Bidders are requested to do the benchmarking for QE package using following two input sets. Bidder has to give the benchmarking results for both the input sets and in a format mentioned in the EoI.
Please follow the instructions below to run the jobs for benchmarking of
Quantum ESPRESSO. In both the following cases, we are looking for how QE
scales across different number of cores/nodes.
- "QE_small.tar.bz2" contains the input files to run the jobs upto 16 nodes (1,2,4,8, and 16).
In this case, we want you to run the job within single node as well as
across the nodes (maximum upto 16 nodes). The parameter "nbnd" should be
an integral multiple of number of cores used. In this particular case,
it should always be a number greater than 825.
- "QE_large.tar.bz2" contains the input files to run the jobs on larger number
of nodes. These jobs have high memory requirements. We would like you to run
these particular jobs on N/4, N/3 and N/2 nodes and see how the scaling goes.
The parameter "nbnd" should be an integral multiple of number of cores used.
In this particular case, it should always be a number greater than 7206.
Basic commands for executing the job
mpirun -np [No of Cores] pw.x > OUT
or
mpirun -np [No of Cores] pw.x -input scf.in > OUT
The screen-output will be saved in "OUT" file.
Instruction Regarding Submission of Benchmark Results for QE
There is no need to submit wave function files along with benchmark results, because the size of those files are large. But Vendors needs to submit Standard screen-output along with Benchmark results (i. e. OUT file, see above section).
Important Instruction
For any clarification regarding QE benchmarking, the bidders are advised
to contact
Ms. Deepashri Saraf : deepashri@cms.unipune.ac.in
Prof. Anjali Kshirsagar : anjali@cms.unipune.ac.in
Dr. Deepak Bankar : deepak@cms.unipune.ac.in
