I'd like to run gpu grid on my server. Do I require X windows to run it? Will nvidia drivers work fine?

I'd like to run gpu grid on my server. Do I require X windows to run it? Will nvidia drivers work fine?

No you don't have to run X, you just need to load the nvidia kernel module and create the device nodes, before starting the client.

#!/bin/bash

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 $N`; do
    mknod -m 660 /dev/nvidia$i c 195 $i
  done

  mknod -m 660 /dev/nvidiactl c 195 255
else
  exit 1
fi

---

Crunching on Linux: Fedora 11 x86_64 / nVidia 185.18.36 driver / CUDA 2.2

Thank you!

Ok, I've tried the script, named it nvidia on ubuntu-server, and this is the output:

mknod: `/dev/nvidia0': File exists
mknod: `/dev/nvidiactl': File exists

But after running boinc I get:

01-Oct-2009 17:36:11 [---] No CUDA-capable NVIDIA GPUs found
01-Oct-2009 17:36:11 [---] No coprocessors

Ok, I've tried the script, named it nvidia on ubuntu-server, and this is the output:

mknod: `/dev/nvidia0': File exists
mknod: `/dev/nvidiactl': File exists

OK, so module is loaded and the /dev nodes already exist.

But after running boinc I get:

01-Oct-2009 17:36:11 [---] No CUDA-capable NVIDIA GPUs found
01-Oct-2009 17:36:11 [---] No coprocessors

Have you got the CUDA toolkit installed? libcudart.so is available for runtime binaries to find?

[clivem@c7p6t:~]$ ldconfig -p | grep cudart
        libcudart.so.2 (libc6,x86-64) => /usr/lib64/nvidia/libcudart.so.2
        libcudart.so.2 (libc6) => /usr/lib/nvidia/libcudart.so.2
        libcudart.so (libc6,x86-64) => /usr/lib64/nvidia/libcudart.so
        libcudart.so (libc6) => /usr/lib/nvidia/libcudart.so

EDIT: If you think that's not the problem, grab deviceQuery-2.2.tgz, extract, make sure the 2 files are executable (or chmod them) and run 'deviceQuery'. (This uses the libcudart.so that'll prove the lib is available from the toolkit install.) If that fails, run 'deviceQueryDrv' which doesn't use the toolkit, just the driver libs. (If that succeeds but the former has failed you need to make sure libcudart.so is installed and in the LD path.)

Assuming, that a CUDA capable device is reported from 'deviceQuery', BOINC should find it OK.

---

Crunching on Linux: Fedora 11 x86_64 / nVidia 185.18.36 driver / CUDA 2.2

No, I haven't installed cudatoolkit. I needed to install just nvidia drivers when I was using ubuntu with gui, so I did the same with ubuntu server.
What is the difference?

No, I haven't installed cudatoolkit. I needed to install just nvidia drivers when I was using ubuntu with gui, so I did the same with ubuntu server.
What is the difference?

Erm, I don't know a great deal about the Ubuntu side of things, but I don't understand how that could have been working without the libcudart.so, which is not a part of the driver release. You have to get it from the toolkit.

My recollection from the BOINC code is that it dlopens libcudart.so. If it can't find it then you're going to be told that CUDA is not available or no CUDA devices found. (They've been messing with the messages recently, so what you get depends on the version of BOINC you have installed.)

EDIT: Assuming you're running the 185.xx driver release (CUDA 2.2) you can d/l the toolkit, cudatoolkit_2.2_linux_64_ubuntu8.10.run

---

Crunching on Linux: Fedora 11 x86_64 / nVidia 185.18.36 driver / CUDA 2.2

double post...

Ok, here it is. On a computer where ubuntuGUI/boinc/cuda works. I downloaded the drivers:
NVIDIA-Linux-x86_64-190.32-pkg2.run
and BOINC/cuda works, in BOINC directory I have file libcudart.so and the test you provided gives:

CUDA Device Query (Driver API) statically linked version 
There is 1 device supporting CUDA

Device 0: "GeForce 8600 GT"
  CUDA Capability Major revision number:         1
  CUDA Capability Minor revision number:         1
  Total amount of global memory:                 536150016 bytes
  Number of multiprocessors:                     4
  Number of cores:                               32
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       16384 bytes
  Total number of registers available per block: 8192
  Warp size:                                     32
  Maximum number of threads per block:           512
  Maximum sizes of each dimension of a block:    512 x 512 x 64
  Maximum sizes of each dimension of a grid:     65535 x 65535 x 1
  Maximum memory pitch:                          262144 bytes
  Texture alignment:                             256 bytes
  Clock rate:                                    1.19 GHz
  Concurrent copy and execution:                 Yes
  Run time limit on kernels:                     Yes
  Integrated:                                    No
  Support host page-locked memory mapping:       No
  Compute mode:                                  Default (multiple host threads can use this device simultaneously)

Test PASSED

On ubuntu server I also installed the same drivers and it gives:

/deviceQueryDrv 
CUDA Device Query (Driver API) statically linked version 
Cuda driver error 3 in file 'deviceQueryDrv.cpp' in line 76.

So that is what is strange to me, the same procedure gives a different result. And when I used ubuntu desktop version on the same computer everything worked fine on that computer, but now when I installed server version I get this problem.

So you think that installing the cuda toolkit will solve the problem?

I installed the cudatoolkit the same problem arises:

/deviceQueryDrv
CUDA Device Query (Driver API) statically linked version
Cuda driver error 3 in file 'deviceQueryDrv.cpp' in line 76

I installed the cudatoolkit the same problem arises:

/deviceQueryDrv
CUDA Device Query (Driver API) statically linked version
Cuda driver error 3 in file 'deviceQueryDrv.cpp' in line 76

OK, what version of the nVidia driver did you install on the server?

What is the output from 'ldd ./deviceQueryDrv'?

---

Crunching on Linux: Fedora 11 x86_64 / nVidia 185.18.36 driver / CUDA 2.2

/deviceQueryDrv
CUDA Device Query (Driver API) statically linked version
Cuda driver error 3 in file 'deviceQueryDrv.cpp' in line 76

More.....

Error 3 == CUDA_ERROR_NOT_INITIALIZED. So the nvidia.ko is not loaded?

Make sure it is loaded - 'lsmod | grep nvidia'

Make sure the /dev nodes exist - 'ls -l /dev/nv*'

---

Crunching on Linux: Fedora 11 x86_64 / nVidia 185.18.36 driver / CUDA 2.2