Call for testing gmp-pi

[deanjo]

In search of a better pi benchmark for PTS, I came across http://h2np.net/pi/. With some modifications that I have done I'm hoping this will replace super-pi as it is not optimized at all for processors. This one at least will apply the performance patches to gmp and give more accurate results as to a processors capability when optimized. The modified verstion can be found here:

http://www.filefactory.com/file/agf1819/n/gmp-pi_tar_gz

Please read the readme as it contains all that is needed to install. Don't worry everything it compiles is kept locally in the make directory and will not screw up any of your systems libraries.

Let me know if it compiles, what distro, system specs and results. It would also be helpful if you could make sure that it's maxing out the processor usage during testing

Thank you
Dean

Feedback always welcome.

Just to start off a few results of what I'm looking for:

openSUSE 11.1 64-bit, Phenom 9850 @ stock speeds
-compiles fine
Results from threads.txt
Threads: 4 Time: 131.305

openSUSE 11.1 64-bit 4200+ X2 @ stock speeds
-compiles fine
Results from threads.txt
Threads: 2 Time: 304.430

[deanjo]

openSUSE 11.1 64-bit, Phenom II 955 @ stock speeds
-compiles fine
Results from threads.txt
Threads: 4 Time: 99.003

[Aradreth]

Archlinux X86_64 Intel(R) Core(TM)2 Duo CPU T6400 @ 2.00GHz
-complied fine except for some odd reason it did not detect a C2D processor

Try to setup GMP for Intel Core2 or Xeon, wait....
Did not detect Intel Core 2 CPU.  Patch not installed
checking build system type... pentium3-unknown-linux-gnu
checking host system type... pentium3-unknown-linux-gnu

Threads: 2 Time: 798.969

[deanjo]

Archlinux X86_64 Intel(R) Core(TM)2 Duo CPU T6400 @ 2.00GHz
-complied fine except for some odd reason it did not detect a C2D processor

Try to setup GMP for Intel Core2 or Xeon, wait....
Did not detect Intel Core 2 CPU.  Patch not installed
checking build system type... pentium3-unknown-linux-gnu
checking host system type... pentium3-unknown-linux-gnu

Threads: 2 Time: 798.969

Interesting. What distro Aradreth?

Edit, nevermind just read it. Why is you proc being seen as a P3, hmmm?

[Zhick]

CPU ist an Intel C2D E6600 @ 2.4 Ghz. Gentoo ~amd64

CPU is Intel(R)
Try to setup GMP for Intel Core2 or Xeon, wait....
Detected Intel Core 2 CPU

Seems to patch and compile fine, but the compiler is called with mtune=k8? Wouldn't be mtune=core2 better?
Anyway: Threads: 2 Time: 397.813 sec
Edit: Wait what? The C2D is slower than the X2 4200+ in this bench? I guess somethings definitly not setup right.

[Dewni]

Ubuntu Jaunty 32-bit, Athlon64x2 5600+ @ stock speeds (2.9Ghz)
-compiles fine
Results from threads.txt
Threads: 2 Time: 696.551

[deanjo]

CPU ist an Intel C2D E6600 @ 2.4 Ghz. Gentoo ~amd64

Seems to patch and compile fine, but the compiler is called with mtune=k8? Wouldn't be mtune=core2 better?
Anyway: Threads: 2 Time: 397.813 sec
Edit: Wait what? The C2D is slower than the X2 4200+ in this bench? I guess somethings definitly not setup right.

No, the results are right (as far as AMD's being faster then their intel rivals). When it comes to GMP with assembly support the intels are much slower then the AMD's. Can you try the original to see if it calls mtune=k8 there.

http://h2np.net/pi/pi_thread_e.html

http://h2np.net/pi/pi_record_e.html

Seems there is a newer version of the original out as well now which no longer uses the patches.

[leef]

uname: Linux 2.6.26.3 i686 GNU/Linux
distro: Debian testing
CPU 96% gmppi, 4% other programs

Debian testing 32-bit, Athlon64 3200 @ 2390 MHz
-compiles fine
Results from threads.txt
Threads: 1 Time: 1690.708

By the way, I didn't understand what the intent of this new tool is. Is it a better way to stress the CPU? Anyway, good luck with it.

[deanjo]

uname: Linux 2.6.26.3 i686 GNU/Linux
distro: Debian testing
CPU 96% gmppi, 4% other programs

Debian testing 32-bit, Athlon64 3200 @ 2390 MHz
-compiles fine
Results from threads.txt
Threads: 1 Time: 1690.708

By the way, I didn't understand what the intent of this new tool is. Is it a better way to stress the CPU? Anyway, good luck with it.

Just general benchmarking, plus it really does push a CPU to it's max, something a lot of multi-threaded apps don't

[Zhick]

Seems there is a newer version of the original out as well now which no longer uses the patches.

Just tried 4.3.1. It compiles with -mtune=core2 by default and needs only 379 seconds to finish (with full kde running, contrary to 397s with your patched 4.2.2, mtune=k8 and no X Server running at all).
Edit: Just tried with 4.3.1 and -march=core2 and it finished in Time: 316.282 sec. I guess since your script is trying to optimize as much as possible you might as well add CFLAGS="-march=foo" to ./configure in your script.