( ESNUG 417 Item 10 ) ------------------------------------------- [09/08/03]

Subject: ( ESNUG 416 #3 ) AMD's Opteron & Intel's Influence Over Synopsys

> From cost and SPECint performance I would like to buy AMD Opteron servers.
> Even though the Itanium 2 has better SPECfp performace, I believe I would
> prefer two AMD Opteron machines to the one Intel Itanium machine I could
> buy for roughly the same amount of money. ...
>
> I will also mention that I have been able to get quotes for AMD Opteron
> machines with delivery times within the month.  For the Intel Itanium 2
> servers, I have been able to get quotes, but NO ONE has been able to give
> me a delivery schedule.  I know the Intel Itanium 2 severs exist.  I saw
> them at DAC.  I just don't know how I would get one in my hands.
>
>     - Maynard Hammond
>       Scientific-Atlanta, Inc                    Lawrenceville, GA


From: Steve DiBartolomeo <steve=user  company=artwork not bomb>

Hi, John,

I agree with Maynard.  The AMD Opteron willl deliver great bang-for-the-buck
when compared with Itanium.  The major software vendors just need a little
time.  Most of their 64-bit Linux roadmaps were put together a year ago when
Opteron was still an AMD promise.  Only in the last month or two have enough
benchmarks come out that clearly show Opteron is going to give great
performance at a reasonable cost.

While Intel can easily sow FUD prior to release, once the pudding was made
and enough people have had a taste the balance will shift.

Further it is only in the last 30 days or so that workstation motherboards
for Opteron have begun trickling out.  (Again, Intel pressure on the MOBO
manufacturers ... I understand some MOBOs mfgs are shipping Opteron MOBO's
in plain white boxes with no logo ...)

A year ago if you were putting your plans in place would you have bet with
AMD against Intel/HP/IBM?  Nope.

IBM recently announced a major supercomputer using Opteron.  My motherboard
specialist is now hoping that new Opteron MOBOs will be built with 8 memory
slots; 8x1GB is cheaper than 4x2GB if you only need 8GBs and if you need
16 GB well then 8x2GB is possible.

    - Steve DiBartolomeo
      Artwork Conversion Software

         ----    ----    ----    ----    ----    ----   ----

From: [ Mighty Mouse ]               

Hi John...

PLEASE make this anonymous!

Have worked previously at Synopsys for many years, the answer to Maynard's
question is obvious.

Why isn't Synopsys supporting Opteron?

  1) It's too much work to port to both AMD and Intel right away.  Synopsys
     wants to wait to see what the potential market is before porting to two
     platforms.

  2) Intel is a HUGE customer of Synopsys.  The part of the Synopsys sales
     force that services Intel is _very_ influential.  And at Synopsys, the
     sales force rules.  AMD is a Synopsys customer too, but not in the same
      category as Intel.

So the decision to go with Itanium is not based on engineering or cost 
analysis.  It's 100% business.  Anything you can do to make one of your 
biggest customers happy, you do.

    - [ Mighty Mouse ]

         ----    ----    ----    ----    ----    ----   ----

From: William Natter <wnatter=user  domain=nortelnetworks hot calm>

Hi, John,

Many people will have the same reaction, but I have to ask: did Maynard
require that much memory even when creating models for your sub-blocks?
He's certainly heard of Interface Logic Models (ILMs) and other sub-block
modeling techniques?

    - William Natter
      Nortel Networks                            Nepean, Ontario, Canada

         ----    ----    ----    ----    ----    ----   ----

From: Thomas Fairbairn <tomf=user  domain=pdd.3com pot palm>

John,

Maynard was concerned with using Itanium and Opteron in 64-bit mode for the
extra memory in his post.  It's worth pointing out that the Opteron solution
can be run in 32-bit mode which will run all the existing Linux apps.  Of
course, you loose the ability to use the extra memory.

    - Tom Fairbairn
      3Com Europe Ltd.                           Hemel Hempstead, UK

         ----    ----    ----    ----    ----    ----   ----

From: Hai Vo-Ba <hai=user  domain=hpeshai2.fc.hp bought tom>

Hi John,

I've done quite a bit of EDA benchmarking (mostly Synopsys and Cadence
plus internal tools) and found SPECint2000_base to be a pretty good
performance predictor.  The trick is paying attention to the OS and
compilers behind the benchmark numbers.  Here are some of the top
single-CPU SPEC numbers from spec.org as of August 4th, 2003

          SPECint2000      
  Co.      Base Peak  System
  -----    ---- ----  ---------
  HP       1322 1322  HP Server rx2600 (1.5GHz Itanium2 6M, HP-UX 11i)
  Dell     1242 1294  Precision Workstation 650 (3.06GHz Xeon w/ 1MB L3)
  Intel    1221 1261  Intel D875PBZ motherboard (3.2GHz P4)
  Dell     1204 1234  PowerEdge 6650 (2.8GHz Xeon MP)
  AMD      1095 1170  A4800 (1.8GHz Opteron)
  IBM      1077 1113  IBM eServer pSeries 690 Turbo (1.7GHz Power4+)
  AMD      1044 1080  ASUS A7N8X Motherbrd rev. 2.0, 2.2GHz Athlon XP 3200+
  AMD       960   --  A4800 (1.8GHz Opteron w/ gcc33 -m32, Suse LES8)
  Fujitsu   776  905  PRIMEPOWER650 (1350MHz SPARC64 V)
  Sun       642  722  Sun Blade 2000 (1.2GHz UltraSPARC III Cu)

           SPECfp2000      
  Co.      Base Peak  System
  -----    ---- ----  ---------
  HP       2119 2119  HP Server rx2600 (1.5GHz Itanium2 6M, RedHat AS v2.1)
  IBM      1598 1699  IBM eServer pSeries 690 Turbo (1.7GHz Power4+)
  Dell     1267 1285  Precision Workstation 360 (3.2GHz P4, DDR400)
  IBM      1202 1266  IBM eServer pSeries 690 Turbo (1.3GHz Power4)
  Dell     1173 1186  Precision Workstation 650 (3.06 GHz Xeon)
  HP       1124 1482  AlphaServer GS1280 7/1150
  AMD      1122 1219  A4800 (1.8GHz Opteron)
  Dell     1103 1120  PowerEdge 6650 (2.8 GHz Xeon MP)
  Fujitsu  1096 1340  PRIMEPOWER650 (1350MHz SPARC64 V)
  HP       1019 1365  HP AlphaServer ES45 68/1250
  Sun       953 1118  Sun Blade 2000 (1.2GHz UltraSPARC IIIi)
  AMD       873  982  ASUS A7N8X Motherbrd rev. 2.0, 2.2GHz Athlon XP 3200+


For instance, DC-2002.05-SP1 on 1.5GHz Itanium2 6M (HP-UX) runs about 1.3x
faster than a 1.6GHz Opteron (Suse).  Back in May that was the fastest
Opteron I could find.

Also, given the cost of EDA tools, I believe system performance is more
important than price/performance.  The cost of one rx2600 with 24GB of RAM
is still only a fraction of that of the 2 PhysOpt licenses that run on
that box.

     - Hai Vo-Ba
       Hewlett Packard                           Ft. Collins, CO

         ----    ----    ----    ----    ----    ----   ----

From: [ The Cat In The Hat ]
To: Maynard Hammond <maynard.hammond=user  domain=sciatl spot calm>

Hi, Maynard, 

John, please post me anonymously. 

We been benchmarking Xeons, P4s, Opterons and Itaniums for a while.  Our
general conclusions: 

  - While SpecInt/FP give first order insight, real design benchmarks 
    yield totally different results. 
  - In the 32 bit single processor space, using EDA tools with large 
    runtime image sizes, P4 3 Ghz 800 Mhz FSB with DDR400 memory performs
    30-50% faster than their 2.8 Ghz Xeon 533 Mhz FSB counterparts.
    (Workstation vs rack mount server).  The Opteron 1.8 Ghz is par in
    performance with P4 800 Mhz FSB in smaller image size simulations,
    but out performs it in larger memory size images.  Similarly in SystemC
    simulations, at a 30 Meg runtime image size, the three machines are
    10-15% of each other, but with a 300 Meg runtime image, P4  800 Mhz
    FSB was 60% faster than Xeon 533 FSB, and Opteron was 3X faster than
    Xeon 533 FSB. 
  - In general, Opterons scale better with runtime memory size, due to
    better main memory architecture. 
  - In general, two jobs on dual processor take 5-10% performance hit on 
    Opteron,  30-40% on Xeon.
  - Running 32 bit binaries under 64 bit linux for Opteron allows for 4G 
    process size vs traditional 3G. 
  - In summary for 32 bit space problems, based on current HW offerings,
    Opterons should be used, followed by P4 3.2 Ghz 800 Mhz FSB. (DDR400). 
  - In regard to Itanium, early results are mixed in 4G+ simulations.  It 
    had 3X+ performance relative to Sun 750 Mhz.  However, on sub 1G
    simulations, we saw par performance with the Sun, and it was beat by
    32 bit Xeon/P4/Opteron.  I would consider our experiments void at this
    time.  We need to understand the behavior of the EDA tool to comment
    more.  In defense of Itanium, folks are quoting 1.5 Ghz parts which are
    very expensive.  Secondly a traditional box for physical design has
    16G+ memory which is the main cost of the hardware.  Finally HW is
    normally 10-15% of overall simulation cost, so I would not loose sleep
    over the cost of HW in the EDA space. 

In general everyone on ESNUG should contact their EDA vendors for AMD-64
ports of relevant apps, as they will have found the Opteron to be a very
compelling 32 bit EDA simulation platform.

    - [ The Cat In The Hat ]

         ----    ----    ----    ----    ----    ----   ----

From: Maynard Hammond <maynard.hammond=user  domain=sciatl spot calm>
To: [ The Cat In The Hat ]

[ The Cat In The Hat ], 

Thanks for your help.  I hope everyone takes your advice and requests an AMD
64-bit port of their tools.  My company is very conservative.  We just can't
buy all the licenses needed.  Machines are easier to get up to a point.
After they get to a certain cost it becomes hard to get them also.
Therefore, I am always looking at what I can get cheaply but runs fast.  I
want to be able to compile as much as I can in a given time period.

BTW, have you found anyone other than HP selling the Itanium 2 systems? 

    - Maynard Hammond
      Scientific-Atlanta, Inc                    Lawrenceville, GA

         ----    ----    ----    ----    ----    ----   ----

From: [ The Cat In The Hat ]
To: Maynard Hammond <maynard.hammond=user  domain=sciatl spot calm>

Maynard,

Dell and IBM both have Itaniums, also Tier 2 folks like Racksaver.  The
former 2 don't seem to have any discount relative to HP.  Maybe you have
relationship and can work the price.  In fact for PrimeTime, we found 3-6X
performance over Sun on Itanium.  Everyone here loves the Opterons, but
given we don't have any promises from Vendors, we will go with Itaniums
for the Physical Design space for now and Opterons for 32 bit.  If apps
show up, we will add memory upgrades.  Also it will take a few more months
just for the all the Itanium ports to be available.  Synopsys is ahead
but others are behind.

Lets see how it goes.

    - [ The Cat In The Hat ]

         ----    ----    ----    ----    ----    ----   ----

From: Mark Wroblewski <mw1=user  domain=agere plot rom>

Hi John,

Saw Maynard's article on DeepChip.  Sorry no data, only an observation:

Possibly the most likely reason that SNPS and the other CAD vendors haven't
worked very hard on Opteron ports is that until recently there have been no
top-tier computer makers with Opteron rigs.  That has changed in as much as
IBM has announced plans for an eServer line based on dual-processor Opteron
pizza boxes.  Hopefully others like Dell will follow suit.  Bwuhahaha!

I can't speak for all of Agere, but I can say there is interest in this
issue here at my site based on a conversation about your article yesterday.
We think AMD did it right, will garner a sizeable amount of support because
of its backward-compatibility with 32-bit applications, and the price point
just got even lower with their announcement of some new chips in the family
and lower prices on the chips they introduced about a month ago.

    - Mark Wroblewski
      Agere Systems                              Longmont, CO

         ----    ----    ----    ----    ----    ----   ----

From: Michael Neilly <mneilly=user  domain=yahoo lot bomb>

John,

Maynard Hammond asks why Synopsys won't support 64-bit linux on Opteron.
I do not know Synopsys' specific reason but I can tell you for certain
that the same problem exists for other EDA tools.

We use Fintronic's Finsim on 64 bit linux but it requires a node locked
license because Macrovision does not support Opteron and doesn't seem very
motivated to do so.  The same problem exists for Novas' tools.

Until FlexLM can be run on Opteron you're not likely to see support of
that platform.  Why someone at AMD hasn't pounded down their door I
don't know...

    - Michael Neilly

         ----    ----    ----    ----    ----    ----   ----

From: Mike Stabenfeldt <stabie=user  company=stabie-soft hot mom>

John,

Not all EDA companies are jumping on Intel first.  I announced my Opteron
port this week.  I find the price/performance of the AMD Opteron a little
better and it was easy to actually buy a box.

    - Mike Stabenfeldt
      Stabie-Soft

         ----    ----    ----    ----    ----    ----   ----

From: James Guzzo <james.t.guzzo=user  company=intel taught tom>

Hi John,

Here is some of the data that Maynard requested.

> Are there other vendors that offer Itanium 2 solutions other than HP?

Yes, in fact there are 17 different vendors offering Itanium 2 based systems
today.  For a complete list: http://www.intel.com/buy/wtb/wtb1015.htm


> Does someone have better analysis that would make me want to buy and
> Intel Itanium 2 solution?

Yes.

a. Maynard's analysis uses 1.4 GHz/4 MB Itanium 2 system pricing and 1 GHz
   Itanium 2 performance numbers.  Spec numbers for 1.4 GHZ/4 MB Itanium 2
   configurations have not yet been officially published by the system
   vendors, so I can't give you those numbers, but they should be
   substantially better than the 1 GHZ/3 MB system config numbers that he
   used in his analysis.  As a reference point, the HP ZX6000 1.5 GHz/6 MB
   Itanium 2 based system posts Specint = 1315 & SpecFP = 2106 scores.  You
   can do your own scaling.  See http://www.spec.org for details.

b. Maynard's Itanium 2 system pricing is overstated by several thousand
   dollars depending on which Itanium2 system supplier he chose.  General
   quotes start from $12K for a dual proccessor 1.4 GHz/ 4 MB Itanium 2
   system with 8 GB of memory, and $22K for a 16 GB system.


> I will also mention that I have been able to get quotes for AMD Opteron
> machines with delivery times within the month.  For the Intel Itanium 2
> servers, I have been able to get quotes, but NO ONE has been able to give
> me a delivery schedule.  I know the Intel Itanium 2 severs exist.  I saw
> them at DAC.  I just don't know how I would get one in my hand.

Called a few of the vendors to get an idea of delivery schedules for
Itanium 2 systems.  Rackable Systems, CA Digital, and Microway are all
quoting scheduled delivery dates anywhere between 3-5 weeks.

    - James Guzzo
      Intel Corporation

         ----    ----    ----    ----    ----    ----   ----

From: Maynard Hammond <maynard.hammond=user  domain=sciatl spot calm>

Hi, John,

I have been hearing (phone calls) from several Synopsys people.  (5 now).
They have all been supportive.  I haven't had a negative response yet!

I've also heard (grapevine, rumor mill, probably should be taken with a
grain of salt) that some groups in Synopsys had actually started porting
to the Opteron, but that the word came from Aart himself to stop it.

    - Maynard Hammond
      Scientific-Atlanta, Inc                    Lawrenceville, GA

         ----    ----    ----    ----    ----    ----   ----

From: John Ferguson <john_ferguson=user  company=mentor lot palm>

Hi John,

Just wanted to send you a quick follow-up.  The 2003.4 release of Calibre is
now available.  With this version, we now officially support the Itanium2
processors for running 64 bit Linux with Red Hat Advanced Server 2.1.  We
now also officially support the Opteron processors for 32 bit Linux runs.
64 bit Linux support on Opteron is still scheduled for later this year.

    - John Ferguson
      Mentor Graphics                            Wilsonville, OR


 Sign up for the DeepChip newsletter.
Email
 Read what EDA tool users really think.


Feedback About Wiretaps ESNUGs SIGN UP! Downloads Trip Reports Advertise

"Relax. This is a discussion. Anything said here is just one engineer's opinion. Email in your dissenting letter and it'll be published, too."
This Web Site Is Modified Every 2-3 Days
Copyright 1991-2024 John Cooley.  All Rights Reserved.
| Contact John Cooley | Webmaster | Legal | Feedback Form |

   !!!     "It's not a BUG,
  /o o\  /  it's a FEATURE!"
 (  >  )
  \ - / 
  _] [_     (jcooley 1991)