<html><div style='background-color:#ffffff'><DIV>
<P>The only thing you're not looking at is the new AMD "Hammer", which here's a brief list of telltale features...</P>
<P>[copied from link below]</P>
<P><EM>...the architecture's pervasive use of the HyperTransport interconnect, a scalable system bus with support for single-processor and multiprocessor configurations, and an integrated DDR memory controller. </EM><BR></P>
<P><A href="http://www.eet.com/story/OEG20011016S0092">http://www.eet.com/story/OEG20011016S0092</A><BR></P>
<P>The interesting part being the fact that the DDR controller is embedded into the CPU. This means that with a duel or quad CPU set, you <FONT color=#ff0000>DOUBLE</FONT> and <FONT color=#ff0033>QUADRUPLE</FONT> your RAM bandwidth. Tie that to Hypertransport interconnects and you get a very new and large RAM bus subsystem to expect very different results...follow that with the fact that AMD hinting of the 2ghz range, well...read for yourself...</P>
<P>Here's another article on the "Hammer" : <A href="http://arstechnica.com/wankerdesk/01q4/hammer/hammer.html">http://arstechnica.com/wankerdesk/01q4/hammer/hammer.html</A></P>
<P>Sincerely,</P>
<P>Chad Owens</P></DIV>
<DIV></DIV>
<DIV></DIV>>From: Eugene Leitl <EUGENE.LEITL@LRZ.UNI-MUENCHEN.DE>
<DIV></DIV>>To: <BEOWULF@BEOWULF.ORG>
<DIV></DIV>>Subject: CCL:RIMM vs SDRAM (fwd)
<DIV></DIV>>Date: Tue, 23 Oct 2001 19:10:57 +0200 (MET DST)
<DIV></DIV>>
<DIV></DIV>>---------- Forwarded message ----------
<DIV></DIV>>Date: Tue, 23 Oct 2001 08:34:14 -0600
<DIV></DIV>>From: Scott Anderson <ANDERSON@CHEM.UTAH.EDU>
<DIV></DIV>>To: chemistry@ccl.net
<DIV></DIV>>Subject: CCL:RIMM vs SDRAM
<DIV></DIV>>
<DIV></DIV>>Dear Tom:
<DIV></DIV>>
<DIV></DIV>>Regarding your question about RIMM (RDRAM) v.s. SDRAM (or DDR SDRAM) for
<DIV></DIV>>Gaussian:
<DIV></DIV>>
<DIV></DIV>>I have done a fair bit of comparing machines for Gaussian, as I am planning
<DIV></DIV>>to buy a bunch in the near future. I have been using an MP2 Opt Freq job
<DIV></DIV>>that takes a few hours as my benchmark. I have also run much longer Direct
<DIV></DIV>>Dynamics trajectory jobs on several different configurations. The MP2 job
<DIV></DIV>>was deliberately chose to avoid a lot of disk i/o because I really wanted to
<DIV></DIV>>focus on the computational speed. I have compared 1 and 1.4 GHz Athlons,
<DIV></DIV>>PIIIs, and P4s. The conclusion is that for Gaussian, the P4 is 20-30%
<DIV></DIV>>faster than the Athlons for the same clock speed, and since you can get P4s
<DIV></DIV>>with faster clocks, the P4 ends up being substantially faster. This speed
<DIV></DIV>>difference is quite surprising because the P4 floating point unit is
<DIV></DIV>>substantially inferior to that of the Athlon, and as far as I have been able
<DIV></DIV>>to determine, the rate deterimining step for these fast chips is memory
<DIV></DIV>>speed. Here the RDRAM really shines. I have also compared Athlons with
<DIV></DIV>>PC133 SDRAM and PC2100 DDR ram, which is theoretically substantially faster.
<DIV></DIV>>For Gaussian jobs, I see no difference.
<DIV></DIV>>
<DIV></DIV>>One interesting thing to look at in this regard is the review in the Tom's
<DIV></DIV>>Hardware review of the new Athlon XP.
<DIV></DIV>>http://www6.tomshardware.com/cpu/01q4/011009/index.html They compare
<DIV></DIV>>several different Athlons with different memory configurations, and also
<DIV></DIV>>2GHz P4s with both RDRAM and DDR memory. The Athlon XP looks great on most
<DIV></DIV>>of the benchmarks -- equalling or exceeding the P4, even though the P4 clock
<DIV></DIV>>speed is much higher. The one area where the XP fails, however, is memory
<DIV></DIV>>speed. Note that the P4 with RDRAM does 50% better than the P4 with DDR,
<DIV></DIV>>and nearly a factor of two better than the Athlons with either PC133 or DDR
<DIV></DIV>>memory.
<DIV></DIV>>
<DIV></DIV>>My conclusions:
<DIV></DIV>>1. Gaussian is probably memory-bandwidth limited, and you should spend the
<DIV></DIV>>extra money RDRAM.
<DIV></DIV>>2. DDR memory doesn't really make a huge difference -- certainly nothing
<DIV></DIV>>like the factor of two you might expect.
<DIV></DIV>>3. The ideal configuration is very application dependent. For Gaussian, I
<DIV></DIV>>would go with the P4. If you are running a mix of programs, you really have
<DIV></DIV>>a tough decision.
<DIV></DIV>>
<DIV></DIV>>Have fun
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>>Prof. Scott L. Anderson
<DIV></DIV>>Department of Chemistry
<DIV></DIV>>University of Utah
<DIV></DIV>>315 S. 1400 E. Rm 1216
<DIV></DIV>>Salt Lake City, UT 84112
<DIV></DIV>>(801)585-7289
<DIV></DIV>>FAX(801)581-8433
<DIV></DIV>>www.chem.utah.edu/chemistry/faculty/anderson/anderson.html
<DIV></DIV>>----- Original Message -----
<DIV></DIV>>From: "Tom Kuppens" <TOM@HARTREE4.RUG.AC.BE>
<DIV></DIV>>To: <CHEMISTRY@CCL.NET>
<DIV></DIV>>Sent: Tuesday, October 23, 2001 5:07 AM
<DIV></DIV>>Subject: CCL:RIMM vs SDRAM
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>> > Hi there,
<DIV></DIV>> >
<DIV></DIV>> > We have several linux machines that run Gaussian. Now we want to add some
<DIV></DIV>>new
<DIV></DIV>> > power to our park i.e. a PIV - 1700MhZ with 512M RAM (SDRAM or RIMM).
<DIV></DIV>> >
<DIV></DIV>> > Is there someone who can provide me with some info on the Gaussian and/or
<DIV></DIV>> > overall performance with RIMM versus SDRAM modules.
<DIV></DIV>> >
<DIV></DIV>> > Thanks in advance,
<DIV></DIV>> >
<DIV></DIV>> > Tom Kuppens
<DIV></DIV>> > Ghent University
<DIV></DIV>> >
<DIV></DIV>> > --
<DIV></DIV>> > tom@hartree4.ac.be
<DIV></DIV>> > http://studwww.rug.ac.be/~tkuppens/quantum/
<DIV></DIV>> > ---
<DIV></DIV>> > It seems to me, Golan, that the advance of civilization is nothing
<DIV></DIV>> > but an exercise in the limiting of privacy.
<DIV></DIV>> > -- Janov Pelorat in Asimov's Foundation's Edge
<DIV></DIV>> >
<DIV></DIV>> >
<DIV></DIV>> >
<DIV></DIV>> > CHEMISTRY@ccl.net -- To Everybody | CHEMISTRY-REQUEST@ccl.net -- To
<DIV></DIV>>Admins
<DIV></DIV>> > CHEMISTRY-SEARCH@ccl.net -- archive search | Gopher: gopher.ccl.net
<DIV></DIV>>70
<DIV></DIV>> > Ftp: ftp.ccl.net | WWW: http://www.ccl.net/chemistry/ | Jan:
<DIV></DIV>>jkl@osc.edu
<DIV></DIV>> >
<DIV></DIV>> >
<DIV></DIV>> >
<DIV></DIV>> >
<DIV></DIV>> >
<DIV></DIV>> >
<DIV></DIV>> >
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>>-= This is automatically added to each message by mailing script =-
<DIV></DIV>>CHEMISTRY@ccl.net -- To Everybody | CHEMISTRY-REQUEST@ccl.net -- To Admins
<DIV></DIV>>MAILSERV@ccl.net -- HELP CHEMISTRY or HELP SEARCH
<DIV></DIV>>CHEMISTRY-SEARCH@ccl.net -- archive search | Gopher: gopher.ccl.net 70
<DIV></DIV>>Ftp: ftp.ccl.net | WWW: http://www.ccl.net/chemistry/ | Jan: jkl@osc.edu
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>>
<DIV></DIV>>_______________________________________________
<DIV></DIV>>Beowulf mailing list, Beowulf@beowulf.org
<DIV></DIV>>To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
<DIV></DIV></div><br clear=all><hr>Get your FREE download of MSN Explorer at <a href='http://go.msn.com/bql/hmtag_itl_EN.asp'>http://explorer.msn.com</a><br></html>