Hello,
i know it is possible to run several cuda WUs simultaneously.
but how can i do it in linux (ubuntu 10.10 64bit)?
thank you!
You need to put together an app_info.xml file with the relevant apps, files, and other configuration including setting GPU usage to correct value so that multiple work units can run at once. I have been meaning to put one together for Linux but haven't had the time yet.
This app_info.xml seems to be working for me so far with multiple WUs per GPU core in Linux. I will post a link instead of posting directly in the forum as to not clutter up the thread.
GT 240
want to buy another, but don't know which, because there is no any performance statistic for use in E@H
The GT 240 has 96-shader cores. It could be that with a single work unit, the GPU is already at close to max usage. There should be a tool called nvidia-smi included with the Linux Nvidia drivers that can give you that information. You would want to start this command with the -a flag to see that information.
I think a GTX 275, GTX 4xx, or the new GTX 560 should be able to see some benefit of running more than one work unit at a time. I see decent gains running two work units at once via a 295, which includes two lower clocked core versions of what the 275 comes with. You also need one CPU core for every GPU work unit that you run.
[error] No application found for task: i686-pc-linux-gnu 0 SSE2; discarding
[error] No application found for task: i686-pc-linux-gnu 0 BRP3cuda32nv270; discarding
Have tried all kind of options in my app_info.xml.
If I understood it well your app_info.xml blocks loading new files you need.
Remove or rename it.
Restart Boinc and all necessary files will be loaded automatically.
I don’t know if this solves your problem.
Maybe you need special files for running two CUDA wu’s, I don’t know. http://einsteinathome.org/node/195553 has a lot of info.
And eh pssssst..... Mr Einstein is still alive. He is crunching now. Don’t believe the press.
How to run several CUDA WUs in Linux?
)
You need to put together an app_info.xml file with the relevant apps, files, and other configuration including setting GPU usage to correct value so that multiple work units can run at once. I have been meaning to put one together for Linux but haven't had the time yet.
This app_info.xml seems to be
)
This app_info.xml seems to be working for me so far with multiple WUs per GPU core in Linux. I will post a link instead of posting directly in the forum as to not clutter up the thread.
app_info.xml
Set 0.500000 in the file based on the number of units you want to run per GPU. For 2 work units per core, set 0.5 and for 3, 0.33.
thank you, it's work! but
)
thank you, it's work! but almost with no performance gain (
RE: thank you, it's work!
)
What kind of graphics card are you running?
GT 240 want to buy another,
)
GT 240
want to buy another, but don't know which, because there is no any performance statistic for use in E@H
RE: GT 240 want to buy
)
The GT 240 has 96-shader cores. It could be that with a single work unit, the GPU is already at close to max usage. There should be a tool called nvidia-smi included with the Linux Nvidia drivers that can give you that information. You would want to start this command with the -a flag to see that information.
I think a GTX 275, GTX 4xx, or the new GTX 560 should be able to see some benefit of running more than one work unit at a time. I see decent gains running two work units at once via a 295, which includes two lower clocked core versions of what the 275 comes with. You also need one CPU core for every GPU work unit that you run.
Hi! Here's a rather
)
Hi!
Here's a rather typical output of
nvidia-smi -afor my GT 240, sitting in a Core2Quad 8200 Linux box:For comparison, you can see finished CUDA tasks on that card here:
http://einsteinathome.org/host/1767091/tasks&offset=0&show_names=0&state=3
I like the GT 240 because it is reasonable fast but consumes less than 75 W under full load (it doesn't require an extra power connector).
HB
How can I get rid of these
)
How can I get rid of these messages?
[error] No application found for task: i686-pc-linux-gnu 0 SSE2; discarding
[error] No application found for task: i686-pc-linux-gnu 0 BRP3cuda32nv270; discarding
Have tried all kind of options in my app_info.xml.
@Ziegenmelker If I
)
@Ziegenmelker
If I understood it well your app_info.xml blocks loading new files you need.
Remove or rename it.
Restart Boinc and all necessary files will be loaded automatically.
I don’t know if this solves your problem.
Maybe you need special files for running two CUDA wu’s, I don’t know.
http://einsteinathome.org/node/195553 has a lot of info.
And eh pssssst..... Mr Einstein is still alive. He is crunching now. Don’t believe the press.
Here is an updated
)
Here is an updated app_info.xml file with v1.08 application.
app_info.xml
There you can grab the latest files referenced in the above xml file from here.