Pentium Pro vs. Pentium II

From: "Peter R. Schreiner" <prs $#at#$ organik.uni-erlangen.de>
Organization: Institut fuer Organische Chemie, Universitaet Goettingen
Subject: Pentium Pro vs. Pentium II
Date: Tue, 23 Sep 1997 15:17:49 +0200 (MET)
 Dear CCLers,
 The question
 "Does anyone have experience with numerical applications
 (Ab initio codes etc.) on Pentium Pro vs. Pentium II under Linux systems?"
 triggered quite a number of useful responses. It looks like
 PCs are finally replacing average type workstations. It is still
 not clear whether or not the Pentium II series really
 is better that the Pentium Pros - at least they are different
 in price.
 Here are the responses:
 You can find some timings concerning PC GAMESS benchmarks on Pentium Pro and
 Pentium II systems under Windows NT on the URLs:
 http://classic.chem.msu.su/gran/gamess/index.html
 http://classic.chem.msu.su/gran/gamess/performance.html
 May be these can somewhat help to estimate performance under Linux.
 Regards,
     Alex. A. Granovsky
     Moscow State University
 -----------------------------------------------------------
 gran $#at#$ classic.chem.msu.su
 %%%%%%%%%
 I am really interested in any informations you will receive.
 My personal experience, up to now, is only with the Pro 256kb cache.
 I used it to run DFT programs, and the performance, with big systems
 (30 atoms or more) are excellent. It is 85% of an IBM rs6000 Mod. 3CT
 with 1MB of L2 cache memory. And it costs roughly 20%. Remarkable!
 As for the PII, I have no experience. Anyhow, I "think" that the Pro
 even if the clock is slower, should perform better. The L2 cache in
 the Pro is fully integrated on the processor and thus run at the
 same speed of the processor itself.  On the PII, instead, the
 the L2 memory runs at 1/2 of the speed of the processor.
 Finally, the Pro is somewhat cheaper than the PII 266 MHz.
 Hope it helps,
 Best regards
 Luigi
 Luigi Cavallo
 Dept. of Chemistry
 Univ. of Naples, ITALY
 Email cavallo $#at#$ chemna.dichi.unina.it
 %%%%%%%%%%
 While I don't have any timings or benchmarks for Pentium II systems,
 I suggest you have a look at http://www.specbench.org/ for SPECint95 and
 SPECfp95 results.  From what I've seen the pentium II is much better in
 floating point and better in integer than the pentium pro.
 Also direct your web browser at
 http://www.intel.com/procs/perf/PentiumII/spec95int.htm
 and
 http://www.intel.com/procs/perf/PentiumII/cpumark32.htm
 and
 http://www.intel.com/procs/perf/PentiumII/nortonsi32.htm
 To the bottom of this email I've attached the output of cpu benchmarks I ran on
 a dual PPRO 200 (the fact that it is dual means nothing; the SMP code for Linux
 is bad right now and the benchmarks are not multithreading).
 good luck,
 Rich
 BBBBBB   YYY   Y  TTTTTTT  EEEEEEE
 BBB   B  YYY   Y    TTT    EEE
 BBB   B  YYY   Y    TTT    EEE
 BBBBBB    YYY Y     TTT    EEEEEEE
 BBB   B    YYY      TTT    EEE
 BBB   B    YYY      TTT    EEE
 BBBBBB     YYY      TTT    EEEEEEE
 BYTEmark (tm) Native Mode Benchmark ver. 2 (10/95)
 NUMERIC SORT    :  Iterations/sec.:         70.24  Index:   1.80
 STRING SORT     :  Iterations/sec.:          4.35  Index:   1.94
 BITFIELD        :  Iterations/sec.:   16976692.52  Index:   2.91
 FP EMULATION    :  Iterations/sec.:          3.90  Index:   1.87
 FOURIER         :  Iterations/sec.:       1393.26  Index:   1.58
 ASSIGNMENT      :  Iterations/sec.:          0.71  Index:   2.71
 IDEA            :  Iterations/sec.:        138.29  Index:   2.12
 HUFFMAN         :  Iterations/sec.:         63.19  Index:   1.75
 NEURAL NET      :  Iterations/sec.:          0.71  Index:   1.13
 LU DECOMPOSITION:  Iterations/sec.:         52.81  Index:   2.74
  ...done...
 ===========OVERALL=============
 INTEGER INDEX:           2.119
 FLOATING-POINT INDEX:    1.700
  (90 MHz Dell Pentium = 1.000)
 ===============================
 %%%%%%%%%%%%%%
   I don't have any real quantitative answers for you, but thought I might
 pass along my experience with running Gaussian 94 under Linux on a single
 Pentium Pro 200MHz with 64MB RAM and a fast/wide SCSI hard drive.  The
 couple of runs that I did to compare the speed with an Indigo2 (single
 R8000 75MHz processor) seemed to give a factor of about 3-4 in cpu time.
 That is, the Pentium Pro 200 was 3-4 times slower than the SGI.  Considering
 the relative costs this is pretty good in my opinion.  I compiled Gaussian94
 using f2c/gcc.  I sometimes wonder if Gaussian would run faster if compiled
 with a good commercial fortran compiler (e.g. Absoft).  But even the
 educational price ($524.25 US) is too pricey for me right now.
   My guess is that you will not see much of an improvement with the Pentium II
 for these types of floating point intensive calculations.  I believe that
 most of the advantage of the Pentium II comes from MMX capabilities which
 you will find useless with Linux.  At least this is what I recall reading
 in Linux Journal or Unix Review.  Floating point is probably only marginally
 faster than a Pentium Pro.
       Hope this helps - John Bushnell      bushnell $#at#$ gaucho.ucsb.edu
 %%%%%%%%%%%%%%
 We are implementing a system called LoBoS, short for Lots of Boxes on
 Shelves, which consists of 64 dual Pentium Pro machines.  We organize them
 to do supercomputing with molecular dynamics calculations.  We tested the
 Pentium Pro and the Pentium II prior to selecting the Pentium Pro.
 1. The Pentium Pro gives equivalent performance.  Our benchmark which
    consists of extensive floating point (CHARMM) took 2 hours on a 200
    MHz Pentium Pro.  On a 266 MHz Pentium II the time was 2 hours times
    (200/266).
 This is not surprising since a PentiumII is a Pentium Pro core with the
 MMX instructions added on.  MMX, which is entirely integer operations,
 doesn't do anything for floating point operations.  It might do something
 for pointer references and integer math but compilers do not currently use
 MMX for these operations.
 The PentiumII has some new features.  One which improves performance is
 better IO to the system bus.  However, this is essentially offset by the
 fact that the L2 cache runs at half the CPU speed.  Essentially we saw a
 wash between these two design changes.
 Buy the Pentium Pro !  It gives you MUCH more for the money.
 Eric Billings
 LoBoS Project
 NIH
 %%%%%%%%%%%%
 I tested calculations using Gaussian 94, Revision D.4
 1. #N b3p86/gen 5d freq : ClO2 calculation using 79 basis
   Pentium Pro 200MHz : 256 mins
   Pentium II  233MHz : 192 mins
   CrayT3E (DEC Alpha EV5.6 450MHz, 900MFLOPS, not parallelized) : 39 mins
 2. #N b3p86/gen 5d opt=z-matrix : Cl2O calculation using 131 basis
   Pentium     133MHz : 336 mins
   Pentium Pro 200MHz : 149 mins
   Pentium II  233MHz : 130 mins
   CrayT3E (DEC Alpha EV5.6 450MHz, 900MFLOPS, not parallelized) : 56 mins
 -----------------------------------------------------------------------
 Son, Sang-kil                   zannavi $#at#$ orbit.kaist.ac.kr     \\____//
                                                                (. .) |
 Quantum Chemistry Laboratory                                   | " | |
 Dept. of Chemistry, KAIST, South Korea                       CC|___/CCC
                         http://cmsrisc.kaist.ac.kr/qclab/        U
 -----------------------------------------------------------------------
  ...Peter
 --
     ////
 ___|--00___________________________________________________________________
    C   ^     Dr. Peter R. Schreiner
     \ ~/     Institut fuer Organische Chemie
     <><>     Georg-August Universitaet Goettingen
              Tammannstr. 2                    Phone: +49-(0)551-393287
              D-37077 Goettingen, Germany      FAX:   +49-(0)551-399475
 	     http://www.gwdg.de/~pschrei
 ___________________________________________________________________________