meeting with everyone in Pande lab on Tuesday about SMP WU's and then readjust the bonus factors on 80xx projects.
If you look at the spreadsheet here https://docs.google.com/spreadsheet/ccc ... W1Gak8tbmc you will see that p7611 is at the verry bottom of the scale for PPD which is where 8XXX is now. Perhaps you may want to look and compare, it might give you a little better idea of what bonus factor might work.
No it is not you all of the 8xxx projects are now pretty low compared to other projects -20% to 30%. But it does not look like they are going to do anything about it.
I might have to take another look at my crediting code as I'm not sure I'm handling credit correctly for projects which don't have an integer value there, but it feels really low ( compare it with other smp projects where I get 4k on the same slot ).
The real question is how does project 8xxx compare to other uniprocessor cores when it's assigned to a uniprocessor client. They points for the hybrid projects which can be assigned to either uniprocessor or smp clients/slots are a compromise.
I would think the real question would be how many of these WU's or any WU's are being done by uniprocessor compared to -smp multi processor machines I do about 1 1/2 WU's per hour on a i7 980X so around 36 a day. My guess is the majority of these are being done on -smp rigs thus should be benchmarked to reflect that.
