As the Work Unit Crunches

Come join the... uh... er... fold.

Moderators: just brew it!, farmpuma

Re: As the Work Unit Crunches

Postposted on Wed Dec 14, 2011 11:24 pm

PhilipMcc wrote:Ragar Dan - looking at my stats on extreme oc - it appears my recent wu's were credited between 6pm and 9pm on 12/14. Much smaller point totals than yours. Have you been credited yet?

They seem to be crediting everyone today and until they catch up, in some odd fashion or another. I'm not sure if it was just in those 2 updates or if they'll continue further, but it may well be as you say.

I know one of my WU's submitted yesterday morning was putatively worth ~58000 points because it was a bigadv WU that was restarted after ~4 hours, and then there was an hour where my machine decided to lock up or something, which appears to be the HD giving me warning it's going to die at some point even though I can't get any software to show me a problem with it. Be that as it may, none of my extreme oc updates quite look like they contain that one, but maybe I'm just wishing for more points than I would have gotten had things gone normally. But I think not.
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5354
Joined: Sun Jan 20, 2002 7:00 pm

Re: As the Work Unit Crunches

Postposted on Thu Dec 15, 2011 11:04 am

Definitely a major glitch in the servers from 12/10 thru 12/13 -- looks like WUs turned in on 12/11 and 12/12 got credited on the 14th.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37493
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: As the Work Unit Crunches

Postposted on Wed Dec 21, 2011 4:10 pm

I had another problem the other day on my VMwared Linux client. After restarting it, for some reason it acted as though it was caught in an infinite loop or something, making no progress. So I killed the fah6 process and waited a bit, and then restarted it. I thought it was OK then, but did not notice that it chose to wipe out my personal information in client.cfg without telling me. So for a couple of days it wasn't able to do -bigadv (if any were available) because of there being no passkey, and worse, was crediting user Anonymous and team 0 with all the WU's turned in. I discovered it last night.

Makes one wonder. But it's a bit late in the life of this software to care much any more.
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5354
Joined: Sun Jan 20, 2002 7:00 pm

Re: As the Work Unit Crunches

Postposted on Sun Feb 19, 2012 5:11 pm

One of my systems just completed a WU. The end-of-WU disk thrash on this one caused the hard drive to make so much noise I thought the bearings on the CPU fan were starting to fail! (The system in question isn't currently mounted in a case, so there was no HDD activity light to let me know the disk was getting hammered...)
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37493
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: As the Work Unit Crunches

Postposted on Mon Dec 09, 2013 10:28 pm

This thread, like many folders on TR's team, seems to have taken a vacation from folding. Howdy, everyone.

I started up again in late October when I saw a thread on the front page about the Frankenbot, just to see how things would go, and maybe pass someone I'm currently near to doing so that when I get passed, I'm not pushed as far down the list.:D So I checked out this forum and saw there's new client software, and downloaded it, and I must say it seems a lot worse than the earlier version. The old SMP software used to use all available threads, now it uses 1 less than the total available no matter what. OK, but it also seems to be performing worse, so I have to wonder whether that particular change was an improvement.

But I also notice that, unlike the earlier version, when you wake up your display (on Windows 7, at least), the GPU client just quits instead of slowing slightly. It won't operate at all for me unless I let the display go back to sleep. Is that because my driver version is too old and I should upgrade to allow use of new API features that will keep that from happening? i don't know, but I'm wary of upgrading drivers since my experience with nVidia drivers has been a reduction in GPU folding performance.

And what's worse, some SMP units go into S-L-O-W mode when the GPU client is running. How slow? 50% longer time between frames. I don't recall that sort of decline before, but I admit I forget quite a bit about how things were when I decided to save electricity money and quit.

I would have considered running VMware again, but it appears the new version isn't available for Linux, and I haven't tried experimenting with older versions yet.
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5354
Joined: Sun Jan 20, 2002 7:00 pm

Re: As the Work Unit Crunches

Postposted on Mon Dec 09, 2013 11:00 pm

Ragnar Dan wrote: The old SMP software used to use all available threads, now it uses 1 less than the total available no matter what.


I also found that annoying. You can force it to use all threads in the advanced control options - configure - slots.

Ragnar Dan wrote:But I also notice that, unlike the earlier version, when you wake up your display (on Windows 7, at least), the GPU client just quits instead of slowing slightly. It won't operate at all for me unless I let the display go back to sleep


You're lucky, on mine it just locks up the machine if it decides to fold at all. (I have a Radeon 7790 ) so I just terminated the gpu slot.

That's the Win7 x64 box, all my linux servers are still chugging along ok though.
warhead
Gerbil
 
Posts: 35
Joined: Wed Jul 27, 2005 1:37 pm
Location: North Texas

Re: As the Work Unit Crunches

Postposted on Mon Dec 09, 2013 11:02 pm

Ragnar Dan wrote:The old SMP software used to use all available threads, now it uses 1 less than the total available no matter what.

FWIW you can still force it to use all cores with the --cpus=n (where n is the desired number of threads) command line argument.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37493
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: As the Work Unit Crunches

Postposted on Mon Dec 09, 2013 11:30 pm

just brew it! wrote:
Ragnar Dan wrote:The old SMP software used to use all available threads, now it uses 1 less than the total available no matter what.

FWIW you can still force it to use all cores with the --cpus=n (where n is the desired number of threads) command line argument.

Thanks for the reply. I just lost my response to it and have to get to bed now, so I'll try again tomorrow... hopefully I'll remember before it gets late. :oops:

But I did see that and found it a pain in the neck, compared to prior version considering the new UI feature (which I didn't discover for quite a while)... so I tested other methods. Anyway, more tomorrow.


Edit: Sorry warhead, didn't notice your reply ... Will try that... Nite all.
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5354
Joined: Sun Jan 20, 2002 7:00 pm

Re: As the Work Unit Crunches

Postposted on Mon Dec 09, 2013 11:50 pm

As others have stated, for the CPU client you can force its folding slot to use n threads. I actually did not know better so I forced it to n-1 knowing that my nVidia GPU client needs a core and I want a bit of CPU left to handle my other stuff. Sounds like a sensible default to me.

As for the GPU wake problem It is all in that Folding Power slider. Unless you set it to "Full", when you wake from the screensaver the folding client stops. IMO it is actually good behaviour of "getting out of the way" and I like the design. With a lot of apps using GPU acceleration these days (Vista/Win7, IE, Chrome), chances are when you are not idle that the screensaver is running, you are doing something. Like you, I want the max points so i slide the bar to Full, but say if I want to play a video, using either a dedicated player or Flash, my video will lag unless I pause the GPU folding client anyways.

The thing I still have not gotten over yet, is that they created a whole new set of command line parameters and I did not have time to learn the ins and outs of those yet.
Image
The Model M is not for the faint of heart. You either like them or hate them.

Gerbils unite! Fold for UnitedGerbilNation, team 2630.
Flying Fox
Gerbil God
 
Posts: 24286
Joined: Mon May 24, 2004 2:19 am

Re: As the Work Unit Crunches

Postposted on Tue Dec 10, 2013 1:45 am

I have a six-core/twelve-thread processor, and I never notice any CPU cores at zero. I use Process Lasso, and keep the folding slider on "full". Process Lasso seems to duck CPU usage whenever I use it for something else such as rendering, and when the render is done, PL lets things go back to helter-skelter folding in the CPU slot.

But no such luck with the GPU slot because when I start up a Civilization V game, it will usually crash within 5 minutes or so. So I have to manually duck the folding when I get ready to play a game.
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1543
Joined: Tue May 25, 2004 7:41 pm

Re: As the Work Unit Crunches

Postposted on Tue Dec 10, 2013 11:56 am

Keep in mind the amount of CPU cores F@H decides to use is now variable. It's load based and depends on where you move the slider to.

For my 4770K it will use 8 threads when at full, and change to 7 threads if I selection "medium". It would switch between 7 and 8 cores frequently so I just locked it at 7.
Kougar
Gerbil Team Leader
 
Posts: 245
Joined: Tue Dec 02, 2008 2:12 am
Location: Texas

Re: As the Work Unit Crunches

Postposted on Tue Dec 10, 2013 11:44 pm

Well, I learned some things tonight. I checked again and was wrong in what I said about a new version for Linux. I believe my actual problem was that it's newer than I can use on my old VMware player and guest Linux OS. Sorry for the errors, there.

Tonight, for some reason, the "Full" slider choice seems to be working properly. When I did that things worked better, but as others have mentioned it makes the system quite a bit less responsive (for example even in typing this reply). Yes, my video card is ancient and weak (GTS 450), but I noticed its driver claims to be from March 2013, so it likely isn't that. Perhaps I was OC'ed too far w/o enough and proper voltage increases and that's what screwed things up. I'm stuck at ~3.6 GHz presently on a 2.8 GHz i7-930, but that's decent enough. If I use it much longer I'll probably end up replacing it, but I'm often using other machines anyway and as mentioned before this bout of Folding is temporary.

FF: I never run a screensaver, but have the display sleep after 15 minutes, though from the context, it seems like you may have written a word other than the one you meant. Since the drive light always flashes when Folding, that, along with the warmth produced, is enough indication it's doing its thing. I'm running it "Light" when working on it, and when done with that, in the evenings, it goes to "Medium", and now that it's working better, perhaps "Full."

I would like to know, respecting Kougar's comments, whether the GPU client is using a lot more data so that it needs a core to itself more than it used to, when every now and then it would ask for data to be moved back and forth and would only slightly reduce the output of the CPU.

Anyway, thanks for all the replies, everyone. (I know I'm late again, but I spent some time learning things before I made my reply, and am using the folding PC instead of an iPad to reply so as not to lose it like happened last night.)
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5354
Joined: Sun Jan 20, 2002 7:00 pm

Re: As the Work Unit Crunches

Postposted on Wed Dec 11, 2013 1:27 am

Ragnar Dan wrote:FF: I never run a screensaver, but have the display sleep after 15 minutes, though from the context, it seems like you may have written a word other than the one you meant. Since the drive light always flashes when Folding, that, along with the warmth produced, is enough indication it's doing its thing. I'm running it "Light" when working on it, and when done with that, in the evenings, it goes to "Medium", and now that it's working better, perhaps "Full."

I would like to know, respecting Kougar's comments, whether the GPU client is using a lot more data so that it needs a core to itself more than it used to, when every now and then it would ask for data to be moved back and forth and would only slightly reduce the output of the CPU.

Anyway, thanks for all the replies, everyone. (I know I'm late again, but I spent some time learning things before I made my reply, and am using the folding PC instead of an iPad to reply so as not to lose it like happened last night.)

Nvidia GPU folding traditionally requires a "dedicated" core (even if it is an HT "fake" core), for some reason, or the ppd goes to the tank. I remember reading about that the new OpenCL-based x17 client should address that, but I am not sure. Are you getting x15 or x17 WUs?
Image
The Model M is not for the faint of heart. You either like them or hate them.

Gerbils unite! Fold for UnitedGerbilNation, team 2630.
Flying Fox
Gerbil God
 
Posts: 24286
Joined: Mon May 24, 2004 2:19 am

Re: As the Work Unit Crunches

Postposted on Wed Dec 11, 2013 6:10 pm

I can't speak about "needing" a CPU core for NVIDIA GPU folding.... this certainly used to be the case when NVIDIA folding actually used a full core's worth of overhead, but that was circa 2007/8... Given 1 less thread on a Hyperthreaded system will have only a tiny impact in perf anyway, I prefer to leave mine at 7.

I'd assume a fair amount of data is being moved around inside the GPU, but I wouldn't know. What I can tell ya is that as time progresses, F@H's GPU clients get better and better at maximizing GPU utilization of varying architectures. Open up GPU-Z and move to the GPU utilization % tab, and any 400-series cards should be around 99% utilization now when folding, or at least my 480 is. Even after the 500 parts came out GPU utilization used to only be around 70%, then it became 80%, then 90%... When my GPU started maxing out at 99% any GPU accelerated stuff lags like hell regardless of if the CPU client running or not. I'm not actually aware of any way to "dial down" the % of the GPU used unlike what is possible with the CPU client. There's a % utilization slider under F@H's settings for CPU clients but not the GPUs, unless I missed it.
Kougar
Gerbil Team Leader
 
Posts: 245
Joined: Tue Dec 02, 2008 2:12 am
Location: Texas

Re: As the Work Unit Crunches

Postposted on Thu Dec 19, 2013 3:36 pm

FF: I haven't gotten any x17 WU's. All x15 that I've noticed.

I really would like to use the v. 7 folding software, since it runs all the folding devices from the same software and you don't have to look around for different log files, and it takes care of renaming and moving those to a logs directory when a restart is done, too. All very useful. But the v. 6 software still works faster for me on SMP units, for whatever reason. It claims to map NT from 8 to 8, and whether that means anything or not, it definitely produces faster, and the frames come more consistently, which is to say they don't vary like the later version's do according to the GPU's behavior. All in all it produces several thousand PPD more. So I switched back to using it on Sunday (and I know how to run the SMP part as a service, while the other version I think I read somewhere about having that feature but have not looked into it). On v. 7 I could go from around 10 minutes a frame to over 35, and I'm still not sure why it does that. So I think their software needs to be improved a bit yet.

A further change I made was that on Monday night I decided to see what I could do with the CPU's speed since it had been more than cold enough for a number of days, and had settings saved (and mostly forgotten about) in the BIOS which seem to have worked. That's added about 138 MHz more (* (4+x)) which helps it produce even faster. Though, it's supposed to get a lot warmer in the next couple of days so maybe I'll have to reduce the speeds back down again, although the Vcore voltage is enough higher that I'm not sure it will make much difference. But anyway, it should produce something in excess of 200,000 points per week with little trouble, and my object should be obtained by the 27th of this month. If there was an old version of GPU software for Linux I'd throw one in the Linux box running notfred's folding setup, too.

This all reminds me of the money folding wants to eat, though. I began thinking of an SSD for it so the checkpoints wouldn't slow it down, and even a liquid cooling system when I saw that Corsair H60 on sale mentioned over the weekend. But I stopped myself in time. :o (Especially after reading the folding blog the other day and seeing my machine won't be getting bigadv WU's for much longer.)

Next month's bills should be interesting all the same. :wink:
Ragnar Dan
Gerbil Elder
Silver subscriber
 
 
Posts: 5354
Joined: Sun Jan 20, 2002 7:00 pm

Re: As the Work Unit Crunches

Postposted on Thu Dec 19, 2013 4:22 pm

If you're running a competently fast i5 or i7, checkpoints won't slow anything down. I say don't fiddle too much, because it's a waste of time. Set it up and let it fly.

If you're running a Pentium II, then checkpoints might slow you down...but you're already slow to begin with at that point.

Eating money. Well, yes it can, especially if you're like me and you want to get the most Work Units/points for your electricity dollar. But in the long run, endless fiddling with the configuration would be time better spent taking a part-time job to add a more modern CPU or GPU to any folding system that might be getting long in the tooth.
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1543
Joined: Tue May 25, 2004 7:41 pm

Re: As the Work Unit Crunches

Postposted on Thu Dec 19, 2013 4:42 pm

Ragnar Dan wrote:FF: I haven't gotten any x17 WU's. All x15 that I've noticed.
That's odd. My GTX 660 is getting x17s exclusively these days. May be the 450 is too weak, but even the Folding blog said the x15 should be end of life.

Ragnar Dan wrote:On v. 7 I could go from around 10 minutes a frame to over 35, and I'm still not sure why it does that. So I think their software needs to be improved a bit yet.
I don't pay attention much to individual frames. I look at the average over a few frames, or I just trust the PPD projection of the v7 client. One thing to consider though is that may be the old client does not give you the new WUs? I think before Frankie in October I was still on the v6 WinSMP client and I never got any A4s. Now with the v7 client I am getting A4s a lot.

Ragnar Dan wrote:(Especially after reading the folding blog the other day and seeing my machine won't be getting bigadv WU's for much longer.)
Damn, they are really pushing the envelope here and bias people to do GPU folding, points wise balanced with costs. Getting a 32-core system to get the bigadv points is going to be a really expensive proposition compared to a system with a couple of GPUs. :x
Image
The Model M is not for the faint of heart. You either like them or hate them.

Gerbils unite! Fold for UnitedGerbilNation, team 2630.
Flying Fox
Gerbil God
 
Posts: 24286
Joined: Mon May 24, 2004 2:19 am

Re: As the Work Unit Crunches

Postposted on Sat Dec 21, 2013 7:08 pm

Ragnar Dan wrote:FF: I haven't gotten any x17 WU's. All x15 that I've noticed.


You won't get Core 17 projects unless you configured your v7 GPU client with the "advanced" or "beta" flags. Vanilla installs do not yet qualify for them.
Kougar
Gerbil Team Leader
 
Posts: 245
Joined: Tue Dec 02, 2008 2:12 am
Location: Texas

Re: As the Work Unit Crunches

Postposted on Sat Dec 21, 2013 8:03 pm

Kougar wrote:
Ragnar Dan wrote:FF: I haven't gotten any x17 WU's. All x15 that I've noticed.


You won't get Core 17 projects unless you configured your v7 GPU client with the "advanced" or "beta" flags. Vanilla installs do not yet qualify for them.

I am configured for Core 17 WU's via the Advanced flag and I have not gotten one in a long while. I keep getting Core 15 WU's, specifically variations of 8018. Been on them since a week or so after we put Frankie back to sleep.

17's are supposed to be out of beta and any one can run them now. Advanced flag or no Advanced flag. But to get to them we have to clear out any 15 WU's that are still around. Hopefully the 8018 series will run out soon.
Khali
Gerbil XP
 
Posts: 322
Joined: Thu Dec 20, 2012 5:31 pm

Re: As the Work Unit Crunches

Postposted on Tue Jan 21, 2014 9:15 am

Lots of news and information this month from the Pande Group folding blog - http://folding.stanford.edu/home/blog

Including -
New GPU assignment server and back-up
New v7 (7.4.0) client with access to the new GPU back-up AS and a correction for the bonus points calculations
The hows and whys of BigAdv work units (including end of life date)
and much much more
Image Image
.* * M-51 * *. .The Whirlpool Galaxy.
farmpuma
Minister of Gerbil Affairs
Silver subscriber
 
 
Posts: 2311
Joined: Mon Mar 22, 2004 12:33 am
Location: Soybean field, IN, USA, Earth .. just a bit south of John .. err .... Fart Wayne, Indiana

Re: As the Work Unit Crunches

Postposted on Wed Jan 22, 2014 6:26 pm

farmpuma wrote:Lots of news and information this month from the Pande Group folding blog - http://folding.stanford.edu/home/blog

Including -
New GPU assignment server and back-up
New v7 (7.4.0) client with access to the new GPU back-up AS and a correction for the bonus points calculations
The hows and whys of BigAdv work units (including end of life date)
and much much more

Amazing what a mini revolt/riot on the folding forums will accomplish. One of the biggest complaints was the lack of communication with the donors. Presto, we have blogs almost on a daily basis since then. Hope they keep it up.
Khali
Gerbil XP
 
Posts: 322
Joined: Thu Dec 20, 2012 5:31 pm

f@h tee shirt

Postposted on Sun Jan 26, 2014 11:15 am

farmpuma
Minister of Gerbil Affairs
Silver subscriber
 
 
Posts: 2311
Joined: Mon Mar 22, 2004 12:33 am
Location: Soybean field, IN, USA, Earth .. just a bit south of John .. err .... Fart Wayne, Indiana

Re: As the Work Unit Crunches

Postposted on Mon Jan 27, 2014 8:56 pm

Get the app, get the t-shirt! :P
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1543
Joined: Tue May 25, 2004 7:41 pm

Previous

Return to TR Distributed Computing Effort

Who is online

Users browsing this forum: Yahoo [Bot] and 3 guests