NVIDIA HPC
NVIDIA HPC is a Comprehensive Suite of Compilers, Libraries and Tools for HPC
Using NVIDIA HPC
Loading the Module
To use NVIDIA HPC in our HPC Cluster, one needs to load the corresponding module. To know the dependencies of NVIDIA HPC, run
module spider nvhpc
Load the appropriate module (version):
module load nvhpc/22.2
Running a Sample Job
Copy the sample C++ source code "sortParallel.cpp" from /usr/local/doc/NVHPC to your home directory:
cp /usr/local/doc/NVHPC/sortParallel.cpp .
Request a GPU node:
srun -p gpu --gres=gpu:1 --mem=8gb --pty bash
Load nvhpc module:
module load nvhpc
Compile the code using nvhpc:
nvc++ -stdpar -o cusort sortParallel.cpp
Run:
time ./cusort
output:
Testing with 1000000 doubles...
Serial: Lowest: 3348 Highest: 4.29497e+09 Time: 5.141516ms
Serial: Lowest: 3348 Highest: 4.29497e+09 Time: 2.559334ms
Serial: Lowest: 3348 Highest: 4.29497e+09 Time: 2.473932ms
Serial: Lowest: 3348 Highest: 4.29497e+09 Time: 2.676512ms
Serial: Lowest: 3348 Highest: 4.29497e+09 Time: 2.555062ms
real 0m0.773s
user 0m0.138s
sys 0m0.214s
For batch job submission, refer to HPC Interactive & Batch job submission.