icc really works pretty well

[jhu]

i compiled povray using gcc 2.95 and icc 7.0. the benchmark.pov scene was used for rendering.
cpu: 800mhz celeron
icc flags: -O3 -xK -ipo
gcc flags: the regular ones supplied

Total Time (gcc): 2 hours 45 minutes 19.0 seconds (9919 seconds)
Total Time (icc): 1 hours 51 minutes 25.0 seconds (6685 seconds)

 

[Bremen]

This isn't a surprise, icc is optimized specifically for intel chips (x86 in particular). gcc on the other hand supports a vast miriad of differing architectures. The only problem with icc is programs aren't written with it in mind, and occasionally something won't compile correctly. Although tighter adherence to standards in gcc itself should help this.

 

[Armitage]

yea, icc rocks. I got about 20% improvement on one of my sims vs. gcc 3.1. And that was on a PIII, supposed to be even better on a P4.

 

[drag]

-strike that.. sorry I am just being brain-dead again

 

[n0cmonkey]

Link?

 

[Bremen]

drag, you do realize, that we are all extremely curious as to what your brain dead mind said? :0)

 

[jhu]

did some more tests on my own scenes. turns out that it all depends on the scene. gcc is faster in some, icc in others.

 

[GigaCluster]

I'd be more interested in comparing the latest version of GCC to ICC. 2.95 is quite old.

 

[drag]

How big is the binary it produced? Is there any (noticable) difference in stability?

 

[Armitage]

Originally posted by: drag
How big is the binary it produced? Is there any (noticable) difference in stability?

I've had zero stability problems with it ... but then, I was only using the demo version for some limited benchmarking purposes. The binaries produced were significantly bigger IIRC. We're getting the Intel compilers & performance libraries with our new Xeon cluster

There's definitely some wiggle room though. I've seen some benchmarks, and heard second-hand of cases where gcc was as good or faster then icc.

Intel Compilers

 

[Buddha Bart]

brain dead here too

 

[Armitage]

another bit ... required only very minimal code changes to build my stuff with icc vs. gcc 3.1. Can't recall what they were off hand.

 

[Armitage]

Originally posted by: Buddha Bart
who care's how long it takes to compile, the question is does it run faster after it's been compiled.

Where did anybody mention compile time? FWIW, basic compiles were faster then gcc, but it can take longer when you get into profile based builds, etc.

 

[n0cmonkey]

Originally posted by: ergeorge

Originally posted by: Buddha Bart
who care's how long it takes to compile, the question is does it run faster after it's been compiled.

Where did anybody mention compile time? FWIW, basic compiles were faster then gcc, but it can take longer when you get into profile based builds, etc.

Compile time is important to developers, especially on old platforms. That is one reason OpenBSD hasn't yet moved to 3.x.

 

[Armitage]

Yea, compile time is important, although far less then it used to be given the faster hardware, etc.. Used to go to lunch when it was time to rebuild a large project, or home for the night for really big ones. But my tolerance for build time has adjusted accordingly ... my current project only takes about 30 seconds to build & link but it's still tedious enough for me to flip over to /. or AT for a few seconds. And for the big projects, make -j works well on MP rigs, and ppmake on clusters

 

[n0cmonkey]

Originally posted by: ergeorge
Yea, compile time is important, although far less then it used to be given the faster hardware, etc.. Used to go to lunch when it was time to rebuild a large project, or home for the night for really big ones. But my tolerance for build time has adjusted accordingly ... my current project only takes about 30 seconds to build & link but it's still tedious enough for me to flip over to /. or AT for a few seconds. And for the big projects, make -j works well on MP rigs, and ppmake on clusters

There isn't much new hardware in the sparc4m realm. Remember, some people still use older hardware, so compile time *is* important for the developers that actually think about more than x86.

 

[Armitage]

What's special about sparc4m?
In any case, my comment is generic and has nothing to do with x86. In fact, much of my work is done on an SGI. My point is simply that faster hardware & faster compilers make compile time less of an issue then it used to be. Of course, that doesn't help if you're stuck on antique or exceptionally slow hardware for some reason.

BTW ... make -j is especially nice on an 8 way O2K

 

[n0cmonkey]

Originally posted by: ergeorge
What's special about sparc4m?

It is old and basically discontinued. It is definitely a great arch, just really slow.

In any case, my comment is generic and has nothing to do with x86. In fact, much of my work is done on an SGI. My point is simply that faster hardware & faster compilers make compile time less of an issue then it used to be. Of course, that doesn't help if you're stuck on antique or exceptionally slow hardware for some reason.

BTW ... make -j is especially nice on an 8 way O2K

I'll have to try that on a 12 way 4800

 

[Armitage]

Originally posted by: n0cmonkey

Originally posted by: ergeorge
What's special about sparc4m?

It is old and basically discontinued. It is definitely a great arch, just really slow.

In any case, my comment is generic and has nothing to do with x86. In fact, much of my work is done on an SGI. My point is simply that faster hardware & faster compilers make compile time less of an issue then it used to be. Of course, that doesn't help if you're stuck on antique or exceptionally slow hardware for some reason.

BTW ... make -j is especially nice on an 8 way O2K

I'll have to try that on a 12 way 4800

See? What are you complaining about??

 

[n0cmonkey]

Originally posted by: ergeorge

Originally posted by: n0cmonkey

Originally posted by: ergeorge
What's special about sparc4m?

It is old and basically discontinued. It is definitely a great arch, just really slow.

In any case, my comment is generic and has nothing to do with x86. In fact, much of my work is done on an SGI. My point is simply that faster hardware & faster compilers make compile time less of an issue then it used to be. Of course, that doesn't help if you're stuck on antique or exceptionally slow hardware for some reason.

BTW ... make -j is especially nice on an 8 way O2K

I'll have to try that on a 12 way 4800

See? What are you complaining about??

I don't think they will let me take that machine home...

 

[jhu]

It is old and basically discontinued. It is definitely a great arch, just really slow.

you could try cross-compiling