[Wien] Parallel execution on new Intel CPUs

Wed Feb 22 15:16:24 CET 2023

Thank you very much.

No, the mpi-benchmark makes no sense without mpi.

If possible, it would be nice to have the following times too:

Prepare a case.klist with 16 times the Gamma point.
Then prepare a   .machines file with:

1:localhost
1:localhost
omp_lapw1:2

and run   x lapw1 -p

Then increase in .machines the localhost lines  to 4, 8 and 16 lines
(and the 16 line case with both, omp_lapw1:1 and 2)

Thanks
Peter Blaha

Am 22.02.2023 um 11:43 schrieb pluto via Wien:
> Dear Prof. Blaha, Prof. Marks, dear All,
> 
> Below some benchmark results. It seems that for a serial calculation 
> using 8 OMP threads is optimal. This probably has something to do with 
> having 8 fast and 8 slow cores.
> 
> Hardware:
> 13th Gen Intel(R) Core(TM) i7-13700K
> 64 GB of RAM DDR4-3600
> 2 TB drive Samsung NVMe
> ASUS Z690-P D4 mainboard
> 
> I also looked at mpi-benchmark, but I don't have mpi, so I think these 
> tests make no sense.
> 
> Let me know if I shoud add something to this.
> 
> Best,
> Lukasz
> 
> 
> 
> bash-5.1$ pwd
> (...)/WIEN2k_benchmark/Serial/test_case
> 
> bash-5.1$ export OMP_NUM_THREADS=1
> bash-5.1$ echo $OMP_NUM_THREADS
> 1
> bash-5.1$ x lapw1
>   LAPW1 END
> 12.567u 0.216s 0:12.82 99.6%    0+0k 464+37840io 2pf+0w
> 
> bash-5.1$ export OMP_NUM_THREADS=2
> bash-5.1$ echo $OMP_NUM_THREADS
> 2
> bash-5.1$ x lapw1
>   LAPW1 END
> 14.844u 0.248s 0:07.65 197.1%    0+0k 0+37840io 2pf+0w
> 
> 
> bash-5.1$ export OMP_NUM_THREADS=4
> bash-5.1$ echo $OMP_NUM_THREADS
> 4
> bash-5.1$ x lapw1
>   LAPW1 END
> 21.091u 0.372s 0:05.51 389.4%    0+0k 0+37840io 10pf+0w
> 
> bash-5.1$ export OMP_NUM_THREADS=6
> bash-5.1$ echo $OMP_NUM_THREADS
> 6
> bash-5.1$ x lapw1
>   LAPW1 END
> 27.765u 0.490s 0:04.87 580.0%    0+0k 0+37824io 19pf+0w
> 
> bash-5.1$ export OMP_NUM_THREADS=8
> bash-5.1$ echo $OMP_NUM_THREADS
> 8
> bash-5.1$ x lapw1
>   LAPW1 END
> 34.099u 0.605s 0:04.51 769.1%    0+0k 0+37824io 27pf+0w
> bash-5.1$ x lapw1
>   LAPW1 END
> 34.087u 0.616s 0:04.51 769.1%    0+0k 0+37824io 33pf+0w
> bash-5.1$ x lapw1
>   LAPW1 END
> 34.119u 0.629s 0:04.52 768.3%    0+0k 0+37824io 26pf+0w
> bash-5.1$ x lapw1
>   LAPW1 END
> 34.234u 0.579s 0:04.53 768.2%    0+0k 0+37824io 26pf+0w
> 
> bash-5.1$ export OMP_NUM_THREADS=12
> bash-5.1$ echo $OMP_NUM_THREADS
> 12
> bash-5.1$ x lapw1
>   LAPW1 END
> 61.638u 2.193s 0:05.54 1151.9%    0+0k 0+37840io 44pf+0w
> 
> bash-5.1$ export OMP_NUM_THREADS=16
> bash-5.1$ echo $OMP_NUM_THREADS
> 16
> bash-5.1$ x lapw1
>   LAPW1 END
> 82.629u 2.636s 0:05.55 1536.0%    0+0k 0+37840io 63pf+0w
> 
> bash-5.1$ export OMP_NUM_THREADS=24
> bash-5.1$ echo $OMP_NUM_THREADS
> 24
> bash-5.1$ x lapw1
>   LAPW1 END
> 86.794u 3.724s 0:05.48 1651.6%    0+0k 0+37840io 57pf+0w
> 
> 
> 
> 
> bash-5.1$ pwd
> (...)/WIEN2k_benchmark/mpi-benchmark
> bash-5.1$ export OMP_NUM_THREADS=1
> bash-5.1$ echo $OMP_NUM_THREADS
> 1
> bash-5.1$ x lapw1
>   LAPW1 END
> 117.827u 0.921s 1:58.88 99.8%    0+0k 432+162616io 2pf+0w
> 
> 
> 
> 
> On 2023-02-15 01:11, Laurence Marks wrote:
>> Two things:
>>
>> 1) The CPU you have looks interesting. Can you please run and post the
>> benchmark from the Wien2k page for different omp (and mpi would be
>> good). It would be good to know what the "Hybrid Core" architecture
>> does with Wien2k. For mpi elpa is much better -- it can also be better
>> for non-mpi.
>>
>> 2) It is established lore in the DFT community that increasing the
>> "smearing" assists convergence. However, not all lore is true. I am
>> aware of zero evidence for this with the current Wien2k mixer, so I
>> suggest sticking with room temperature rather than 1500K. More
>> important is a well-posed problem. For more see
>> http://www.numis.northwestern.edu/Presentations/DFT_Mixing_For_Dummies.pdf
>>
>> On Tue, Feb 14, 2023 at 5:18 PM pluto via Wien
>> <wien at zeus.theochem.tuwien.ac.at> wrote:
>>
>>> Dear Prof. Blaha,
>>>
>>> Thank you for comments.
>>>
>>> At the moment I have 56 k-points in a big slab of one of the ternary
>>>
>>> magnetic 2D materials. Perhaps I can reduce k-points, something to
>>> test.
>>> Also now I see that my 56 k-points are compatible with 1:localhost
>>> lines
>>> :-)
>>>
>>> Also, for now it does not want to converge after 40 iterations with
>>> TEMP
>>> 0.002, for a while I was trying TEMP 0.004, and now I am trying TEMP
>>>
>>> 0.01. Maybe I should start with a smaller slab...
>>>
>>> Some info you asked for:
>>>
>>> The i7-13700K CPU has 8 P-cores (fast) and 8 E-cores (slow), so 16
>>> total
>>> physical cores. Each P-core has 2 threads, so there are total of 24
>>> threads. Many other new Intel CPUs are the same. I don't think there
>>> is
>>> an easy way to enforce certain task on a certain core, and probably
>>> it
>>> makes no sense, because the CPU for sure has thermal control over
>>> different cores etc.
>>
>>  --
>>
>> Professor Laurence Marks
>> Department of Materials Science and Engineering
>> Northwestern University
>> https://scholar.google.com/citations?user=zmHhI9gAAAAJ&hl=en [1]
>> "Research is to see what everybody else has seen, and to think what
>> nobody else has thought", Albert Szent-Györgyi
>>
>> Links:
>> ------
>> [1] http://www.numis.northwestern.edu
>> _______________________________________________
>> 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
> _______________________________________________
> 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
Email: peter.blaha at tuwien.ac.at    WIEN2k: http://www.wien2k.at
WWW:   http://www.imc.tuwien.ac.at
-------------------------------------------------------------------------