Help Needed With Nvidia Open CL AP App

Message boards : Number crunching : Help Needed With Nvidia Open CL AP App
Message board moderation

To post messages, you must log in.

AuthorMessage
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1817
Credit: 262,693,308
RAC: 44
Australia
Message 1352709 - Posted: 1 Apr 2013, 9:23:38 UTC
Last modified: 1 Apr 2013, 9:24:36 UTC

As a total Noob where these apps are concerned I'd appreciate some advice. I've checked around the different Open CL threads, but they all seem to refer a lot more to ATI cards rather than Nvidia.

Dumbest question first.
1) Does the ap_cmdline_xxx.txt file syntax need to contain the name of the .exe file or just the switches ? e.g. Should it be AP6_xxx_xxx.exe -s, or just (for example) -v N or -ffa N with each entry ? How should these be entered, one switch per line or consecutively e.g -v N -ffa_block_switch N etc. on the same line

2) What driver version is recommended ? The rig I'm using for test purposes has 285.58 installed but the rig that will be the working rig when testing is complete has 310.70 on board.

3) To save me having to reinvent the wheel, does anyone has some starting values in the settings (-ffa_block, -unroll etc.) for use with a GTS250 and a GTX470 ?

4) What CPU usage should I expect with one core dedicated to the Open_CL app ? If running 2 units per Nvidia GPU, do I still have allow one core per unit ?

Thanks in advance.

T.A.
ID: 1352709 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34249
Credit: 79,922,639
RAC: 80
Germany
Message 1352716 - Posted: 1 Apr 2013, 9:38:50 UTC

Just put the parms into the text file.

Like -unroll 6 -ffa_block 6144 -ffa_block_fetch 1536.

You can use each driver but with 258.x CPU usage is much lower.

See my example above, its a good start.

If you use 258.x drivers CPU usage is similar to AMD cards which is very low.
Using new drivers it will consume almost full CPU core.



With each crime and every kindness we birth our future.
ID: 1352716 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1352720 - Posted: 1 Apr 2013, 10:02:26 UTC - in response to Message 1352709.  
Last modified: 1 Apr 2013, 10:04:35 UTC

As a total Noob where these apps are concerned I'd appreciate some advice. I've checked around the different Open CL threads, but they all seem to refer a lot more to ATI cards rather than Nvidia.

Dumbest question first.
1) Does the ap_cmdline_xxx.txt file syntax need to contain the name of the .exe file or just the switches ? e.g. Should it be AP6_xxx_xxx.exe -s, or just (for example) -v N or -ffa N with each entry ? How should these be entered, one switch per line or consecutively e.g -v N -ffa_block_switch N etc. on the same line

2) What driver version is recommended ? The rig I'm using for test purposes has 285.58 installed but the rig that will be the working rig when testing is complete has 310.70 on board.

3) To save me having to reinvent the wheel, does anyone has some starting values in the settings (-ffa_block, -unroll etc.) for use with a GTS250 and a GTX470 ?

4) What CPU usage should I expect with one core dedicated to the Open_CL app ? If running 2 units per Nvidia GPU, do I still have allow one core per unit ?

Thanks in advance.

T.A.

1) Just enter the values consecutively, ie: -unroll 10 -ffa_block 6144 -ffa_block_fetch 1536

2) Driver version really doesn't matter (With the exception that 270 and later drivers use a full core, and that pre-270 drivers use minimal CPU usage, but is slower overall, and you loose the ability to use Cuda42 or Cuda5 apps)

3) There are suggested values at the Bottom of the Readme for the latest AP apps, after that you can tweak it by going up or down on the -unroll (affects memory usage),
or going up or down on the -ffa_block/-ffa_block_fetch values (affects GUI usability, too high and it'll be laggy)
You should find that performance is relatively poor on the GTS250 compared with the Setiathome Cuda apps, and is not worth running on it, and GUI is laggy.

4) Each app instance is going to need a full core available otherwise there will be slowdown (if you're running +270 drivers)

Claggy
ID: 1352720 · Report as offensive

Message boards : Number crunching : Help Needed With Nvidia Open CL AP App


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.