Any reason for supporting only a 2yr old linux system like Fedora 13 and not Fedora 14 nor the most recent 15?
Your latest CUDA Toolkit only supports EoLed Fedora 13!
Same for SUSE 11.2 … guys the latest is 11.4 …
Same for Ubuntu 10.10… the latest is 11.04
Is it really so hard to change two lines, run make and give us the proper libraries for each OS?
Actually the latest CUDA and OpenCL libraries can work on the latest linux distributions. I use openSUSE 11.4 and all OpenCL samples can be complied and run. The only problem is the CUDA C samples can not be complied which need old GCC, that is why Nvidia tells us the CUDA toolkit only support those old distribution. Someone said there are some technical reasons for using old GCC. I don’t know when Nvidia would like to solve it.
In my specific case, it does not work with Fedora 15 due to GLibc mismatch. Also, if you compile your program with the NVIDIA openCL sdk it won’t work with an ATI card because they link using libOpenCL.so.1 instead of libOpenCL.so.
CUDA 4.0 toolkit misses libOpenCL.so apparently too.
It’s supposed that the SDK could be used to use other platform’s implementations but, in practise, not.
Not only the CUDA toolkit but also the GPU computing SDK. It’s a dissaster, nothing compiles. The big question is why they support a 2yr old almost EOLed distro and not the current ones.
I have not used Fedora 15 and know nothing about this problem. Maybe you can try OpenSUSE 11.4( I like it External Image ).
In my mind compiling OpenCL program on Nvidia platform has nothing to do with CUDA toolkit or openCL SDK. It needs libOpenCL.so and libcuda.so which are all included in Nvidia driver. So if you have the Nvidia driver installed, you should be able to compile OpenCL program.
And the Nvidia’s OpenCL SDK can only used on Nvidia platform. And if you want your program to be used on the other platforms, you’d better recompile it.
For GPU computing SDK, what is the error message?
/usr/local/cuda/include/host_config.h:82:2: error: #error – unsupported GNU version! gcc 4.5 and up are not supported!
In file included from /usr/lib/gcc/i686-redhat-linux/4.6.0/…/…/…/…/include/c++/4.6.0/i686-redhat-linux/bits/gthr.h:162:0,
from /usr/lib/gcc/i686-redhat-linux/4.6.0/../../../../include/c++/4.6.0/ext/atomicity.h:34,
from /usr/lib/gcc/i686-redhat-linux/4.6.0/../../../../include/c++/4.6.0/bits/ios_base.h:41,
from /usr/lib/gcc/i686-redhat-linux/4.6.0/../../../../include/c++/4.6.0/ios:43,
from /usr/lib/gcc/i686-redhat-linux/4.6.0/../../../../include/c++/4.6.0/ostream:40,
from /usr/lib/gcc/i686-redhat-linux/4.6.0/../../../../include/c++/4.6.0/iostream:40,
from interval.cu:26:
/usr/lib/gcc/i686-redhat-linux/4.6.0/…/…/…/…/include/c++/4.6.0/i686-redhat-linux/bits/gthr-default.h:251:1: error: paste “_gthrw” y "/* Android’s C library does not provide pthread_cancel, check for
`pthread_create’ instead. */" it does not give a valid preprocessing element
In file included from interval.cu:32:0:
cpu_interval.h:19:0: warning: redefined “__USE_ISOC99” [activated by default]
/usr/include/features.h:267:0: note: this is the ubication of the previous definition
make[2]: *** [obj/i386/release/interval.cu_13.o] Error 1
make[2]: exiting directory `/home/santy/NVIDIA_GPU_Computing_SDK/C/src/Interval’
make[1]: *** [src/Interval/Makefile.ph_build] Error 2
make[1]: exiting directory `/home/santy/NVIDIA_GPU_Computing_SDK/C’
According to the error message, you were compiling CUDA C samples. As I said, compiling CUDA code is only supported by old GCC.
If you just want to use OpenCL, you should enter into the folder ‘/home/santy/NVIDIA_GPU_Computing_SDK/OpenCL’ and type make in terminal to compile OpenCL Samples. It should work.
As I said before, NVIDIA_GPU_Computing_SDK is not necessary for compiling your OpenCL programs. It only has some samples. All you need is the driver and OpenCL header files which can be downloaded from Khronos.
Same errors for OpenCL. GCC version not supported.
In my OpenSUSE 11.4 GCC version is 4.5.1. Maybe 4.6 has some differences. I have no idea about that. External Image
Now I have turned to Intel OpenCL SDK. I will come back when Nvidia releases their OpenCL 1.1 Implementation.
actually its quite easy to redirect nvidia to use that old compiler
details at The Official NVIDIA Forums | NVIDIA
enjoy - scott stensland