Discussion in 'Folding@Home - Join Team Guru3D !' started by iancook221188, Oct 11, 2012.
look like a new V7 about time:banana:
iancook221188: First of all, sorry about the loss of your SSD -that sucks, but good that you weren't down for long because you foresaw the problem.
Anyway, I'm curious as to why my PPD has dropped after installing this beta v7.
Previous to installing this, I was getting WUs with really high credits which in turn, created a high PPD for just this one rig (roughly 40K + all by itself).
Since updating to this beta v7, I've either not gotten any decent WUs, or they're reconfiguring how many PPD each WU is worth and their weighed credits with this release.
I've duplicated the exact settings I had on the previous v7 release but just can't even come close to what I was producing before. Currently, I'm pushing out WU 7610 which has a base credit of 788 and an estimated credit of 3869, with an estimated PPD of 5088 at 10 minutes 57 seconds TPF.
Those numbers don't seem right - before the beta v7 I was getting almost 10 times more PPD.
I've had to try to make up for the loss on my own rig's PPD by increasing the folding on the other 2 rigs (which is something that I really didn't want to do due to age and heat).
As soon as this WU finishes, I'm going to try going back to the stable v7 I had previous and set it up identically and see if I can catch those high credit WUs I was getting before. But any other suggestions would be GREAT!
Here's what I had set with my rig that was producing the best results prior to this beta v7: (this was borne out of trial and error for my first couple weeks, letting one or two WUs go through and then tweaking the settings, letting another couple WUs go through and then to mark any differences in PPD, keeping whichever settings worked the best on my i7 3820 rig)
0 SMP .. CPUs set at 
1 SMP .. CPUs set at 
I know that the bigadv flag shouldn't have anything to do with what WU I was receiving because I don't have a 16 core processor, however, I was getting REALLY high credit WUs with this setting. (I don't know if there is even a place to check what WUs are bigadv or not)
I should mention that I did have my ATI 5970 crunching 2 GPU units as well early in my testing, but it never produced nearly as much as the processor by itself, so the drain on the electricity wasn't worth running that beast any longer.
As for the other machines, they're pretty much irrelevant since they're both usually running Uniprocessor and don't actually provide much number crunching on their own.
Any ideas besides going backwards and re-testing the v7 stable that was producing the best results for me?
I realize that this is FREE for everyone involved and that I shouldn't complain about helping the cause, it's just that I'm all-in on this and really just want to understand better how to help more efficiently. If there's something that I can do on this same rig that produces more and better results, it's now my job to ensure that's what I'm doing.
Still doesn't work for my GTX660....
i saw on the interweb today that cuda 5 was just released. so it's a start.
i my self have not had time to up date my client yet but there has been no change to the core so work unit wise they should be the same from 1.52, i do have to that ive also been getting very low ppd work units at the moment about 50-60% lower. this was on a 3570k with no gpu running on my other rigs i can make it go up a little buy moving a gpu to each core some work gpu units can take 0.05% and some cant take up would of 8/9% not knowing what the ati client is like ( do know that it is a lot high) could it be some times ir take 50% and some time it take 80%?
hehe this is very interesting
on going back on the 2.9 i don't think it should make a difference but you don't know unless you try im going to try the 2.9 soon see what or if it make a difference to ppd
Just as a test, I started a BIG WU with the bigadv flag in a VirtualBox Linux 2.6.
I'm SHOCKED at the PPD for this single unit: 779,000.
Unfortunately, my rig likely isn't going to finish the work on time, so really, it's just a test to see what, if anything, can be tweaked to actually pull off a huge WU like this one.
On the other side (Windows), I can't run anything other than the current Linux VB right now as it's pulling all 8 cores at full blast. Just starting the Windows v7 client brought my rig to it's knees.
Anyway, I'll report back here when this first test is over and perhaps with some good results for the Linux VB test! (HOLY CRAP, 779THOUSAND in a single WU?!!?!!!)
how many frames did it do, it may need a few to calculate it properly. witch work unit is it doing btw
It's working 8101. It's really struggling though. I jumped the gun and restarted the client a couple times because it looked like it wasn't doing anything, but it's only at 1% (2500 out of 250000 steps) after 21 hours, so there's NO way it's not going to meet the deadline in 1D 11H. I'm going to let it run until later tonight anyway just to see if it picks up at all, and then will mark down my progress and settings and try tweaking it a little for another test.
The only thing that sucks is that my PPD isn't going as well as it should because of this testing. However, if I find that I can actually pump out a bigadv WU with some tweaked settings, then it will make up for it in the long run.
After a few days of tweaking settings and playing with it, if it looks like it's not going to be successful, then I'll go back to Windows v7 stable and resume where I left off from there.
Any ideas or settings to try out is appreciated!
well the only thing i can say about bigadv is unless you meet the minimum requirement of 16core/threads you should not be doing them that what Sanford has made clear (they don't like people doing that as it can damage results). some time you can get different smp work unit by using the bigadv flag and not getting the bigadv unit. its probably best just to do non VM bigadv folding for now
^ I agree. I hacked the VirtualBox to show that I had 16 cores for the last 2 tests. It worked and picked up actual bigadv WUs with really REALLY high PPDs, but I knew right away that there was no way the rig was going to be able to pull them off.
I have read over and over that as long as the WU is finished on time is the most important part. The disclaimer to what constitutes an *actual* 16 core system is in itself partly to blame because it doesn't state 16 physical cores but means 16 logical cores. (if it were 16 physical cores then the logic would be that the constraint is that it requires 32 logical cores, precluding MOST computers from working on anything with the bigadv flag)
But enough about that - that test is over for me. I just wanted to see if it was possible on the new X79 architecture with this 3820, and it's clearly not.
Back to stable v7 release, I'm currently working on PID8055 with 196 base points and 1530 estimated credit, with a PPD of 16,500. That's just over a fifth of what I was getting prior to the install of the beta v7 client.
I've got some more tweaking to do on v7 to see if I can pick up some higher PPD units.
No more VM bigadv units from this rig!
one question how are you getting your ppd estimates? the funny thing is that sandford made the dead line so tight so that singe cpu system could not do them that some 16 physical core dual cpu systems can't finesh them on time :cry:
My PPD estimates are coming straight from HFM (based on ALL frames not just the last 3), but they're almost exactly the same as FAHMon report - which, by the way, aren't very far off from what the Folding Client Control actually predicts.
And yes, my test actually validates what I've been saying about Pande Group for a bit now - they've placed such HUGE constraints on what they want their algorithms crunching on that they're neglecting all the work that *could* be done. For instance, I have 8 computers here right now (customer's), but not a single one of them can run anything beyond the standard uniprocessor client and each will only get a couple hundred PPD no matter what I tweak.
Now, don't get me wrong, I fully understand the implications of crunching WUs on a processor that's not actually got as many cores as they want it crunched on, and I also understand that by doing so, it could invalidate their research or at the very least add a bunch of garbage data that they'd have to sort through... I get it.
What I don't get is that they've made it so stringent that real, ACTUAL 16 core systems are having a hard time meeting their deadlines, dropping a huge chunk of their really, *really* good number crunchers out of the equation.
IMO, they need to relax some of their specs so that more people can fold for them, which in return would create a much larger userbase for them to draw from.
Anyway, I digress. Points are points and it's all for a good cause anyway.
i think what happen last year is that the core limit was at 8core/threads but to many people where doing them and the smp projects were getting back logged and all systems where able to bigav raver than just the top 2p /4p systems. i new this move was coming it just that all the system witch where capable of bigadv are not now but that was said by sanford that this was going to happen right from the start
i think people expected that a high end cpu could run the client but sanford all way ment the bigav client for just the 2p /4p system to do it, it was just at the time it started i.e bigadv was when intel came back with hyperthreading so even there quad could do it when sanford wanted the dual quads only and normal cpu's to do smp, it just at the moment the smp seem to be down on point but i think in time they will make it a bit better for that client.
My 3930@4,5 Ghz finished in time more than 25 of these 8101 projects.It's true with 4 hours before the preferred time.If I go with higher ram,instead of 1333, 1600 or 1866 the tfp goes from 32 minutes to 29-30 minutes and that's 7-8 hours before preferred time and that's not to bad.But all of this in Ubuntu.
I'd love to run a couple tests on my own hardware for a couple days using your knowledge if you'd be kind enough to help me.
I have the rig you see in my specs, including the RipjawsZ 2133 RAM just *itching* to be utilized.
So couple quick questions: Is this your daily driver rig? If so, does that mean you're using Ubuntu all the time or is it a VM?
Lastly, would you give me your settings so that I could see if I can pull off a bigadv WU?
I realize that my CPU is lacking 2 physical cores over yours, and that's likely to make a really big difference, but I really would just like to try. Worst case scenario is that my rig simply won't do it in time and I just lose a day or two of crunching and go straight back to Windows and stable v7 @ ~50K PPD.
Thanks in advance for any info you're willing to pass along.
Again, this is really just a test. I'm willing to run 100% on my rig for a couple days, install Ubuntu (or run a live distro?) and just see how it goes...
This is my daily rig that I use for folding: Intel 3930k@4,5, Ram: Kingston Hyperx 8Gb 1866@2000mhz. My video card is a 460gtx but I fold on it when I am in Windows. To fold bigadv at the maximum speed I use Ubuntu 10.10,maybe 12.04 it's good enough too,I'm not sure.I use Ubuntu when I fold bigadv,and to receive a bigadv you have to use a script which tell the client you have 16 cores.Another thing, in general there are a lots of 8101 projects and sometimes 8102. 8102 folds faster and it is easier to finish the work before the preferred deadline.On my rig, if 8101 has a ~30min tpf ,8102 has~25min tpf for the same preffered deadline.If you are lucky you get a 8102 and maybe you can finish it in time. When I installed Ubuntu for the first time I found a very useful link,where you can fold bigadv in Ubuntu and change the cores the client detects.Here's the link:http://www.overclock.net/t/1048365/ubuntu-setting-up-big-bigadv-p6903-for-dedicated-2600k.
If you have other questions I'm glad to help you. Good luck
Thank you ra_alphaomega. That info would have been GREAT if I had a 6-core processor with HT.
This is almost exactly what I did to spoof 12 cores on my i7 3820 - and it just doesn't work. The deadline would have been missed by something like 3 days.
Guess I'll just have to save up and get a decent 3930K, or perhaps wait for a while and see what happens in the LGA2011 arena... gotta be something *really* good sometime soon!
Thanks again for the info.