Kernel Usage Question [Linux]

Moderators: Site Moderators, PandeGroup

Kernel Usage Question [Linux]

Postby Kurtis200200 » Thu Feb 27, 2014 12:28 am

Hey all,

Running arch linux with two GTX 660s and a Q6600 (quad-core), and noticed in 'top' that, although each FahCore_17 is using 100% of a core, "system" usage for the machine is fairly high (averaging ~26.6%).

Should I be concerned about such high "system" usage? Is there some upgrade or configuration I should change to lower that amount, or is it supposed to be that high, and what precisely is counting as "system" usage in FahCore_17?
Kurtis200200
 
Posts: 80
Joined: Sat Jul 06, 2013 4:37 pm

Re: Kernel Usage Question [Linux]

Postby Jesse_V » Thu Feb 27, 2014 12:34 am

FahCore_17 uses OpenCL, which is handled by the drivers. I believe driver processing counts as system/kernel usage, hence the higher score. Not much you can do about it, Nvidia drivers use one CPU core for each card that's doing OpenCL processing. Since you have two GTXs, I recommend using FAHControl to modify the CPU count on the CPU slot so that it uses two less cores than you have available, thus leaving them for use by the drivers. That way your GPUs and your CPU don't fight over resources.
Pen tester at Cigital/Synopsys
User avatar
Jesse_V
 
Posts: 2773
Joined: Mon Jul 18, 2011 4:44 am
Location: USA

Re: Kernel Usage Question [Linux]

Postby Kurtis200200 » Thu Feb 27, 2014 1:16 am

I've actually disabled CPU folding entirely (removed the slot in FAHControl), and have BOINC running but only on the remaining two slots.

Does FahCore_17 like to use >1.0 cores per instance or is it alright to allocate 2 for the two GTX 660s and 2 for BOINC?
Kurtis200200
 
Posts: 80
Joined: Sat Jul 06, 2013 4:37 pm

Re: Kernel Usage Question [Linux]

Postby 7im » Thu Feb 27, 2014 1:45 am

Jesse_V wrote:...snip

Nvidia drivers use one CPU core for each card that's doing OpenCL processing. ...snip
That way your GPUs and your CPU don't fight over resources.


i.e. 2 cores for GPU folding, 2 for that other program.
How to provide enough information to get helpful support
Tell me and I forget. Teach me and I remember. Involve me and I learn.
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: Kernel Usage Question [Linux]

Postby davidcoton » Thu Feb 27, 2014 12:58 pm

Core17 has not previously been reported as using more than one core -- the indication is that more likely a lot of the activity is non-productive polling for completion (? of work on the GPU), and the full core is only needed intermittently. Thus in multi-GPU systems the core allocation for GPUs can be squeezed -- but no experimental evidence yet to prove or disprove this.

Can you pinpoint which process the system activity is connected with? Does it stop when Core17 is paused? When BOINC is paused? Or just keep running (a system process such as file indexing, not linked with folding)?

David
Image
davidcoton
 
Posts: 940
Joined: Wed Nov 05, 2008 3:19 pm
Location: Cambridge, UK

Re: Kernel Usage Question [Linux]

Postby PantherX » Thu Feb 27, 2014 1:36 pm

davidcoton wrote:...Thus in multi-GPU systems the core allocation for GPUs can be squeezed -- but no experimental evidence yet to prove or disprove this...

I remember that a donor used CPU Affinity to ensure that 2 instances of FahCore_17 were locked to a specific Core in Linux. No significant loss of performance was noticed.

I have done the same on my Windows 8 64-bit system too by using Process Lasso.

EDIT -> The GPUs used for both set-ups were Nvidia GPUs. I haven't come across any posts where CPU Affinity was used for AMD GPUs.
ETA:
Now ↞ Very Soon ↔ Soon ↔ Soon-ish ↔ Not Soon ↠ End Of Time

Welcome To The F@H Support Forum Ӂ Chrome Folding App (Beta) Ӂ Troubleshooting "Bad WUs" Ӂ Troubleshooting Server Connectivity Issues
User avatar
PantherX
Site Moderator
 
Posts: 6321
Joined: Wed Dec 23, 2009 9:33 am

Re: Kernel Usage Question [Linux]

Postby bruce » Thu Feb 27, 2014 4:32 pm

davidcoton wrote:Core17 has not previously been reported as using more than one core -- the indication is that more likely a lot of the activity is non-productive polling for completion (? of work on the GPU), and the full core is only needed intermittently. Thus in multi-GPU systems the core allocation for GPUs can be squeezed -- but no experimental evidence yet to prove or disprove this.


From reports gathered here on the forum (not specifically Core17) the high CPU activity has sometimes changed based on the version of the drivers without any change to the science code itself. This is a strong indication supporting non-productive polling (aka "busy waiting") somewhere in the drivers. If someone is able to profile the code, this might provide useful evidence one way or another.
bruce
 
Posts: 21165
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.


Return to V7.3.6 Public Release Windows/Linux/MacOS X

Who is online

Users browsing this forum: No registered users and 1 guest

cron