I don't think there is any mistake: when GPU's are used, OpenMP has the rather marginal role of speeding up some calculations that are not GPU-accelerated
Paolo On Mon, Aug 2, 2021 at 1:01 AM Takahiro Chiba < [email protected]> wrote: > Dear experienced users, > > I have trouble in utilizing OpenMP with my compilation. From the > output file, pw.x 6.8 recognizes "OMP_NUM_THREADS=2", but it took same > time as "OMP_NUM_THREADS=1", and according to PBS batch queue, only > 100% (not 200%) of CPU is used. Therefore, QE 6.8 with GPU is not as > fast as expected. > > I used nvidia HPC SDK 20.9, cuda 10.1, and Intel MKL 2021.2. The node > has two Xeon Gold 6248, one Tesla V100 32GB, and 768GB of RAM. > Benchmark results and make.inc are attached as tarball. > > Could you please point out my mistake? > > ---Sender--- > Takahiro Chiba > 1st-year student at grad. school of chem. sci. and eng., Hokkaido Univ. > Expected graduation date: Mar. 2023 > [email protected] > ----- > _______________________________________________ > Quantum ESPRESSO is supported by MaX (www.max-centre.eu) > users mailing list [email protected] > https://lists.quantum-espresso.org/mailman/listinfo/users -- Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche, Univ. Udine, via delle Scienze 206, 33100 Udine, Italy Phone +39-0432-558216, fax +39-0432-558222
_______________________________________________ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list [email protected] https://lists.quantum-espresso.org/mailman/listinfo/users
