Go Back   Hardware Canucks > HARDWARE CANUCKS COMMUNITY > HardwareCanucks F@H Team

    
Reply
 
LinkBack Thread Tools Display Modes
  #11 (permalink)  
Old February 14, 2012, 07:59 AM
Hall Of Fame
F@H
 
Join Date: Apr 2011
Location: Montreal
Posts: 1,625

My System Specs

Default

you could also use a VM for the SMP folding under Linux and keep the GPUs under Windows in the same time.
__________________
Reply With Quote
  #12 (permalink)  
Old February 14, 2012, 11:49 AM
Linus's Avatar  
Join Date: Apr 2007
Posts: 716
Default

Alright. Maybe I'll hit up Dead Things as I'm pretty green at all of this stuff (including Linux). Would be handy if I could find someone local, but I'll settle for whatever I can get. Not even sure where to begin as I have zero experience with various Linux distros etc.

3.0Charlie, is there a reason you are not using GPUs and CPUs in the same rig? Which is more efficient in terms of PPD/$ these days? I was planning to configure the dual Opteron machine with some GPUs as well, but is that not recommended?
__________________
Please note: I am not here to provide official NCIX support on these forums.

For faster (and official) NCIX service please contact our customer care team at www.NCIX.com (Canada) or www.NCIXUS.com (USA)

Heatware: http://heatware.com/eval.php?id=25647
Reply With Quote
  #13 (permalink)  
Old February 14, 2012, 12:16 PM
G777's Avatar
Allstar
 
Join Date: Mar 2010
Location: Coquitlam,BC
Posts: 541

My System Specs

Default

Quote:
Originally Posted by Linus View Post
Alright. Maybe I'll hit up Dead Things as I'm pretty green at all of this stuff (including Linux). Would be handy if I could find someone local, but I'll settle for whatever I can get. Not even sure where to begin as I have zero experience with various Linux distros etc.

3.0Charlie, is there a reason you are not using GPUs and CPUs in the same rig? Which is more efficient in terms of PPD/$ these days? I was planning to configure the dual Opteron machine with some GPUs as well, but is that not recommended?
The thing is that points generated by SMP folding on the cpu are based on a quick return time, so that usually any points that are gained by gpu folding are negated by the stolen cpu cycles slowing down the cpu client's speed.

At present: Linux>VMWare Linux>Windows. As you're probably realizing truly 'optimizing' a folding rig can be a fairly complicated endevour, depending on how far you really want to go.
__________________
_______________________________________________

HWC Overclocking Team, breaking hardware for fun!

Team Thread
Reply With Quote
  #14 (permalink)  
Old February 14, 2012, 12:52 PM
Top Prospect
F@H
 
Join Date: Jan 2012
Posts: 104

My System Specs

Default

I just don't understand how putting a few CPU cycles towards GPU Folding makes that much of an impact. I would think that the PPD produced from the GPU's would be more than the lost PPD from the CPU. Isn't the CPU utilization from GPU folding like 3-5% for nvidia cards? If this is the case then I can't see it being better to only run CPUs instead of CPU and GPU in the same system.

Has somebody actually tested this? Perhaps somebody can clarify the situation for me and Linus a little as I am sure he is wondering the same thing.
__________________
Reply With Quote
  #15 (permalink)  
Old February 14, 2012, 01:18 PM
SugarJ's Avatar
Moderator
F@H
 
Join Date: Jan 2008
Location: Langley, BC
Posts: 6,140

My System Specs

Default

Even the fastest GPU will only hit 20K PPD, while some of these machines hit well over 200K PPD. Because the bonus points are awarded on a sliding time scale, that 5% cpu time you lose results in more than a 5% decrease in points. It's not worth putting GPU's in a dedicated 2p/4p folding machine.

Edit: It also matters which version of Ubuntu you use. 10.04 apparently has shorter TPF than 11.04 or 11.10.

Last edited by SugarJ; February 14, 2012 at 01:23 PM.
Reply With Quote
  #16 (permalink)  
Old February 14, 2012, 01:18 PM
Dead Things's Avatar
Hall Of Fame
F@H
 
Join Date: Oct 2008
Location: Centre of the Universe
Posts: 1,572

My System Specs

Default

Quote:
Originally Posted by Mathus View Post
I just don't understand how putting a few CPU cycles towards GPU Folding makes that much of an impact. I would think that the PPD produced from the GPU's would be more than the lost PPD from the CPU. Isn't the CPU utilization from GPU folding like 3-5% for nvidia cards? If this is the case then I can't see it being better to only run CPUs instead of CPU and GPU in the same system.

Has somebody actually tested this? Perhaps somebody can clarify the situation for me and Linus a little as I am sure he is wondering the same thing.
Good question - and yes, it's been tried and tested numerous times by myself and many others here who can attest to the same thing. The issue stems from the "S" part of "SMP" - that is, symmetry. Even though a GPU may take only a small proportion of cycles away from a single core in a multi-threaded machine, it wreaks havoc on processing symmetry such that many cycles are displaced and have to be re-assigned. These cycles are appended to the end of other cycles already existing in the same processing thread in accordance with a round-robin schedule. Your kernel scheduler takes care of this in a manner that is seamless for most usage profiles, but significantly negative in terms of performance when dealing with a well-optimized multi-threaded process that is highly-sensitive to asymmetry like FAH.

The problem here is that every time this happens, all of the other processing threads have to stop and wait for the extra appended cycles on the one thread to complete. So while the GPU process may occupy merely 5% of the cycles of a thread, its net affect on the SMP process is much greater. I've seen boxes wherein the net time the fahcore has spent waiting for cycles to catch up has been in excess of 60%. That means out of every 100 minutes of processing time, just 40 were spent doing actual work.

The faster the machine (and thus, the more dependent upon quick return bonuses) the more negative this effect is. In other words, while GPU folding does the same thing to a Q6600 as it does to quad 6174's, for example, the net effect in terms of PPD is exponentially smaller on a Q6600 than the quad-Opty box due to the exponential impact of the QRB's on the SMP PPD of the two machines.

...But most folders don't really care about the "why" of this kind of stuff - mainly the "how" of maximizing the PPD of the hardware available to them, hence the recommendation to segregate GPU and CPU folding machines.

edit - @ the OP, my recommendation would be - if possible - to choose a relatively high-profile member of the NCIX folding team from the GVRD that can maybe even make an appearance on the episode. That way, new members will see the guy from the episode on the team and might be more inclined to ask him questions rather than inundate you with them.
__________________
Follow my folding, mining & benching shenanigans @dt_oc!

Think you can overclock? Then show us what you got!
Join the Hardware Canucks Overclocking team today!

Last edited by Dead Things; February 14, 2012 at 01:27 PM.
Reply With Quote
  #17 (permalink)  
Old February 14, 2012, 01:25 PM
G777's Avatar
Allstar
 
Join Date: Mar 2010
Location: Coquitlam,BC
Posts: 541

My System Specs

Default

Quote:
Originally Posted by Mathus View Post
I just don't understand how putting a few CPU cycles towards GPU Folding makes that much of an impact. I would think that the PPD produced from the GPU's would be more than the lost PPD from the CPU. Isn't the CPU utilization from GPU folding like 3-5% for nvidia cards? If this is the case then I can't see it being better to only run CPUs instead of CPU and GPU in the same system.

Has somebody actually tested this? Perhaps somebody can clarify the situation for me and Linus a little as I am sure he is wondering the same thing.
Quote:
Originally Posted by Dead Things View Post
Good question - and yes, it's been tried and tested numerous times by myself and many others here who can attest to the same thing. The issue stems from the "S" part of "SMP" - that is, symmetry. Even though a GPU may take only a small proportion of cycles away from a single core in a multi-threaded machine, it wreaks havoc on processing symmetry such that many cycles are displaced and have to be re-assigned. These cycles are appended to the end of other cycles already existing in the same processing thread in accordance with a round-robin schedule. Your kernel scheduler takes care of this in a manner that is seamless for most usage profiles, but significantly negative in terms of performance when dealing with a well-optimized multi-threaded process that is highly-sensitive to asymmetry like FAH.

The problem here is that every time this happens, all of the other processing threads have to stop and wait for the extra appended cycles on the one thread to complete. So while the GPU process may occupy merely 5% of the cycles of a thread, its net affect on the SMP process is much greater. I've seen boxes wherein the net time the fahcore has spent waiting for cycles to catch up has been in excess of 60%. That means out of every 100 minutes of processing time, just 40 were spent doing actual work.

The faster the machine (and thus, the more dependent upon quick return bonuses) the more negative this effect is. In other words, while GPU folding does the same thing to a Q6600 as it does to quad 6174's, for example, the net effect in terms of PPD is exponentially smaller on a Q6600 than the quad-Opty box due to the exponential impact of the QRB's on the SMP PPD of the two machines.

...But most folders don't really care about the "why" of this kind of stuff - mainly the "how" of maximizing the PPD of the hardware available to them, hence the recommendation to segregate GPU and CPU folding machines.
Yeah... What he said
__________________
_______________________________________________

HWC Overclocking Team, breaking hardware for fun!

Team Thread
Reply With Quote
  #18 (permalink)  
Old February 14, 2012, 01:42 PM
Top Prospect
F@H
 
Join Date: Jan 2012
Posts: 104

My System Specs

Default

Thanks so much Dead Things! You have finally cleared up everything for me and it makes sense. Given the decrease in performance I may stop folding on both my HD7950 and i7 920 and just folding on my i7 920 if it will be better overall.
__________________
Reply With Quote
  #19 (permalink)  
Old February 14, 2012, 01:50 PM
dandelioneater's Avatar
Hall Of Fame
F@H
 
Join Date: Dec 2010
Location: Kelowna, BC
Posts: 1,114

My System Specs

Default

Linus:

Another program to check out is called: 'FAH gpu tracker" which IMO is great for newcomers to folding. It provides dead easy smp and gpu client setup under windows and provides some basic point monitoring as well. I think it uses the 6.34 folding client, not the latest v7 or whatever it is now, so basically it just doesn't support the better core for AMD gpu's. But AFAIK, cpu and nvidia gpu folding is great on it.

Also, (just my 2cents btw) but with your video maybe you should start with a "beginners guide" to folding. Perhaps hold off on the dual 16 core opteron system, and instead have a 2600K or 3930K setup with some GTX 4xx's or 5xx's or whatever. Then show how to setup the smp and gpu clients. I just feel like someone who is considering getting into folding would rather start with their current hardware to learn what it is about/configure clients/ and get the hang of things. I just think that people could misinterpret your video and think that they need to go out a buy a 2P of 4P system just to start folding on, and without a basic understanding of Folding, they are then dealing with server chips/boards, more advanced client setup, more $$, linux.


Anyways, whatever you choose to do I look forward to seeing your video.
__________________
Reply With Quote
  #20 (permalink)  
Old February 14, 2012, 01:52 PM
Dead Things's Avatar
Hall Of Fame
F@H
 
Join Date: Oct 2008
Location: Centre of the Universe
Posts: 1,572

My System Specs

Default

Quote:
Originally Posted by Mathus View Post
Thanks so much Dead Things! You have finally cleared up everything for me and it makes sense. Given the decrease in performance I may stop folding on both my HD7950 and i7 920 and just folding on my i7 920 if it will be better overall.
In my experience, eight threads is usually around the cut-off of where it stops making sense to combine GPU and CPU folding. My guess is that you are probably coming out ahead with both, but not by as much as you might think. So if it's just raw PPD you're after, then I'm sure the status quo will work for you.

The main metric that most concerns me, though, is PPD/W. I try to limit myself to 30 amps for folding due to the cost of hydro. So squeezing as much PPD out of every watt is important to me. And unfortunately, PPD/W is not a metric on which GPU folding does well. So if you pay for your own hydro, you may find that the extra little bit of PPD you net from running both clients simply isn't worth the extra hydro costs.
__________________
Follow my folding, mining & benching shenanigans @dt_oc!

Think you can overclock? Then show us what you got!
Join the Hardware Canucks Overclocking team today!
Reply With Quote
Reply


Thread Tools
Display Modes

Similar Threads
Thread Thread Starter Forum Replies Last Post
First Time Builder - Help me spend my money bonanza2000 New Builds 22 September 14, 2010 04:00 PM
Completed HELP! How to Spend Your Free Time After Exams? Rebuilding Computers! geokilla Case Mods, Worklog & Gallery 26 July 4, 2010 09:59 AM
CPU and GPU folding at the same time? RLP06 HardwareCanucks F@H Team 4 June 29, 2010 09:45 PM
My time folding farm martin_metal_88 New Builds 35 December 26, 2009 11:43 PM
Time for some Folding Humor!!!! Soultribunal HardwareCanucks F@H Team 31 May 12, 2009 07:49 PM