Intel® Fortran Compiler
Build applications that can scale for the future with optimized code designed for Intel® Xeon® and compatible processors.

Polyhedron benchmarks

Michal_Kvasnicka
Beginner
1,560 Views

The new updated version of well known Polyhedron fortran compilers benchmark was released few days ago. See http://www.polyhedron.com/compare0html

Any comments from INTEL guys?

Michal

0 Kudos
1 Solution
Steven_L_Intel1
Employee
1,560 Views
The numbers I have for what will be 12.0.3 show a 78% improvement for fatigue, 42% for induct, 33% for tfft, 21% for rnflow - all compared to 12.0.0. ac does not yet show much improvement - I am told that it will be better in the next version (not update). On some platforms we expect to be #1 again as of 12.0.3 - the others will improve in the next version.

As always, you have choices in the compiler you use and we encourage you to try any and all compilers to see which best meet your needs. It is our goal to be the best compiler available on all platforms we support.

View solution in original post

0 Kudos
16 Replies
Steven_L_Intel1
Employee
1,560 Views
You may want to check again - Polyhedron notified us that they made an error in their scripts and the Intel compiler shows better in the new results, posted just yesterday. Yes, we see that the Absoft compiler is ahead of us slightly in performance on many of these tests and we found that some optimizations were inadvertently disabled in the 12.0 compiler. These are being fixed and we expect significant improvements in the future.

It is our goal to be the best performing compiler on both Intel and non-Intel processors.
0 Kudos
Michal_Kvasnicka
Beginner
1,560 Views
Hi Steve,

are there any performance enhancements regarding new 12.0.2 release of Intel fortran compiler on Polyhedron benchmark?

Is the new version able to produce better (faster) binaries for AC and FATIGUE codes?

Michal
0 Kudos
Steven_L_Intel1
Employee
1,560 Views
There are improvements in 12.0.2, and will be further improvements in 12.0.3.
0 Kudos
Michal_Kvasnicka
Beginner
1,560 Views
OK ... could you be a bit more specific?

Michal
0 Kudos
Steven_L_Intel1
Employee
1,561 Views
The numbers I have for what will be 12.0.3 show a 78% improvement for fatigue, 42% for induct, 33% for tfft, 21% for rnflow - all compared to 12.0.0. ac does not yet show much improvement - I am told that it will be better in the next version (not update). On some platforms we expect to be #1 again as of 12.0.3 - the others will improve in the next version.

As always, you have choices in the compiler you use and we encourage you to try any and all compilers to see which best meet your needs. It is our goal to be the best compiler available on all platforms we support.
0 Kudos
Michal_Kvasnicka
Beginner
1,560 Views
According to the Polyhedron benchmark published on http://polyhedron.com/compare0html is the IFORT compiler 12.0.3 number "One".

Congratulations!!!
0 Kudos
Steven_L_Intel1
Employee
1,560 Views
Thanks - and we'll continue to improve over time. Our goal is to be the fastest compiler on any platform we support.
0 Kudos
Steven_L_Intel1
Employee
1,560 Views
The Polyhedron site was updated today. I encourage everyone to go look...
0 Kudos
Michal_Kvasnicka
Beginner
1,560 Views

Steve, this is a big achievment ... congratulation!!! Guys you are doing realy good job!!!

I tested recent Intel fortran compiler 12.1 against PGI 11.8 on our in house codes and I can confirm, that Intel compiler provides binaries with excelent speed. There is probably no reason to use any other compiler!

0 Kudos
jimdempseyatthecove
Honored Contributor III
1,560 Views
Steve,

Good job (Intel) on the rankings for this benchmark. I will note a peculiarity with these benchmarks in particular with the TIFF2 results. The i5 vs Phenom indicate different disk and/or disk interface and/or RAID/non-RAID configuration 2.9x faster on Windows and 2.4x faster onLinux. Thismay be due to AVX utilization by the compileror disk subsystem utilization. It would be interesting to find out why.

Jim Dempsey
0 Kudos
Steven_L_Intel1
Employee
1,560 Views
Don't try to compare across systems. It is meant to compare compilers on the same system. Yes, there are OS and platform oddities that can be puzzling, such as why there is less of a gap on AMD Linux than on AMD Windows, but... We are looking at the cases where we're not the best on individual tests as well.
0 Kudos
jimdempseyatthecove
Honored Contributor III
1,560 Views
>>Don't try to compare across systems. It is meant to compare compilers on the same system

Quite true...
0 Kudos
JohnNichols
Valued Contributor III
1,560 Views
If you have the biggest, fastest or meanest then it feels good.

Congrats. Although I do miss my Microsoft version 3.1.3 running on my Compaq Portable with two floopy drives.

I could go and make a cup of tea, drink the cup of tea and have a biscuit all while it compiled, now it is so freaking fast I can barely take a sip of coke.

Coke applauds you , the British Tea Company is selling its Intel stock.

0 Kudos
Steven_L_Intel1
Employee
1,560 Views
Nice blog post by Intel's James Reinders that references these results and more.
0 Kudos
Michal_Kvasnicka
Beginner
1,560 Views
John, do not be sad, try to use: http://www.oldskool.org/pc/throttle

You can slow down your current superfast computer as you wish and enjoy your cup of tea during compilation.
0 Kudos
JohnNichols
Valued Contributor III
1,560 Views

Michal:

Thanks I loved that a lot.

But I am not sad:

1. My new Chinese daughter keeps me happy, she calls me ba ba , which is the Chinese for da da. It is fun to watch the expressions on people face as she says it and I turn around.

2. I am going to buy a sheep T Shirt. If I have to explain why then you probably have not read Church's book.

3. I am in the middle of calling the OSRUSBFX2 from a Windows program. It is a lot of fun. I want to thank all the people on this form that helped me learn Windows program and the excellent Tassie Book.

4. That I am developing the first version in C++ shows how much I have learned about C++ programming in the last few weeks.

JMN

0 Kudos
Reply