BM Seer Unofficial thoughts from an anonymous Sun employee

IBM POWER6 4.7GHz: following the rules?

Tuesday Aug 28, 2007

Benchmarks have rules, if you announce systems you must have them available in different time periods. I'm worried that since I haven't seen any IBM 4.7GHz POWER6 systems at customers that they may not be available yet. Anyone know?

Benchmark Availability Windows & IBM POWER6 May 22 Launch benchmarks

    R6iNotes - 2 months, No Power6 launch benchmark

    SPECweb2005 - 90 days, No Power6 launch benchmark
    SPECmpi2007 - 90 days, No Power6 launch benchmark

    SPEC CPU2006 - 3 months, eight IBM Power6 Launch Benchmarks
    SPEC OMP2001 - 3 months, one Power6 Launch Benchmark
    SPECjAppServer2004 - 3 months, No Power6 launch benchmark
    SPECjbb2005 - 3 months, four IBM Power6 Launch Benchmarks
    SPECmail2001 - 3 months, No Power6 launch benchmark
    SPEC SFS - 3 months, No Power6 launch benchmark

Now it is over 3 months we need to make sure the three spec benchmarks have all of the POWER6 servers shipping in those configurations with the AIX and all other software they listed publicly available. Actually they said the hardware was avail in June. No matter the 3-month rule is the critical one. How can we check on this? Anyone want to try to order one to see what IBM says? or ???

Disclosure statement

SPEC, SPECint, SPECfp, SPECjbb, SPECweb, SPECjAppServer, SPECmail, SPECsfs, SPECmpi, and SPEComp registered trademarks of Standard Performance Evaluation Corporation.

NotesBench R6iNotes More info: www.notesbench.org

from IBM press release:
IBM System p570 1-core (4.7 GHz, 1 chip, 2 cores/chip,1 thread/core) SPECint2006 result of 21.6; IBM System p570 1-core (4.7 GHz, 1 chip, 2 cores/chip,1 thread/core) SPECfp2006 result of 22.3; IBM SPECjbb2005 result of 691,975 bops (86,497 bops/JVM) on a 16-core (8 chips, 32 threads) 4.7 GHz POWER6 IBM System p570 running AIX 5L V5.3. A 16-core IBM System p 570 (4.7 GHz) (8 chip, 32 threads) result (86,624 SPECompMpeak2001). IBM System p 570 results as of May 21, 2007. (IBM results to be submitted by 5/21/07, configuration available 8/23/07)

[8] Comments
Like this post? del.icio.us | furl | slashdot | technorati | digg
Comments:

I'm sure they shipped one to somebody. I'm sure they will sell you one with no discount, and direct demand towards a slower bin.

It's all irrelevant. AMD Barcelona, Intel Tigerton, UltraSPARC T2, these are the processors that matter now. These are processors which will power "Redshift" computing.

Who cares about POWER6? Sure its fast, but so was Alpha. Sure it's great for grand challenge HPC, but so is a Cray vector processor. Sure its a highly available, highly virtualizable enteprise processor but so is a mainframe. Basically POWER6 is chasing the Blueshift market. But then again, IBM is Big Blue, isn't it?

Posted by Mark on August 28, 2007 at 10:01 PM PDT #

Yes, I'll grant you looking at processors it is really becoming pretty irrelevant, but rumors are big blue really rushed out power6 to try to have something to compete. ...and rules are rules. Several years ago, at Sun, I know a bunch of people that really went to extremes to get some SW releases done in time to match the benchmark rules. Anyone having problems should do the same.

Posted by BM Seer on August 29, 2007 at 08:53 AM PDT #

Hm ... and who need all this benchmarks ? CIO want to see SAP, OEBS, TPC, and other REAL app benchmarks. All this benchmarks is available from IBM on Power6, but there are no useful benchmarks from SUN ...

Posted by Triffids on August 30, 2007 at 01:29 AM PDT #

Perhaps someone in Sun has ordered a Power6-based p570 for competitive evaluation by now. Those folks would know if it arrived or not.

Don't forget that it isn't just the hardware which must be generally available.

Posted by rick jones on August 30, 2007 at 08:28 AM PDT #

1600 published results on SPEC CPU, hundreds on SPEC jbb... yes they
are important (may not important if you don't like how the systems you like compare :) ).

Well informed, CIOs and datacenter managers know that you have to look at
a variety of commercial benchmarks: SPEC CPU, SPEC jbb, SPECjAppServer, TPC-H, SAP-SD, SPECweb. ...and benchmark their own applications, it is here where Sun wins lots of business as we typically do better on actual workloads than we do on benchmarks. But Sun has published SEVENTY benchmarks in the past 12 months. IBM has avoided many benchmarks on IBM POWER6 systems.

Well-informed CIOs know that TPC-C has been over-optimized is 15 years old and useless to gauge performance. They also know the new Oracle apps benchmark is more of a sizing guide as systems can run it very differently.

I don't know of anyone that has a 4.7GHz system in the field or in Sun, that is why I asked someone to post their system in comments so we could check.

Posted by BM Seer on August 30, 2007 at 09:49 AM PDT #

I know if a large Sun customer that just ordered four 4.7 Ghz p570's after they ran a head to head benchmark against SPARC IV+'s with 2x as many SPARC cores for in a standard two-tier JAVA/DBMS (oltp and batch). Not only was the p570 environment many times faster then the SPARC IV+, it was 2/3 the price. Do your homework before you post...

Posted by Derek on August 30, 2007 at 12:46 PM PDT #

Oh, I see head to head comparison customer benchmarks against POWER6,
Sun wins most of these US-IV+ 1.95GHz vs p570 4.7Ghz shootouts (on perf and cost). Homework done! it is not about core count but system cost. Of course there will always be some that go the other way.

But I don't know any customer that actually has a 4.7GHz in their hands yet.

Posted by BM Seer on August 30, 2007 at 01:58 PM PDT #

Bullshit, SAP-SD:
(*) SPARC Enterprise Server Model M8000, 16 processors / 32 cores / 64 threads, SPARC64 VI 2.4 GHz, 256 KB L1 cache per core, 6 MB L2 cache per processor
36570 SAPS

IBM System p 570, 8 processors / 16 cores / 32 threads, POWER6 4.7 GHz, 128 KB L1 cache and 4 MB L2 cache per core, 32 MB L3 cache per processor
40070 SAPS

Only oracle will charge additional $480K (16 cores * 0.75 * $40K).

Posted by Triffids on August 31, 2007 at 01:10 AM PDT #

Post a Comment:
Comments are closed for this entry.