◦
◦
–
mcelog package to decode kernel machine check log on x86 machines used by CT memory error check:
–
Additional requirements for accelerator support
In addition to those listed in “Requirements for Cluster Test RPM” (page 5), the following are requirements for clusters with accelerators:
•To work with NVIDIA accelerators, the following NVIDIA software packages must be installed and functional. They should be installed in the order listed:
1.NVIDIA-Linux-x86_64-319.32.run (Or newer)
The installation of NVIDIA drivers must be done with the X server shut down. This can be done by putting the node into runlevel 3 using the telinit command.
2.cuda_5.0.35_linux_64_<OS_used>.run
Install the toolkit and the samples. The nbody sample must be compiled as it is used by Nvidia Accelerator tests.
Once the cuda toolkit is installed, the user environment must include PATHs to the new directories. One way of doing this is to create the file /etc/profile.d/cuda.sh with the following content (assuming cuda is installed in /usr/local/cuda).
PATH=/usr/local/cuda/bin:${PATH}
LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
3.gpucomputingsdk_4.2.9_linux.run
Once installed, the applications must be compiled as follows.
#cd NVIDIA_GPU_Computing_SDK_C
#make
If compile errors related to missing libraries are reported, you will have to create symlinks from existing libraries to the missing names. For example:
#ln
Create the missing symlinks as needed until all the applications have been built.
Cluster Test requires the nbody and deviceQuery binaries present in the SDK directory C/ bin/linux/release.
Add the path to the newly built binaries to your user environment.
To verify the NVIDIA components are working, run the
Getting the CT RPM kit
1.Download the Cluster Test file
HP_Cluster_Test_Software_Vx.x_Clusterx64_rpm_vx.x_XXXX.bin from
2.Set the binary file as executable with the chmod command.
Additional requirements for accelerator support | 15 |