Tesla in Linux without Monitor attached want to use built in GPU to run monitor and Tesla for number
hi,

I've got a few Tesla 2075's, thanks to Nvidia for sponsoring! Really appreciated.
This will kick major butts!

If i put in a tesla at Tyan S4985 board, it already has built in videocard.
I can boot on this card (it's a XGI - not so relevant further except it's not a nvidia).

Installed driver, clicked 'no' integrating it into X, installed cudatools and Nvidia gpgpu SDK.

Yet CUDA doesn't get recognized then.

Again installed driver now clicked 'yes' integrating into X. Guess what? It then BOOTS using the tesla card as gpu.

Now cuda works but it also uses tesla for gpu. Dang! Away 25% of performance! (performance loss % is gamble, in no way
this reflects an accurate measured amount)

In test i didn't even push the bandwidth yet feeding gpu
with all 16 cores generating the primenumber candidates, and soon another 64 cpu cores from cluster might join as feeding just 1 gpu is difficult with a cpu :)
(it's a pci-e 1.0 mainboard)

How to get to work the tesla *without* using it to steer the TFT so using them just to number crunch?

I spammed some dudes under which a few from ***, yet they have a week holiday right now it seems... (government dudes never work when you need 'em), whereas the awake math guys no experience setting up the stuff, the google attempts we did do also resulted in not finding anything... ..so that's why i post here. Hope someone knows.

Thanks a lot for answer!
Vincent Diepeveen
diep@xs4all.nl

p.s. i also tested with a nvidia card in pci slot, yet then the 295.41 (also the .40 ) driver complains that this oldie pci Nvidia card no longer is supported bla bla, whereas i just want to number crunch on the tesla and losing any performance not my intention of course - you number crunch or you do not.
hi,



I've got a few Tesla 2075's, thanks to Nvidia for sponsoring! Really appreciated.

This will kick major butts!



If i put in a tesla at Tyan S4985 board, it already has built in videocard.

I can boot on this card (it's a XGI - not so relevant further except it's not a nvidia).



Installed driver, clicked 'no' integrating it into X, installed cudatools and Nvidia gpgpu SDK.



Yet CUDA doesn't get recognized then.



Again installed driver now clicked 'yes' integrating into X. Guess what? It then BOOTS using the tesla card as gpu.



Now cuda works but it also uses tesla for gpu. Dang! Away 25% of performance! (performance loss % is gamble, in no way

this reflects an accurate measured amount)



In test i didn't even push the bandwidth yet feeding gpu

with all 16 cores generating the primenumber candidates, and soon another 64 cpu cores from cluster might join as feeding just 1 gpu is difficult with a cpu :)

(it's a pci-e 1.0 mainboard)



How to get to work the tesla *without* using it to steer the TFT so using them just to number crunch?



I spammed some dudes under which a few from ***, yet they have a week holiday right now it seems... (government dudes never work when you need 'em), whereas the awake math guys no experience setting up the stuff, the google attempts we did do also resulted in not finding anything... ..so that's why i post here. Hope someone knows.



Thanks a lot for answer!

Vincent Diepeveen

diep@xs4all.nl



p.s. i also tested with a nvidia card in pci slot, yet then the 295.41 (also the .40 ) driver complains that this oldie pci Nvidia card no longer is supported bla bla, whereas i just want to number crunch on the tesla and losing any performance not my intention of course - you number crunch or you do not.

#1
Posted 05/01/2012 07:19 PM   
From the release notes:


----------------------------------------
Linux
----------------------------------------
* In order to run CUDA applications, the CUDA module must be loaded and the
entries in /dev created. This may be achieved by initializing X Windows, or
by creating a script to load the kernel module and create the entries. An
example script (to be run at boot time):
#!/bin/bash

/sbin/modprobe nvidia

if [ "$?" -eq 0 ]; then

# Count the number of NVIDIA controllers found.
N3D=`/sbin/lspci | grep -i NVIDIA | grep "3D controller" | wc -l`
NVGA=`/sbin/lspci | grep -i NVIDIA | grep "VGA compatible controller" | wc -l`

N=`expr $N3D + $NVGA - 1`
for i in `seq 0 $N`; do
mknod -m 666 /dev/nvidia$i c 195 $i;
done

mknod -m 666 /dev/nvidiactl c 195 255

else
exit 1
fi
From the release notes:





----------------------------------------

Linux

----------------------------------------

* In order to run CUDA applications, the CUDA module must be loaded and the

entries in /dev created. This may be achieved by initializing X Windows, or

by creating a script to load the kernel module and create the entries. An

example script (to be run at boot time):

#!/bin/bash



/sbin/modprobe nvidia



if [ "$?" -eq 0 ]; then



# Count the number of NVIDIA controllers found.

N3D=`/sbin/lspci | grep -i NVIDIA | grep "3D controller" | wc -l`

NVGA=`/sbin/lspci | grep -i NVIDIA | grep "VGA compatible controller" | wc -l`



N=`expr $N3D + $NVGA - 1`

for i in `seq 0 $N`; do

mknod -m 666 /dev/nvidia$i c 195 $i;

done



mknod -m 666 /dev/nvidiactl c 195 255



else

exit 1

fi

#2
Posted 05/01/2012 08:36 PM   
Many thanks, i will try this!
Many thanks, i will try this!

#3
Posted 05/02/2012 08:42 AM   
Scroll To Top