CUDA on Fedora Trying to compile SDK examples on Fedora 11

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I installed this rpm nvidia-x11-drv-190.36-1.fc11.i586 according this http://forums.fedoraforum.org/showpost.php…amp;postcount=4

Then I installed cudatoolkit_2.3_linux_32_fedora10.run and cudasdk_2.3_linux.run.

I have this video card http://www.nvidia.com/object/geforce_9200m_gs.html

I get this:

/home/lukas/NVIDIA_GPU_Computing_SDK/C/bin/linux/release

[lukas@pavilion release]$ cd ..

[lukas@pavilion linux]$ cd ..

[lukas@pavilion bin]$ cd ..

[lukas@pavilion C]$ pwd

/home/lukas/NVIDIA_GPU_Computing_SDK/C

[lukas@pavilion C]$ make

make[1]: Entering directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/common'

make[1]: Leaving directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/common'

make[1]: Entering directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/common'

make[1]: Leaving directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/common'

make[1]: Entering directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/common'

make[1]: Leaving directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/common'

make -C src/deviceQuery/

make[1]: Entering directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/src/deviceQuery'

make[1]: Leaving directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/src/deviceQuery'

make -C src/dwtHaar1D/

make[1]: Entering directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/src/dwtHaar1D'

make[1]: Leaving directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/src/dwtHaar1D'

make -C src/matrixMulDrv/

make[1]: Entering directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/src/matrixMulDrv'

/usr/bin/ld: cannot find -lcuda

collect2: ld returned 1 exit status

make[1]: *** [../../bin/linux/release/matrixMulDrv] Error 1

make[1]: Leaving directory `/home/lukas/NVIDIA_GPU_Computing_SDK/C/src/matrixMulDrv'

make: *** [src/matrixMulDrv/Makefile.ph_build] Error 2

This is what I added to .bashrc:

export PATH=$PATH:/usr/local/cuda/bin

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/nvidia

I added path /usr/local/cuda/lib to /etc/ld.so.conf.d/cuda.conf

Would you please help me?

I’m on Fedora 11, using gcc 4.4.1 with workaround http://forums.nvidia.com/index.php?showtopic=104525

2

I moved myself somehow (I installed newest driver I was able to find nvidia-x11-drv-190.36-1.fc11.i586 )

but now i get from ./deviceQuery There is no device supporting CUDA…

I did

ls /dev/ | grep nv

nvidia0

nvidia1

nvidia2

nvidia3

nvidia4

nvidia5

nvidia6

nvidia7

nvidia8

nvidia9

nvidiactl

nvram

Where can be a problem?

3

I added this to grub.conf:
title Fedora (2.6.30.8-64.fc11.i586)
root (hd0,2)
uppermem 524288
kernel /vmlinuz-2.6.30.8-64.fc11.i586 ro root=/dev/mapper/vg_pavilion-lv_root rhgb quiet vmalloc=256MB vga=0x318
pci=nommconf
initrd /initrd-2.6.30.8-64.fc11.i586.img

After reboot:
Oct 3 15:52:58 pavilion kernel: nvidiafb 0000:01:00.0: power state changed by ACPI to D0
Oct 3 15:52:58 pavilion kernel: nvidiafb 0000:01:00.0: PCI INT A → GSI 16 (level, low) → IRQ 16
Oct 3 15:52:58 pavilion kernel: nvidiafb 0000:01:00.0: BAR 3: can’t reserve mem region [0x90000000-0x91ffffff]
Oct 3 15:52:58 pavilion kernel: nvidiafb: cannot request PCI regions
Oct 3 15:52:58 pavilion kernel: nvidia: module license ‘NVIDIA’ taints kernel.

X server is running. I also chmod all the devices in /dev/nvidia* to 666.
DeviceQuery still did not find my CUDA device.
this /sbin/lspci | grep -i NVIDIA | grep “VGA compatible controller” | wc -l returns 1 of course.

Any ideas?

4

Weird…I installed original driver from NVidia and now everything is compiled fine and demos are working.

5

Lukas

You can install the rpmfusion nvidia driver, I do that at home and it seems to work.
But the rpmfusion driver is a bit older than the proprietary nvidia one. For example the
rpmfusion driver is around 185.18.36 but the one from Nvidia is 190.18

instructions here:

[url=“Fedora Nvidia Driver Install Guide”]http://www.mjmwired.net/resources/mjm-fedora-nvidia.html[/url]

the nice thing is you can then use the akmod-nvidia package and then each time you update your kernel the
nvidia driver will magically update itself or at least rebuild from the source for the latest one there is.

but I think you might lose the profiling functionality with the rpmfusion driver

Mark

6

Yes, I know about that possibility. But I need to develop on OpenCL/CUDA and driver in rpmfusion does not have those features. They are present in 190.18 of course and that one is not yet there.