F@H testing on my new i7 system

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

Moderators: just brew it!, farmpuma

F@H testing on my new i7 system

Postposted on Sun Jan 13, 2013 2:31 am

So for some Saturday geek fun, I decided to do some CPU and GPU folding. Having a beer (or two) while testing my new system. I'm logging these points to TRFrankenbot because there are at least a hundred people named "bif-something" and I'm not feeling creative enough at the moment to choose a new ID that doesn't name a person, place, thing, or ship name from Star Wars or Star Trek. People are too predictable; yawn! :roll:

Anyway, I have a Core i7 3930k and an AMD 7970 HD, with 64 GB RAM and Windows 8 Pro 64 bit. I've got both SMP and GPU folding working now, and I'm logging between 25,000 and 35,000 PPD thus far. This number was in the mid-40,000's when I allowed it to use 100% of all 12 threads, but as you'll see below the system wasn't usable.

Here's what I learned on my Saturday Night:

1. When setting up a new Windows 8 PC with an AMD GPU, it's tempting to let Windows Update / Microsoft Update do your work for you and go get all the drivers you should need, right? And it's tempting to use the "autoselect" on the AMD website to tell you what you need, right?. Well, don't do this; or at least take it all with a grain of salt. I was getting a message about OPENCL.DLL missing, and this was causing the GPU client to fail. Eventually, I decided to physically go get, download, and install the whole Catalyst 12.10 package, which fixed it and allowed GPU folding to begin.

2. Even a 6C/12T CPU is not invincible. In order to have ANY reasonable response time in Firefox, I've had to adjust the SMP settings in the F@H configuration. Right now it's set to use only 8 threads. Additionally, I had to adjust the "%CPU Usage" slider (on the Advanced tab) to 80%. Otherwise, there's just not enough compute headroom to do simple end-user tasks. For some reason that I don't yet understand, my old Q6600 quad-core could run F@H at 100% and you'd never know it using Firefox or IE.

3. So this setting in F@H results in the monitor readings as follows:

Core limit = 8
Cpu % limit = 80

Resource Monitor shows CPU% = 84% to 90%
Resource Monitor shows 8 CPU threads in the 80%-ish range and 4 threads close to 100%.

My case is still on the workbench. I have the case doors off because I have one malfunctioning piece of equipment (a fan controller that I'll be RMAing soon). The left door is leaning against the unit so that the fan can blow into the case. Three 230mm case fans, Front (in), Side (in), and Top (out), currently plugged into various motherboard connectors. The Corsair H80i is set up in push-pull blowing from back of case (in). Ambient room temperature is about 79 F (26.1c).

Corsair Link shows CPU% is around 75% and temperature is 53-54 degrees c.
Corsair Link shows H80i cooler temp is around 36.5c
Corsair Link shows GPU temp is 58 c

According to Corsair Link, these are normal when not folding (basically just using web browser, etc):
CPU% <3%
CPU Temp <33c
GPU Temp <32c

I would like to find a way to run F@H at a higher priority, but not if it's going to impact everyday PC usage. I'm willing to shut it off manually when I'm going to do something critical like running one of my DAWs, but I can't put up with simple web browser usage going into the toilet.

Anyway, that's my fun thing for today. Now I'm off to find another beer and maybe make a pizza.
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1546
Joined: Tue May 25, 2004 7:41 pm

Re: F@H testing on my new i7 system

Postposted on Sun Jan 13, 2013 2:58 am

Friendly reminder: if you don't let the WU finish with -oneunit, there will be no points credited for all your test 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: 24378
Joined: Mon May 24, 2004 2:19 am

Re: F@H testing on my new i7 system

Postposted on Sun Jan 13, 2013 3:04 am

Hi!

Not sure what you mean by "-oneunit", but when I'm done with my testing, I'll hit "finish" and let it run for however long it needs to. At this writing, the GPU WU will finish in 4 hours, the SMP WU is going to run for another 18-19 hours. Is that basically what you meant?
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1546
Joined: Tue May 25, 2004 7:41 pm

Re: F@H testing on my new i7 system

Postposted on Sun Jan 13, 2013 3:41 am

BIF wrote:Hi!

Not sure what you mean by "-oneunit", but when I'm done with my testing, I'll hit "finish" and let it run for however long it needs to. At this writing, the GPU WU will finish in 4 hours, the SMP WU is going to run for another 18-19 hours. Is that basically what you meant?

Yup, that should do it. The -oneunit switch is from the older command line clients.

Frankie thanks for your contribution, but he's in his deep sleep now for another 10 months so I am doing it in his honour. ;)
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: 24378
Joined: Mon May 24, 2004 2:19 am

Re: F@H testing on my new i7 system

Postposted on Sun Jan 13, 2013 4:04 am

LOL, I'm just keeping his circuits from going completely dark. :lol:
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1546
Joined: Tue May 25, 2004 7:41 pm

Re: F@H testing on my new i7 system

Postposted on Mon Jan 14, 2013 10:18 pm

Well, son of a gun.

So I started experimenting more with Folding and wanted to see what was causing the muddy responsiveness. I'm using the free version of Process Lasso, with ProBalance enabled. No recurrence of last weekend's laggy UI, and I've even re-raised my folding so that it would use all 12 HT cores and run at 100%. Even without PL active, folding can't seem to slow down my system for normal UI usage, even with several IE windows open.

I have seen a few brief instances where PL temporarily downgrades IE's priority, then sets it back to whatever the norm is. The downgraded priority period rarely lasts for more than 5 seconds and has resulted in no perceptible lags all day Sunday or Monday.

My only guess is that during my testing last Saturday, I may have been bumping up against Windows 8's default weekly maintenance schedules.

For now I'm going to leave Folding at 100% of 12 cores, although I think I would like to eventually set it up so that it pauses folding whenever I open one of my DAW applications or a computer game (or Steam, etcetera). If I can do this with Process Lasso, I think then I can just set up folding to run automatically anytime I boot the system, right?

I've observed a Folding oddity too. On this system, the current PPD display shows only 19,800 (projected), and CPU + GPU folding is running right now. Wow, that seems very low. First of all, it's no higher than when I was only letting it use 8 cores and 80%. Second, My Asus laptop is a couple years old and runs an older i7, and that laptop was generating the same (or higher) PPD back in October. I'm not overclocking either the workstation's CPU or the GPU, and I wasn't doing so on the laptop either.

I read someplace that I could possibly get better folding performance by using Process Lasso to lock the F@Home tasks to the physical CPU cores and not running any of them on the virtual cores. Can anybody venture a guess on that, or maybe have some experience with this?

I suppose it's possible that the point calculations have been changed since the time we all fired up Frankie in October, which would probably mean that anything I do won't result in a significant increase in WU or PPD. I'd have to run some work units on the laptop to get a valid current-day comparison, but right now I really don't feel like doing that. Maybe in a week or two.
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1546
Joined: Tue May 25, 2004 7:41 pm

Re: F@H testing on my new i7 system

Postposted on Mon Jan 14, 2013 10:31 pm

A few thoughts...

The client is generally quite good about getting out of the way when other stuff needs the CPU. It is normally scheduled to run at the lowest possible process priority, so everything else takes precedence.

You will hit a point of diminishing returns when running more threads than the number of physical (not HT) cores. That might be why you're not seeing much difference in production between 8 and 12 threads. However, it could also be that you're running a different type of WU that produces at a different rate on a 3930K (this could also explain why the laptop seemed to do just as well).

Yes, locking the worker threads to physical cores would probably be a good idea.

Are you using a passkey? You need to do that to get the early turn-in bonuses.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37672
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: F@H testing on my new i7 system

Postposted on Mon Jan 14, 2013 11:03 pm

just brew it! wrote:Are you using a passkey? You need to do that to get the early turn-in bonuses.

I think points projections these days from monitoring apps include the bonus already, passkey in use or not.
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: 24378
Joined: Mon May 24, 2004 2:19 am

Re: F@H testing on my new i7 system

Postposted on Mon Jan 14, 2013 11:09 pm

Flying Fox wrote:
just brew it! wrote:Are you using a passkey? You need to do that to get the early turn-in bonuses.

I think points projections these days from monitoring apps include the bonus already, passkey in use or not.

Probably. But he still needs to use the key to actually get credit for the bonus, monitoring app or not.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37672
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: F@H testing on my new i7 system

Postposted on Mon Jan 14, 2013 11:13 pm

The reason I was thinking of pausing folding during DAW or game usage is to reduce these things:

--> Reduce the possibility of an undue audio glitch during microphone or line recording (or during critical listening).
--> Reduce heat generation while I'm actively using the workstation for activities that require that I be in the same room with the computer. I hate heat; it's fatiguing and uncomfortable and is counter productive in my workspace.
--> Reduce fan noise for two reasons: 1) direct noise can be picked up by microphones, and 2) ongoing noise is fatiguing even if I'm not recording with a mic.

Yes, still using Frankie's passkey, so he (and TR) are getting all the bonuses during my testing! :D And yes, it's my understanding too that the PPD is a projection including any expected bonuses. I'll set up my own folding account sometime soon.

It will be awhile before I put this machine "into production"; in other words, when I have the case buttoned up and ready to place under my desk in my home office/recording studio. Still dealing with a defective fan controller and arguing with myself whether or not to add some more SATA ports. Until I get those things worked out, it's just easier to keep the machine on the workbench with the sides partly open and with a single big monitor plugged in.

I also have TONS of software yet to install; most of my Office and web authoring software, plus my DAW, Video, and photo content creation software, and dozens of VST instruments. I knew this would be a long process and for now I'm just taking my time and allowing myself to "wander" a bit, exploring here and there things that I have never had a chance to explore before. Such as this folding testing, learning how Process Lasso works, and so forth.

It's been rewarding, too. I'm learning a lot and having a lot of fun, although now that the holidays are over, other parts of my life are beginning to suck time away from this like a black hole. :( Oh well, can't have everything AND time to enjoy it all too, right? :)
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1546
Joined: Tue May 25, 2004 7:41 pm

Re: F@H testing on my new i7 system

Postposted on Mon Jan 14, 2013 11:41 pm

You may need to use scripts to coordinate these shutdown/restart work, along with the big programs that you are going to run.
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: 24378
Joined: Mon May 24, 2004 2:19 am

Re: F@H testing on my new i7 system

Postposted on Tue Jan 15, 2013 2:30 am

I like the scripting idea. Maybe a script can be made to issue the "pause" command to the F@H engine upon instantiation of the DAW or game executable, and then the "unpause" could be issued when the game or DAW program is ended, eliminating the need for me to manually navigate to the F@H UI and punch the buttons myself?

Or maybe it would be acceptable to use Process Lasso to just set the folding processes' priority to "lowest" or even "zero CPU time", thereby effectively pausing any folding progress while I'm using the DAW or game, and once the DAW or game is shut down, then return the folding processes to their original "normal" priority? Well, if it can do that, I mean.

My thinking here is that F@H is not strenuous or resource intensive even if "just sitting there" without getting any CPU. Memory usage is very low, there's no issue with disk file locking or pending allocations, and even F@H's own check pointing process should allow proper recovery of the inflight F@H work units even if the worst should happen and Windows were to crash while using the DAW software or a game.

Pause/unpause might be better if F@H takes a checkpoint when pause is invoked, but I don't know if F@H does that.

Your thoughts (anybody)?
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1546
Joined: Tue May 25, 2004 7:41 pm

Re: F@H testing on my new i7 system

Postposted on Tue Jan 15, 2013 12:11 pm

The actual number crunching processes (fah_[core].exe) are already running at the lowest priority possible, that's what JBI has been trying to tell you. So I don't see it as a problem, especially on the CPU side. This is a well known problem for the OS scheduler and for the most part Idle processes behave pretty nicely.

For the GPU client, I am not entirely sure how it works but there does not seem to be "OS level" construct to swap the GPU programs in and out of the GPU. What I noticed is that the GPU will be dragged down with the GPU client working. So I would think about making the GPU client stop while you are working.
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: 24378
Joined: Mon May 24, 2004 2:19 am

Re: F@H testing on my new i7 system

Postposted on Tue Jan 15, 2013 12:58 pm

Oh, okay. Lots to learn, lots to think about.

Thanks.
BIF
Gerbil Jedi
Gold subscriber
 
 
Posts: 1546
Joined: Tue May 25, 2004 7:41 pm


Return to TR Distributed Computing Effort

Who is online

Users browsing this forum: No registered users and 2 guests