Intel Compiler?

[mjlef]

If anyone has access the the latest Intel C++ compiler, can they let me know if it makes faster executablea than gcc? Around 5 years ago it seemed gcc was faster, but before spending a lot for it, I was wondering what people have found for 64 bit chess programs.

[Dann Corbit]

I used the Intel compiler for some time, but we found a couple problems:

1. The Intel toolset does not work properly on machines that are not Intel CPU powered. Some features of the debugger and profiler are turned off, for instance.

2. We found a bug in code generation and they took eons to fix it.

On the other hand, it produced the best assembly code for Intel chip sets as far as speed goes.

We have not used it for some years now, so I do not know if these restrictions still apply.

[velmarin]

I would say that it is slower, not much, but slower.
With the Stockfish versions less than five percent on my machine an AMD Phenom a few years already.
You can try it, three months of free evaluation will suffice.

Speaking of the topic is very curious how Embarcadero RAD Berlin is based on Clang, Microsoft also has Clang support in its latest environments,
Even Intel with its free compiler.
But Clang also seems a bit slower than GCC.

[bhamadicharef]

What about Intel VTune Amplifier performance profiler ... It used to be popular in my (old) days...

[jdart]

VTune is a nice tool. But while they have tried to streamline it and it does have some pre-canned analyses, I would still say it is best for users who have a fairly deep understanding of Intel chip and system architecture. It is very Intel-specific.

I never really got a performance boost out of the Intel compiler on my code, compared to current-generation gcc and MSVC. Your mileage may vary though.

--Jon

[MikeB]

If anyone has access the the latest Intel C++ compiler, can they let me know if it makes faster executablea than gcc? Around 5 years ago it seemed gcc was faster, but before spending a lot for it, I was wondering what people have found for 64 bit chess programs.

I had the 2016 version for a bit before the license expired. It made very marginally faster single thread/core exe's 9 (2% max faster) and the multiple threads appeared t be worse than gcc. ( longer to reach depth x). Even clang now is pretty close to gcc which is pretty close to icc. As a hobbyist, it was clearly not worth purchasing.

I had an educators license, using the fictitious school name "Bear Mountain University". Before my father -in-law passed away, he always claimed he graduated from "Bear Mountain University". A high school graduate, he went on to run the largest oil refinery in Philadelphia as the Superintendent. To say he was a "quick study" would be an understatement. When I tried to re-up the license, it was denied ( for obvious reasons) . They do have a free linux version available for those who participate in open source projects on github. Not sure why they exclude the Mac folks who participate.

[tttony]

I have the ICC 2016 with MSVC 2013 and ICC it's slower than MSVC, my engine it's written in C99


With MSVC 2013

Code: Select all

info depth 8 score cp 0 nodes 82830362 qnodes 77604426 time 9968 nps 8309627 pv d2d4

With ICC 2016

Code: Select all

info depth 8 score cp 0 nodes 82830362 qnodes 77604426 time 10640 nps 7784808 pv d2d4

Maybe need some optimization tweaks