I am trying to get CUDA5 to work on a shared cluster.
When I invoke the local gcc to build large applications it works just fine. When I use the CUDA wrapper I get an error saying that it can’t find certain parts of glib.
Why is CUDA looking at /lib64/libc.so.6, if I specified the local gcc library directory?
[uid002@n001 cuda5test]$ ldd /home/ex/uid002/cuda/lib64/libcudart.so
/home/ex/uid002/cuda/lib64/libcudart.so: /lib64/libc.so.6: version `GLIBC_2.14' not found (required by /home/ex/uid002/cuda/lib64/libcudart.so)
linux-vdso.so.1 => (0x00007fff277ff000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007f5328da6000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f5328b89000)
librt.so.1 => /lib64/librt.so.1 (0x00007f5328981000)
libstdc++.so.6 => /home/ex/uid002/gcc/lib64/libstdc++.so.6 (0x00007f532867d000)
libm.so.6 => /lib64/libm.so.6 (0x00007f53283f8000)
libgcc_s.so.1 => /home/ex/uid002/gcc/lib64/libgcc_s.so.1 (0x00007f53281e3000)
libc.so.6 => /lib64/libc.so.6 (0x00007f5327e67000)
/lib64/ld-linux-x86-64.so.2 (0x000000335ae00000)
And
[uid002@n001 cuda5test]$ echo $LD_LIBRARY_PATH
/home/ex/uid002/cuda/lib64:/home/ex/uid002/gmp/lib:/home/ex/uid002/mpfr/lib:/home/ex/uid002/mpc/lib:/home/ex/uid002/gcc/lib64
I have tried to reinstall the CUDA package, and this was no help.
This error means that the
libcudart.solibrary was linked on a system withglibc-2.14or later. You can only use that library if your system (both at link time and at runtime) hasglibcversion >= 2.14. Your system does not satisfy prerequisite forlibcudart.so. You’ll have to use older version oflibcudart.so, or upgrade your system.(Be careful: upgrading
glibcincorrectly is a sure way to render the system un-bootable.)Glibcis not part ofGCCdistribution, and is completely independent. “local gcc library directory” is unlikely to havelibc.soin it.