| View previous topic :: View next topic |
| Author |
Message |
bdub
Black Belt 3rd Degree
Joined: 19 Feb 2003
Posts: 3348
Location: Washington D.C.
|
 Posted: Sat Oct 20, 2012 3:22 pm Post subject: new client... |
 |
|
wow this new folding client program is pretty good. it uses all cores, and detects the gpu. my gpu only shows as getting used at about 10% with spikes up to 20%. i've only been running it a short time, as i've just now installed it. are there any tweaks i should know about?
hmmm. now it's showing 0% gpu use, i guess i had the graphic part open and it was showing that. how does it process through the gpu? does it only work on cuda enabled nvidia cards, or can it work on my old ati 3850?
i think i'm just doing a demo project, and haven't downloaded a real project yet. any tips on getting this thing started?
_________________
my main rig...
asrock 970 extreme3
AMD athlonII X3 440
zalman cpns5x performa hs/fan
crucial ballistix 2x4gb sport ddr3-1333
powercolor 3850 256bit/512meg vga
antec neo he 650r
kingston hyperx 3k ssd 120GB
seagate 500GB HDD 32M cache
seagate 750GB HDD 32M cache
hitachi 1TB HDD 32M cache
lg wh14ns40 bd burner
optiarc ad-7240s sata dvdrw (nec chipset) |
|
| Back to top |
|
 |
Karlsweldt
Enlightened Master
Joined: 12 Nov 2003
Posts: 18694
Location: 07438
|
| Posted: Sun Oct 21, 2012 4:55 am Post subject: |
|
|
Stanford U. has some tech info plus recommendations in this link about GPU folding.. http://folding.stanford.edu/English/FAQ-ATI .
And WIKI has a lengthy article, also.. http://en.wikipedia.org/wiki/Folding@Home .
Most newer graphics cards are likely candidates for the F@H process.. but older ones may not be.
The F@H client must be able to detect the CPU or GPU core model, and apply the "divisor" process so each core can be optimally used, while not interfering with needs by the user or donor.
Depending on the disseminating process, some steps may be extremely intricate, or very simple.. thus a reading of maximum or minimum core use at any time.
_________________
F@H.. to solve mankind's maladies.. in our lifetimes! |
|
| Back to top |
|
|
bdub
Black Belt 3rd Degree
Joined: 19 Feb 2003
Posts: 3348
Location: Washington D.C.
|
| Posted: Sun Oct 21, 2012 12:12 pm Post subject: |
|
|
it seems the gpu core can't dl a work unit. from the log...
20:07:26:WARNING:WU00:FS00:Failed to get assignment from 'assign-GPU.stanford.edu:80': Empty work server assignment
is there something i need to set to get the gpu folding?
the links above are not much help for this version of the folding program (v7).
_________________
my main rig...
asrock 970 extreme3
AMD athlonII X3 440
zalman cpns5x performa hs/fan
crucial ballistix 2x4gb sport ddr3-1333
powercolor 3850 256bit/512meg vga
antec neo he 650r
kingston hyperx 3k ssd 120GB
seagate 500GB HDD 32M cache
seagate 750GB HDD 32M cache
hitachi 1TB HDD 32M cache
lg wh14ns40 bd burner
optiarc ad-7240s sata dvdrw (nec chipset) |
|
| Back to top |
|
|
evasive
Mobo-fu Master
Joined: 06 May 2001
Posts: 36479
Location: Netherlands, Breda
|
|
| Back to top |
|
|
Karlsweldt
Enlightened Master
Joined: 12 Nov 2003
Posts: 18694
Location: 07438
|
| Posted: Mon Oct 22, 2012 5:08 am Post subject: |
|
|
From the Currently Running Projects List, looks like not many GPU-intensive projects available.. http://fah-web.stanford.edu/psummary.html
But lots of GROMACS.. some with high points.
_________________
F@H.. to solve mankind's maladies.. in our lifetimes! |
|
| Back to top |
|
|
bdub
Black Belt 3rd Degree
Joined: 19 Feb 2003
Posts: 3348
Location: Washington D.C.
|
| Posted: Mon Oct 22, 2012 6:41 am Post subject: |
|
|
now installed on a laptop with quad-core sandy bridge and quadro 3000m vga chip (and at a new location... not my home machine)... everything is running fine.
i wonder if the hyper-threading machines would be better to set affinity on threads set core count to 4, as to get multi threading but have each physical core at full power? i guess i would have to physically set the correct cpu core affinity in the task manager too, (or it could pick the virtual core and its physical core on its own)...
or would breaking the project into 8 pieces with 8 cpu cores working half strength on each piece be faster? this is the default setting.
_________________
my main rig...
asrock 970 extreme3
AMD athlonII X3 440
zalman cpns5x performa hs/fan
crucial ballistix 2x4gb sport ddr3-1333
powercolor 3850 256bit/512meg vga
antec neo he 650r
kingston hyperx 3k ssd 120GB
seagate 500GB HDD 32M cache
seagate 750GB HDD 32M cache
hitachi 1TB HDD 32M cache
lg wh14ns40 bd burner
optiarc ad-7240s sata dvdrw (nec chipset) |
|
| Back to top |
|
|
Karlsweldt
Enlightened Master
Joined: 12 Nov 2003
Posts: 18694
Location: 07438
|
| Posted: Tue Oct 23, 2012 5:19 am Post subject: |
|
|
Best to set maximum core usage to no more than 90% per client/core designation. This allows some 'headroom' for background processes.
For my present setup, I run a Pentium 4 Prescott 2.8 with HT, (ASRock Conroe 865) and I have two separate F@H processes active (different drives, folder names). No problem with operations, and can run other apps easily.
The F@H clients are designed to optimize their performance, yet readily cede to any user needs on the computer. The newer OS versions can control multiple-core (true) CPU types, and assign semi-dedicated processes to each as required. Best to allow the OS to relegate usage of processing per actual core, rather than manual settings. "Handing off" a process between cores can evoke delays and errors.
_________________
F@H.. to solve mankind's maladies.. in our lifetimes! |
|
| Back to top |
|
|
|
