[Wien] Intel compilers
Gavin Abo
gabo13279 at gmail.com
Sun Jul 18 16:27:09 CEST 2021
I haven't did a benchmark [1] on my computer using both gfortran and
ifort of oneAPI to compare, so I don't know quantitatively for my
system. I'm just using an AMD processor computer with 4 cores which
seems to be good enough for running a single small unit cell calculation
or for troubleshooting some WIEN2k issues. However, it is not so great
for doing today's cutting edge research calculations which can require
multiple calculations for structure optimization, convergence tests,
computationally expensive hybrid calculations, etc. For those, a hpc
(high performance computer) cluster is desirable.
Qualitatively, I haven't not noticed a difference in speed of
calculations between gfortran and ifort of the older Parallel Studio
with my computer. The main reason I had switched to gfortran was that
Intel with Parallel Studio had discontinued the free non-commercial
license and the non-commercial license was too expensive to purchase.
Though, I could now reconsider using Intel's compilers again as oneAPI
provides them free for non-commercial use again as seen on Intel's FAQ
page [2] were it states:
/Are all of the oneAPI Toolkits available for free?//
//
//Yes, all of the oneAPI Toolkits are available for free download and
use for commercial and non-commercial purposes./
A secondary reason was gfortran seemed more stable as WIEN2k seemed to
break every once in awhile with each new Intel compiler release such as
with file I/O as they were improving their compliance with Fortran
standards [3]. The WIEN2k developers were superb in addressing errors
from Intel's I/O changes but the memory issue [4,5] was something only
Intel could address. Since I could not use the newer Parallel Studio
version with the memory leak fix after the trial period expired and
there was not a free non-commercial license to continue using it, it was
only natural back then to use gfortran instead.
It is also hard to know if the speed of calculations on a computer
system will better with one compiler over another as it could depend on
the hardware, compiler settings, and software (e.g., libraries) used on
that system. Since each computer system can be unique, you would have
to test yourself which has the faster calculation speed for you.
If I recall correctly, another user benchmarked gfortran and ifort
getting similar speed results for the two. You can read more about at [6].
If your system has Intel processors, the Intel compilers/libraries are
probably better to use as you would expect the Intel software engineers
to have optimized those for their processors. On the other hand, the
Intel compilers/libraries might have just barely get working support by
Intel such that they may be less optimized for AMD processors. Refer to
[7] having the statement:
/
/
/Intel's compilers may or may not optimize to the same degree for
non-Intel microprocessors for optimizations that are not unique to Intel
microprocessors/
For an AMD processor computer system, you might recall that one WIEN2k
user was trying to improve the speed by using AOCL [8,9] instead of
Intel MKL but I don't know if they ever got it to work.
[1] http://susi.theochem.tuwien.ac.at/reg_user/benchmark/
[2]
https://software.intel.com/content/www/us/en/develop/articles/oneapi-commercial-faq.html
[3]
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg17542.html
[4]
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg18798.html
[5]
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg19081.html
[6]
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg16159.html
[7] https://en.wikipedia.org/wiki/Intel_C%2B%2B_Compiler
[8]
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg20424.html
[9] https://developer.amd.com/amd-aocl/
On 7/17/2021 1:17 PM, delamora wrote:
> I use intel compilers because they are faster than gfortran. Is this
> what you see or you find then to be with same speed?
>
> Saludos
>
> Pablo
>
> ------------------------------------------------------------------------
> *De:* Wien <wien-bounces at zeus.theochem.tuwien.ac.at> en nombre de
> Gavin Abo <gabo13279 at gmail.com>
> *Enviado:* viernes, 16 de julio de 2021 09:06 p. m.
> *Para:* wien at zeus.theochem.tuwien.ac.at <wien at zeus.theochem.tuwien.ac.at>
> *Asunto:* Re: [Wien] Intel compilers
>
> I have been using gfortran recently. So I don't know about oneAPI as
> I have never used it. According to [2] that was in the previous
> email, the oneAPI Base Toolkit would be needed for MKL and the oneAPI
> HPC toolkit would be needed for the C (icc) and Fortran (ifort)
> compiler (and Intel MPI if you need that).
>
>
> [1]
> https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg21134.html
> <https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg21134.html>
>
> [2] https://www.scivision.dev/intel-oneapi-fortran-install/
> <https://www.scivision.dev/intel-oneapi-fortran-install/>
>
>
> On 7/16/2021 8:00 PM, delamora wrote:
>> Thank you Gavin Abo and Laurence Marks.
>> It seems that I can download it from Mexico.
>> Now; I want it for the WIEN2k, so what package I download;
>> Base Kit or All Kit ?
>> I supose that the Base Kit would be OK, but then do I need the HCP
>> Kit or the loT Kit of the Render Kit?
>> I ask because before the oneAPI there was the Parallel Studio that
>> took a lot of memory in the computer.
>>
>> Saludos
>>
>> Pablo
>>
>> ------------------------------------------------------------------------
>> *De:* Wien <wien-bounces at zeus.theochem.tuwien.ac.at>
>> <mailto:wien-bounces at zeus.theochem.tuwien.ac.at> en nombre de
>> Laurence Marks <laurence.marks at gmail.com>
>> <mailto:laurence.marks at gmail.com>
>> *Enviado:* viernes, 16 de julio de 2021 08:28 p. m.
>> *Para:* A Mailing list for WIEN2k users
>> <wien at zeus.theochem.tuwien.ac.at>
>> <mailto:wien at zeus.theochem.tuwien.ac.at>
>> *Asunto:* Re: [Wien] Intel compilers
>> Note: this is for the US, probably not for every country.
>>
>> ---
>> Prof Laurence Marks
>> "Research is to see what everyone else has seen, and to think what
>> nobody else has thought", Albert Szent-Györgyi
>> www.numis.northwestern.edu <http://www.numis.northwestern.edu>
>>
>> On Fri, Jul 16, 2021, 20:24 Gavin Abo <gabo13279 at gmail.com
>> <mailto:gabo13279 at gmail.com>> wrote:
>>
>> The Intel compilers (e.g., C and Fortran) currently seem to be
>> free for everyone with oneAPI without Priority Support from Intel
>> as seen at:
>>
>>
>> https://software.intel.com/content/www/us/en/develop/articles/free-intel-software-developer-tools.html
>> <https://urldefense.com/v3/__https://software.intel.com/content/www/us/en/develop/articles/free-intel-software-developer-tools.html__;!!Dq0X2DkFhyF93HkjWTBQKhk!BE815WRgDxTHbLBQEba9pFnKi7uazdHVymBvAltAP4QMMloTB75ba_nEMROoxaYNBroi3Q$>
>>
>> If you need the Priority Support from Intel, you have to buy
>> that. Though, you could use the free 'Community Support' instead
>> of the 'Priority Support' as seen at:
>>
>> https://www.scivision.dev/intel-oneapi-fortran-install/
>> <https://urldefense.com/v3/__https://www.scivision.dev/intel-oneapi-fortran-install/__;!!Dq0X2DkFhyF93HkjWTBQKhk!BE815WRgDxTHbLBQEba9pFnKi7uazdHVymBvAltAP4QMMloTB75ba_nEMROoxaY3yW43-g$>
>> https://software.intel.com/content/www/us/en/develop/tools/oneapi/support.html
>> <https://urldefense.com/v3/__https://software.intel.com/content/www/us/en/develop/tools/oneapi/support.html__;!!Dq0X2DkFhyF93HkjWTBQKhk!BE815WRgDxTHbLBQEba9pFnKi7uazdHVymBvAltAP4QMMloTB75ba_nEMROoxaYq4nh_Uw$>
>>
>> On 7/16/2021 5:28 PM, delamora wrote:
>>> Dear WIEN community;
>>> Are there free intel compilers for academic use?
>>>
>>> Thanks
>>>
>>> Pablo
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20210718/c405a305/attachment.htm>
More information about the Wien
mailing list