From owner-freebsd-questions Thu Feb 13 08:28:57 1997 Return-Path: Received: (from root@localhost) by freefall.freebsd.org (8.8.5/8.8.5) id IAA21673 for questions-outgoing; Thu, 13 Feb 1997 08:28:57 -0800 (PST) Received: from forwiss.tu-muenchen.de (root@forwiss.tu-muenchen.de [131.159.128.1]) by freefall.freebsd.org (8.8.5/8.8.5) with ESMTP id IAA21665 for ; Thu, 13 Feb 1997 08:28:48 -0800 (PST) Received: from suncog13.forwiss.tu-muenchen.de (hafner@suncog13.forwiss.tu-muenchen.de [131.159.128.67]) by forwiss.tu-muenchen.de (8.8.5/V5) with SMTP id RAA28349 for ; Thu, 13 Feb 1997 17:28:39 +0100 (MET) Received: by suncog13.forwiss.tu-muenchen.de (SMI-8.6/SMI-4.0) id RAA23584; Thu, 13 Feb 1997 17:28:35 +0100 To: freebsd-questions@freebsd.org Subject: Re: UltraSPARC and MicroSPARC vs Pentium Pro ? References: <199702120330.TAA15056@f30.hotmail.com> <199702121833.KAA18506@freefall.freebsd.org> From: Walter Hafner Date: 13 Feb 1997 17:28:35 +0100 In-Reply-To: jmb@freefall.freebsd.org's message of 12 Feb 1997 19:44:28 +0100 Message-ID: Lines: 108 X-Mailer: Gnus v5.2.25/XEmacs 19.14 Sender: owner-questions@freebsd.org X-Loop: FreeBSD.org Precedence: bulk jmb@freefall.freebsd.org (Jonathan M. Bresler) writes: > i dont have a www server benchmark numbers available, > but i do have results for an excellent cpu/cache/memory > benchmark called "Hint". > > quick dirty answer: > Integer: the intel boxes kill the snot out of *all* suns > Float: the ultras outperform intel boxes. > > long answer: > the sparc architecture is limited in its ability to perform > integer operations. my suspicion is that the memory bandwidth > is not up to the task. (surely, its not the cpu itself, but > rather feeding data and instructions to the cpu that is the > limiting factor.) > > some performance ratio: (re 586-90) > > integer: > cpu data set size > 10kB - 1MB > > ross 125: 65% - 90% of a 586-90 (yes less) > ultra 167: 55% - 80% > sparc 20: 40% - 60% > > ppro 200: 350% - 400% > ppro 150: 250% - 300% > > get the Hint benchmark and hammer some systems. > read the paper to appreciate the work that these guys > have done for everyone. > > http://www.scl.ameslab.gov/scl/HINT/HINT.html > > note: the interactive graphing tool uses floating point > data, not integer. (these guys are doing finite > element analysis and the like.) so the number that > you see will be different (as i said above) Well, we're doing image processing over here. Our application runs on quite a lot of platforms, so we implemented a little benchmakr ourselves. It does mainly a 'laws' filter on different image types. The 'laws' filter is a linear texture filter. Simply put, a matrix is pushed over the image, that does some mathematical stuff to the image. :-) It is a _typical_ image processing application. We didn't cover image loading and display! I just recompiled the benchmark. Here are the results: 133 Mhz Pentium, Asus Board [~/source/makedir]: uname -a FreeBSD pccog4.forwiss.tu-muenchen.de 2.2-BETA_A FreeBSD 2.2-BETA_A #0: Fri Jan 3 16:42:03 MET 1997 hafner@pccog4.forwiss.tu-muenchen.de:/usr/src/sys/compile/PCCOG4 i386 [~/source/makedir]: /usr/proj/horus/freebsd/bin/hbench HORUS Benchmark byte time = 2.44 val = 1.06 int4 time = 1.11 val = 0.935 float time = 7.12 val = 1.57 region time = 2.45 val = 2.09 sum time = 13.1 val = 1.47 [~/source/makedir]: /usr/proj/horus/freebsd/bin/hbench -ref HORUS/C V4.10 Byte Int4 Float Region | Sum ======================================================== HP 712/60: 1.0 1.0 1.0 1.0 | 1.0 HP 715/50: 0.7 0.7 0.8 0.9 | 0.8 HP 720: 0.7 0.6 0.8 1.0 | 0.8 HP 735/125: 1.6 1.3 1.5 2.3 | 1.8 HP K260/180: 3.7 4.0 3.0 6.7 | 4.7 UltraSPARC/143: 2.4 3.0 2.4 3.7 | 2.9 UltraSPARC/167: 2.9 3.5 2.9 4.7 | 3.6 SPARC 10/40: 0.9 0.8 0.9 1.4 | 1.0 SPARC 20/60: 1.3 0.8 1.0 2.2 | 1.5 Indigo2 R4400/250: 2.4 1.9 2.7 3.8 | 2.9 Indi R5000/180: 2.3 1.3 1.9 3.5 | 2.5 DEC Alpha/100: 1.0 0.9 1.0 1.1 | 1.0 DEC Alpha/275: 3.2 2.4 2.8 5.1 | 3.7 DEC Alpha/400: 4.9 4.9 5.2 8.0 | 6.0 As you can see from the reference: - Overall performance is about the same as a Sparc 20 - float is actually much faster on the Pentium as on the Sparc 20. Considering the price, the Pentium is of course the best you can get - at least for image processing! (PC's have faster and better graphic boards too, compared to typical workstations!) BTW: A P-Pro 200 has an overall benchmark of 3.0 ... faster than a Ultra 143 or Indigo 2! I can't give you exact results since our P-Pro is currently in San Jose (SPIE conference exhibit). -Walter -- Walter Hafner_____________________________ hafner@forwiss.tu-muenchen.de *CLICK* Wenn das so weiter geht, gibt es hier in DE bald mehr Internetprovider als Kunden :) Martin Imlau in 283.6941T990T1916@berlin.snafu.de