Intel C++ 5.0 Compiler speed measuring

[Benedikt]

Hi,

i got the Intel C++ Compiler 5.0 today, and I tested it with FLOPS benchmark (and other stuff), and I noticed performance increases up to 55% (on my Athlon) over the Mingw Compiler 2.95.2-1!!!
Although the Athlon T-Bird 1200 isn't from Intel, there is some nice performance increase at some calculations. I compared Intel C++ 5.0 (optimized) with Mingw Compiler 2.95.2-1 (gcc for windows) and Intel C++ 5.0 (no optimizations) on my Athlon and on my old 366Mhz Pentium II - Notebook.
Wow - I know it isn't possible to get such high increases in standard software, but think of Windows XP compiled with this baby - it would probably run smooth on a Celeron 500 or something... 🙂
Do you have any experiences with new compilers, for example gcc 3.0 or other ones? How fast are they compared to the Intel (I've read that the Intel 5.0 is the fastest compiler around at the moment)? How much speed increase did you notice?

Here are my results:

AMD Athlon 1200, ASUS A7V133, 256 MB SDRAM PC133; Intel C++ Compiler 5.0, no optimizations

FLOPS Benchmark:

1st calculation: 892.5414
2nd calculation: 488.2834
3rd calculation: 918.7249
4th calculation: 767.6929
5th calculation: 786.1492
6th calculation: 849.1850
7th calculation: 274

AMD Athlon 1200, ASUS A7V133, 256 MB SDRAM PC133; Intel C++ Compiler 5.0, /O3 /QaxM (MMX) /Qunroll /Qipo

1st calculation: 890.3242
2nd calculation: 488.2834
3rd calculation: 945.7786
4th calculation: 770.0020
5th calculation: 821.8753
6th calculation: 867.5431
7th calculation: 274


AMD Athlon 1200, ASUS A7V133, 256 MB SDRAM PC133; Mingw 2.95.2-1, full optimization -fomit-frame-pointer -funroll-loops -O3

1st calculation: 573.5777
2nd calculation: 391.5228
3rd calculation: 755.8180
4th calculation: 687.2483
5th calculation: 700.0471
6th calculation: 796.7375
7th calculation: 267.6205

Pentium II 366, Sony VAIO PCG-F305, 128 MB SDRAM PC66; Intel C++ Compiler 5.0, no optimizations

1st: 85.7375
2nd: 67.9869
3rd: 110.4625
4th: 110.9506
5th: 100.6944

Pentium II 366, Sony VAIO PCG-F305, 128 MB SDRAM PC66; Intel C++ Compiler 5.0, /O3 /QaxM (MMX) /Qunroll /Qipo

1st: 163.7876
2nd: 92.9605
3rd: 192.4557
4th: 167.2255
5th: 128.5474

Pentium II 366, Sony VAIO PCG-F305, 128 MB SDRAM PC66; Mingw 2.95.2-1, full optimization -fomit-frame-pointer -funroll-loops -O3

1st: 121.4421
2nd: 97.5663
3rd: 160.0118
4th: 140.9226
5th: 119.8212
6th: 151.7301

 

[rc5]

How's that comparing with M$ VC++ 7.0?

 

[Benedikt]

I just know Visual C++ 6.0, which is approx. as fast as MingW.

 

[Remnant2]

The IntelC compiler is good stuff. In general you can get 30%+ over an antedated compiler, and then another 30% if you use profile-guided optimizations.

Compared to MSVC6, it's more like 5-10% faster without the PGO, can be more depending on the app. Profile-guided ops are the big deal, that improves performance significantly.

MSVC7 is supposed to be quite a bit improved, on-par with the Intel compiler without PGO. IntelC has the advantage that it can sometimes auto-vectorize loops, and use sse/sse2. OTOH, I played around with it for a while on a variety of apps, and its' not entirely stable -- you can get very strange program logic errors from some of the more aggressive optimizations.

 

[zephyrprime]

I've found the Intel compiler to be about 25% faster than VC6. I've also found that it is incompatible with some code.

 

[rc5]

So the next question is: Where can I get Intel C++ compiler for free? Well, what's its full name? 🙂