Re: Mersenne: Testing times
Bill Rea wrote: Is this about right for this type of system? Details:- Ultra-5 270Mhz CPU, 128Mb RAM Solaris 7 gcc version 2.8.1 For a comparison, I've got some Ultra2s with 2x 200MHz CPUs under Solaris 2.5.1. One is double-checking some numbers in the 4.6M range (with a 256k FFT) and is getting approx 0.4s per iteration, and another is checking a number in the 2.4M range (128k FFT) and is getting 0.22s per iteration. MacLucasUNIX was compiled with egcs 1.1.2 with -mcpu=ultrasparc -O6 -funroll-loops -funroll-all-loops for options. For 7902277 I'd guess an FFT size of 512k, so you should be getting roughly double the iteration time divided by 270/200 for CPU speeed - roughly 0.6s per iteration at a guess. What compiler options were you using for gcc? I can send you my ultrasparc binary if you want to test that. 0.4 * 2 * (270/200) Also, I've tested MacLucasUNIX with the Sun CC compiler using the -fast option and it's about a third faster than when compiled with gcc, but there are warnings like:- The -fast option is unsuitable for programs that require strict conformance to the IEEE 754 Standard. Should I be using this option? Good question - sounds scary. Unless you get some informed opinion, I'd be staying clear of that... Simon. _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
Re: Mersenne: Testing times
On Wed, Aug 04, 1999 at 09:50:42PM -0600, Aaron Blosser [EMAIL PROTECTED] wrote: It's been noted that GCC isn't the best at compiling under Solaris, for this type of code anyway. Someone else (my brother?) will probably have more info on this. I admin several Sparc machines with Solaris which are currently running SETI@Home which I'd like to switch over to searching for primes. I was originally holding out until there was a PrimeNet client for Sparc, but what the hell - I log into the machines all the time anyways. However, I don't have a Sun compiler - does anyone have precompiled, best-optimized MacLucasUNIX binaries for Sparc processors? I have a Sparc 5, a couple of Sparc 10/20s, and four CPUs on an E450 which sit idle much of the time. Thanks, Bryan -- Bryan Fullertonhttp://www.samurai.com/ Core Competency Samurai Consulting "No, we don't do seppuku." Can you feel the Ohmu call? _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
RE: Mersenne: Testing times
-Original Message- From: Aaron Blosser [mailto:[EMAIL PROTECTED]] Sent: Wednesday, August 04, 1999 10:51 PM To: Mersenne@Base. Com Subject: RE: Mersenne: Testing times I've got several UltraSPARCs I've pressed into service and currently have 7 exponents checked out being tested with MacLucasUNIX. It's doing fractionally more than one iteration every second. That's judging by the times between checkpoint files at 5000 iterations between checkpoints. So far it's used 716.7 hours of CPU. At this rate it's probably done a bit over 2,500,000 iterations or is about one-third of the way there. Is this about right for this type of system? Details:- Ultra-5 270Mhz CPU, 128Mb RAM Solaris 7 gcc version 2.8.1 It's been noted that GCC isn't the best at compiling under Solaris, for this type of code anyway. Someone else (my brother?) will probably have more info on this. Yeh, Sun's CC is WAY better than GCC is for UltraSparc code. I think this is due to the fact that GCC (even 2.95) isn't optimized for V8plus instructions. Anyway, I was compiling with this... OPT=-fast -xO4 -xtarget=ultra -xarch=v8plusa -xsafe=mem -xdepend -xparallel -dalign (Note: If you are running Solaris 7, you can specify -xarch=v9 which might get you some performance gain) I think that I can speed it up by adding: -xlibmil And I'm not sure if IEEE754 is needed, this can be turned off by: -fsimple=1 -fns (I'm not sure what speed increase would be seen here) Umm... I haven't tried the -parallel flag, but I doubt there is much the compiler would find to parallelize in MacLucasUNIX Some day, I'll get the time to write the assembly for the Sparc and stuff... but first I need to figure out who in our company holds the Sun Workshop license, cuz I was just playing around with the demo... I'll try and find some place to put a binary for UltraSparcs sometime today and let the list know or whatever. Oh, plus I hacked in some timing code so as to get some benchmarks, nothing super special, but it is nice to know what iteration your box is on, and how fast its performing etc... Plus, some day I might get around to porting it to our AS/400 or something. That'd be interesting... -Jeremy _ Unsubscribe list info -- http://www.scruz.net/~luke/signup.htm Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers