[Wien] Serial Benchmark i9-9900K

Mon Apr 19 10:41:33 CEST 2021

Thank you very much for your data. I've put them on our website together 
with our own I7-7820X (3.6 GHz) machines.

It shows impressively that for a single core job the cpu frequency is 
important (15.4 vs. 23.5 sec with 4.7 vs. 3.6 GHz), while under full 
load (8 cores - 6.4 vs. 6.1 sec !!!) the higher frequency does not help 
at all since the memory bus speed determines the overall performance.

In a "throughput test" (many k-points) it is still the best to run 8 
parallel jobs (for 8 k-points, with one thread) or 4 parallel jobs with 
2 threads (OMP_NUM_THREADS=2), depending on the speed of your disk/NFS 
system.

If anybody has one of the new 10th generation CPUs, it would be 
interesting to see their performance.

Best regards
Peter Blaha

Am 18.04.2021 um 21:14 schrieb mpeet at mrc-lmb.cam.ac.uk:
> Hi, I ran serial benchmark after compiling with oneapi intel fortran. 
> The Eigen values are similar to the test cases, although there are 
> additional lines.
> 
> I'm still unsure if it is useful to run the parallel benchmark on a 
> single machine.
> 
> Intel Core i9-9900K (4.70 Ghz)  ASUSTek PRIME Z390-P memory 32Gib 2 x  
> DIMM DDR4 Synchronous 2666 MHz in bank 0 and 2 (HMA82GU6JJR8N-VK).
> 
> ifort (IFORT) 2021.1 Beta 20201112
> 
> cc gcc (Debian 8.3.0-6) 8.3.0
> 
> cores   time(s)
> #N      Wien21.1   Wien19.2
> 1        15.45 15.39
> 2 9.14      9.1
> 3        7.64
> 4 7.09      7.08
> 6        6.57
> 8 6.43      6.42
> 16       6.53 6.54
> 
> Wien2k  19.2
> 
> 
> OMP_NUM_THREADS=1
> 15.234u 0.115s 0:15.39 99.6%    0+0k 704+37840io 3pf+0w
> 
> OMP_NUM_THREADS=2
> 17.708u 0.191s 0:09.10 196.5%    0+0k 0+37824io 0pf+0w
> 
> OMP_NUM_THREADS=4
> 27.267u 0.188s 0:07.08 387.5%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=8
> 48.819u 0.536s 0:06.42 768.5%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=16
> 56.134u 1.275s 0:06.54 877.6%    0+0k 0+37840io 0pf+0w
> 
> 
> 
> WIEN2k_21.1 (Release 14/4/2021)
> 
> serial benchmark test x lapw1, matrix size 3481
> OMP_NUM_THREADS=1
> 15.230u 0.200s 0:15.45 99.8%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=2
> 17.746u 0.264s 0:09.14 196.9%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=3
> 22.057u 0.344s 0:07.64 293.0%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=4
> 27.145u 0.399s 0:07.09 388.2%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=6
> 37.519u 0.567s 0:06.57 579.4%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=8
> 49.004u 0.620s 0:06.43 771.6%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=10
> 50.762u 0.939s 0:06.49 796.4%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=12
> 53.143u 1.075s 0:06.54 828.8%    0+0k 0+37840io 0pf+0w
> 
> OMP_NUM_THREADS=16
> 55.914u 1.493s 0:06.53 879.0%    0+0k 0+37840io 0pf+0w
> 
> 
> grep HORB *output1*
> test_case.output1_10core:       TIME HAMILT (CPU)  =     7.6, HNS = 7.3, 
> HORB =     0.0, DIAG =    36.7, SYNC =     0.0
> test_case.output1_10core:       TIME HAMILT (WALL) =     0.8, HNS =     
> 0.9, HORB =     0.0, DIAG =     4.7, SYNC =     0.0
> test_case.output1_16core:       TIME HAMILT (CPU)  =    11.5, HNS =     
> 8.9, HORB =     0.0, DIAG =    36.7, SYNC =     0.0
> test_case.output1_16core:       TIME HAMILT (WALL) =     0.8, HNS =     
> 0.9, HORB =     0.0, DIAG =     4.7, SYNC =     0.0
> test_case.output1_1core:       TIME HAMILT (CPU)  =     2.7, HNS = 1.6, 
> HORB =     0.0, DIAG =    11.0, SYNC =     0.0
> test_case.output1_1core:       TIME HAMILT (WALL) =     2.7, HNS =     
> 1.6, HORB =     0.0, DIAG =    10.9, SYNC =     0.0
> test_case.output1_2core:       TIME HAMILT (CPU)  =     2.8, HNS = 2.0, 
> HORB =     0.0, DIAG =    12.9, SYNC =     0.0
> test_case.output1_2core:       TIME HAMILT (WALL) =     1.4, HNS =     
> 1.0, HORB =     0.0, DIAG =     6.5, SYNC =     0.0
> test_case.output1_4core:       TIME HAMILT (CPU)  =     3.5, HNS = 3.4, 
> HORB =     0.0, DIAG =    20.3, SYNC =     0.0
> test_case.output1_4core:       TIME HAMILT (WALL) =     0.9, HNS =     
> 0.9, HORB =     0.0, DIAG =     5.1, SYNC =     0.0
> test_case.output1_8core:       TIME HAMILT (CPU)  =     5.5, HNS = 7.0, 
> HORB =     0.0, DIAG =    36.7, SYNC =     0.0
> test_case.output1_8core:       TIME HAMILT (WALL) =     0.7, HNS =     
> 0.9, HORB =     0.0, DIAG =     4.7, SYNC =     0.0
> 
> Best wishes,
> 
> Mathew
> 
> 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> SEARCH the MAILING-LIST at:  http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
> 

-- 
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-165300             FAX: +43-1-58801-165982
Email: blaha at theochem.tuwien.ac.at    WIEN2k: http://www.wien2k.at
WWW:   http://www.imc.tuwien.ac.at
-------------------------------------------------------------------------