BM Seer Unofficial thoughts from an anonymous Sun employee

UltraSPARC T2 servers better/lower-price than IBM high-end Power6

Thursday Dec 06, 2007

In the past posting there was a comment that one can't compare high-end Power6 systems with what they call low-end T2 systems. Quite simply UltraSPARC T2 systems can out perform Power6 on important datacenter workloads and much lower cost. Little need to buy from the ultra-expensive-IBM.

  • SAP-SD, clearly a high-end ERP benchmark with SAP and database. Sun T5120 Beats ultra-expensive 4-core IBM p570 4.7 GHz POWER6 by 7%. Oracle Database used on Sun's SAP-SD.
  • SPECjAppServer, another application tier and database-tier benchmark. Good for high-end servers, clearly UltraSPARC T2 is in that class and much less expensive the Power6. UltraSPARC T2 used as Oracle Database server. Sun T5120 67% faster 4-core IBM p570 4.7GHz Power6
    App Server: 3.8 better power-perf & 7x better SWaP
    Oracle Server: 3.4 better power-perf & 14x better SWaP
  • SPECjbb, another application benchmark with lots of big servers. Sun T5220 9% faster than 4-core IBM p570 4.7GHz POWER6, 2.5x better power-performance & 5x better SWaP
  • SPECint_rate2006: UltraSPARC T2 1.4GHz beats best 1-chip(2-core) IBM POWER6 4.7GHz by 29%
  • No SPECweb2005 on power6, OK web is not an application for Power6 otherwise they would have published. However IBM did publish on the IBM p550 Power5 (an embarrassingly low result). Evidently the even more expensive are reputably faster Power6 couldn't get to the level of performance.

Pricing dataquoted to a customer!
IBM p570 POWER6 4.7GHz 4-core, 32GB $252K USD
IBM p570 POWER6 4.7GHz 4-core, 64GB $313K USD ($78K/core!!!)

So I'm left with 4-core Power6 is way to expensive and doesn't have any more redundancy features that UltraSPARC T2 which outperforms it.

The only public info that lists IBM component prices that I know is at: http://www.tpc.org/results/individual_results/IBM/IBM_570_10000GB_20071015_ES.pdf

Disclosure statement

SPEC, SPECint reg tm of Standard Performance Evaluation Corporation. Results from www.spec.org as of 11/13/07. Sun Blade T6320 (UltraSPARC T2, 1 chip, 8 cores), 78.6 SPECint_rate2006. Sun SPARC Enterprise T5220 (UltraSPARC T2, 1 chip, 8 cores), 78.5 SPECint_rate2006.Sun SPARC Enterprise T5220 gccfss (UltraSPARC T2, 1 chip, 8 cores), 78.0 SPECint_rate2006. IBM p570 (POWER6, 1 chip, 2 cores), 60.9 SPECint_rate2006. Sun SPARC Enterprise T5220 (UltraSPARC T2, 1 chip, 8 cores), 62.3 SPECfp_rate2006. IBM p570 (POWER6, 1 chip, 2 cores), 58.0 SPECfp_rate2006. SPEC, SPEComp reg tm of Standard Performance Evaluation Corporation. Other results from www.spec.org as of 11/13/07. Sun SPARC Enterprise T5220 (1 chip, 8 cores, 64 threads, 1.4 GHz) 16208 SPECompM2001. Sun SPARC Enterprise T5220 (1 chip, 8 cores, 64 threads, 1.4 GHz) 16208 SPECompM2001. IBM p5 520 (1 chip, 2 cores, 4 threads, 1.9 GHz) 8174 SPECompM2001. Sun SPARC Enterprise T5220 (1 chip, 8 cores, 64 threads, 1.4 GHz) 16208 SPECompM2001. Sun SPARC Enterprise T5220 (8 cores, 1 chip) 37001 SPECweb2005. IBM p5 550 (4 cores, 2 chips) 7881 SPECweb2005. SPEC, SPECweb reg tm of Standard Performance Evaluation Corporation. Results from www.spec.org as of Nov 13, 2007. Sun SPARC Enterprise T5220 server power consumption taken from measurements made during the benchmark run. Power is average measured watts during benchmark run.IBM 550 power specifications calculated by applying 70% of the Maximum Watts published in “Facts and Features Report”, 11/14/06, posted at ftp://ftp.software.ibm.com/common/ssi/rep_sp/n/PSB01628USEN/PSB01628USEN.PDF. SPECjbb2005 Sun SPARC Enterprise T5120 (1 chip, 8 cores) 192055 SPECjbb2005 bops, 24007 SPECjbb2005 bops/JVM, Sun SPARC Enterprise T5220 (1 chip, 8 cores) 192055 SPECjbb2005 bops, 24007 SPECjbb2005 bops/JVM, Sun SPARC Enterprise T5120 (1 chip, 8 cores) 170153 SPECjbb2005 bops, 170153 SPECjbb2005 bops/JVM, Sun SPARC Enterprise T5220 (1 chip, 8 cores) 170153 SPECjbb2005 bops, 170153 SPECjbb2005 bops/JVM, Sun SPARC Enterprise T5120/T5220 results submitted to SPEC, IBM p570 (1 chip, 2 cores) 88089 SPECjbb2005 bops, 88089 SPECjbb2005 bops/JVM, IBM p570 (2 chips, 4 cores) 175474 SPECjbb2005 bops, 87737 SPECjbb2005 bops/JVM, IBM p505Q (2 chips, 4 cores) 63544 SPECjbb2005 bops, 31772 SPECjbb2005 bops/JVM, SPEC, SPECjbb reg tm of Standard Performance Evaluation Corporation. Results as of 10/08/2007 on www.spec.org. The 2-core IBM p570 POWER6 system requires 4 RU or 4 times the rack space of a Sun T5120 and consumes on average 1040 Watts of power. The 4-core IBM p570 POWER6 system requires 4 RU or twice the rack space of a Sun T5220 and consumes on average 1040 Watts of power. IBM p6 570 2-core & 4-core power specifications from 80% of maximum report power consumption published here, 06/07/07, posted here. The IBM p505Q POWER5+ system requires 1 RU of rack space and consumes on average 320 Watts of power. IBM p505 power specifications from 80% of maximum report power consumption published in ?Facts and Features Report?, 03/27/06, posted here. 1 IBM p570 (4 cores, 2 chips) and 1 IBM p550 (4 cores, 2 chips) 1197.51 SPECjAppServer2004 JOPS@Standard. 1 IBM p505Q (4 cores, 2 chips) and 1 IBM p550(4 cores, 2 chips) 618.38 SPECjAppServer2004 JOPS@Standard. SPEC, SPECjAppServer reg tm of Standard Performance Evaluation Corporation. Results from www.spec.org as of 10/10/2007. The IBM p570 POWER6 system requires 4RU or twice the rack space of a Sun T5220 and consumes on average 1040 Watts of power. The IBM p6 570 power specifications from 80% of maximum report power consumption published 06/07/07, posted here. The IBM p505Q requires 1 RU of rack space and consumes on average 320* Watts of power. The IBM p505 power specifications from 80% of maximum report power consumption published in "Facts and Features Report", 03/27/07, posted here. The IBM p550 requires 4RU of rack space and consumes on average 770* Watts of power. The IBM p5 power specifications calculated by applying 70% of the power numbers published in ?Facts and Features Report?, 3/10/06, posted < href=http://www-03.ibm.com/servers/eserver/pseries/hardware/factsfeatures.html> here. Two-tier SAP Standard Sales and Distribution (SD) standard SAP ERP 2005 application benchmark: SPARC Enterprise Model T5120 (1-way, 1 proc, 8 cores, 64 threads) 1 x 1.4 GHz UltraSPARC T2, 64GB memory, 2175 SD Benchmark users, 1.91 sec avg response time, Cert#20070xx, Oracle 10g, Solaris 10; SPARC Enterprise Model T2000 | Sun Fire T2000 (1-way, 1 proc, 8 cores, 32 threads) 1 x 1.4 GHz UltraSPARC T1, 64GB memory, 1100 SD Benchmark users, 1.91 sec avg response time, Cert#2007051, Oracle 10g, Solaris 10; IBM System p 570 (2-way, 2 processors, 4 cores, 8 threads) 2 x 4.7 GHz POWER6, 32GB memory, 2035 SD Benchmark users, 1.99s avg resp time, Cert#2007037, Oracle 10g, AIX 5L Version 5.3; SAP, R/3, mySAP reg TM of SAP AG in Germany and other countries. More info www.sap.com/solutions/benchmark.

[10] Comments
Like this post? del.icio.us | furl | slashdot | technorati | digg
Posted at 02:30PM Dec 06, 2007 in IBM Follies
Tags:
Comments:

Perhaps people are thinking that if the T2 is so cheap, "it's just obvious" that it can't be high end. How wrong those people are!

Posted by James C. McPherson on December 06, 2007 at 03:09 PM PST #

No, people see x86 line in IBM server witch 2 socket servers beat T2 and 2-4 times cheaper.
Power6 is analog SPARC64 systems in Sun's line they doesn't compete with T2 and x86 ...

Posted by iTriffids on December 07, 2007 at 12:16 AM PST #

Analog? is that because they do different workloads? Use different electrons? I think your classifications are artificial.

Actually many customers have compared Power6 to US T2 and chosen T2 for their workloads. There are many availability features in US T2 servers, so yes it is of the class of Power6. It is now looking like
the US T2 servers are a class above for availability, performance and of course at a much better $/perf.

I just saw a real workload customer testing where a US T2 server beats a 4-socket QC x86 on performance, price, and lots on $/perf. And the US T2 was a huge lead in power-performance. I'll try to get this data made public.

When I see pricing comparisons like yours I often see the comparison on small memory configs or base prices of small configs. We covered this in the past.

Posted by BM Seer on December 07, 2007 at 10:41 AM PST #

>We covered this in the past.

Yes, but it seems you doesn't remember:
http://blogs.sun.com/bmseer/entry/spec_cpu2006_ultrasparc_t2_exactly

Sun Fire X4450 (4 Quad-Core Intel Xeon X7350, 2.93 GHz, 1066 MHz FSB, 130W) 64Gb RAM less than $50K

Sun Fire T5220 (1 CPU, 64Gb RAM) $60K

Your customer must check SPECjAppServer, SAP-SD and personal ... 8 T2 cores has no chances vs 16 XEON.

Posted by iTriffids on December 08, 2007 at 09:30 AM PST #

The customer ran there real workload (for real stuff 8-cores beat
16-core Xeon). Because that didn't match your beliefs... you want to point them in other directions?

So you must believe that IBM is super expensive based and T2 can beat Power6 on the same workloads for customers. ...but you want to call them
different classes?

Posted by BM Seer on December 10, 2007 at 02:25 PM PST #

The x86 vs. US-T2 argument is not really relevant. The two servers are really complementary, and often target similar market segments.

But the difference is this. You can run the same SPARC/Solaris ISA/ABI platform on an entry, sub-$25K server up to a multi-million dollar mainframe class data center server. No other processor/OS platform offers that.

IBM tried to create an entry POWER/AIX play with PowerPC970 and failed. Then it tried again with POWER5 using its "Quad-Core Module" socket doublers, and has had limited success. POWER6 is not an entry processor, but that hasn't stopped IBM from cramming two lobotomized POWER6 processors into its obsolete blade server for yet another attempt to create an entry POWER play.

At the same time, IBM bifurcated its POWER ABI by trying to push Linux on its entry POWER platforms and positioning AIX on its high-end systems. The problem is IBM created a situation where there was absolutely no value add to use Linux on LPARs on POWER5 QCM compared to Linux on VMware on x86.

Now IBM will tell you the big advantage of Linux on POWER is you can move code from x86 to POWER with a simple recompile and run your Linux apps on POWER. But for large, enterprise apps on POWER, they push customers to AIX.

Compare that to Solaris. You can move code between SPARC and x86 with guaranteed code compatibility. From VMware and Xen on x86 to LDOMs and hardware Domains on SPARC. One code base from one to 128 threads.

This is extremely compelling to enterprises who desire to reduce the number of different OS revisions they are trying to manage.

Posted by Mark on December 10, 2007 at 07:46 PM PST #

FWIW, SPEC released their (first?) power benchmark this week, along with a slate of initial results from AMD, Fujitsu Siemens, Intel and HP.

http://www.spec.org/power_ssj2008/

Perhaps this will lessen the incentive to "guesstimate" competitive power consumption.

Posted by rick jones on December 12, 2007 at 08:11 AM PST #

>Because that didn't match your beliefs...
My, SPEC,SAP and many others ...

>T2 can beat Power6 on the same workloads for customers. ...but you want to call them different classes?

MySQL can beat oracle on the same workloads. T2 can't scale more than 1 CPU, x86 poorly scale after 4 CPU. Power6&Sparc best price/perfomance can show after 8 CPU and beaten by x86 on less than 4 CPU.

Posted by iTriffids on December 12, 2007 at 12:24 PM PST #

wow, thanks Rick for the tipoff on the SPECpower_ssj2008 announce.
Thanks, I missed it.

Memory is now a huge component of system wattage. I really hope vendors didn't do things like use tiny slow disks, turn off NICs, configure small memory with big dimms, or do BIOS changes -- anyone explore that yet?

I'm worried that the entries I saw were only 4GB, 8GB, & 16GB! BUT, those same CPUs are coupled with 64GB or 32GB in other benchmarks. Looks like some "dumpster diving" on watt-hungry memory.

Why don't HP, IBM, Dell, etc. just publish the watts they use on all
benchmarks. Why hide excessive watts on benchmark runs with 32GB or 64GB?

Sun publishes measured watts on US T2 on a wide variety of benchmarks published, others hide - that speaks VOLUMES!

SPECpower_ssj and SPEC are registered trademarks of Standard Performance Evaluation Corporation. www.spec.org

Posted by BM Seer on December 12, 2007 at 02:00 PM PST #

Since you brought-up "speaking VOLUMES" where _was_ the T2 in the initial set of SPECpower_ssj2008 announcement?-)

WRT the memory, I know that in my experience in benchmarking to date, when neither power nor price were among of the components measured, it was easier to just fill the DIMM slots and go. Particularly if the same system was going to be cycled through a number of benchmarks. There were/are enough other things to have to consider. I don't think I'd read too much into differences in memory quantities.

WRT NICs on or off, that seems like a reasonable thing to me - if the workload being measured doesn't need the second, third or fourth NIC in the system, and if disabling it somehow saves a few watts, so be it. The folks at lesswatts.org are talking about things like that.

Keep in mind that this is being positioned as a _first_ step - get a methodology in place for measuring (not guessing...) power, apply it to a first benchmark, gain experience and see about broadening it from there.

Posted by rick jones on December 12, 2007 at 06:47 PM PST #

Post a Comment:
Comments are closed for this entry.