BM Seer Unofficial thoughts from an anonymous Sun employee

Current SPECjbb2005 Sun Fire V890 2.1GHz

Wednesday Jun 20, 2007

The Sun Fire V890 with 8x 2.1 GHz UltraSPARC IV+ obtained a result of 244846 SPECjbb2005 bops, 30606 SPECjbb2005/JVM on the server-side Java benchmark.

The Sun Fire V890 is 40% faster than the expensive 4-core IBM p570 (4.7 GHz power6). I'll leave it to the reader to check the system prices to see the "per core performance" comparisons don't work at best, and are in fact are disingenuous because they make you compare systems of very different costs to you. You'll spend more for IBM.

At the high-end the IBM p570 16-core ($Megabucks) is only 2.8 times faster than the Sun Fire V890 2.1GHz (16-core). Again check the prices of the servers to really understand what you are getting.

Postscript:IBM will drone on every time about performance/core or trying to equate systems on a per-core basis. The reality is IBM's cores cost so much more than anyone elses cores. Get very suspicious any time you see IBM saying things like "8-core IBM system vs. 8-core Sun system" or "perf/core IBM wins". Price out one of these comparison and you WILL BE TRULY AMAZED at the smoke & mirrors they are using. Also note how many times they will mention performance on 16-core and they do a price comparison on 8-core with lots of memory on the Sun system and a sparse config on the IBM system.

If you want to see a chip to chip comparison see BM Seer's posting of UltraSPARC T1 result. Now that box rocks.

SPECjbb2005 Performance Chart (ordered by performance, bops : SPECjbb2005 Business Operations per Second (bigger is better)

System Date Processors Performance
(Chips, Cores, Threads) GHz/ Type SPECjbb2005
bops 		JVMs SPECjbb2005
bops/JVM
IBM p570 power6 6/07 (8, 16, 32) 4.7 power6 691,975 8 86,497
Sun Fire V890 6/07 (8, 16, 16) 2.1 US-IV+ 244,846 8 30,606
IBM p570 power6 6/07 (2, 4, 16) 4.7 power6 175,474 2 87,737
Sun Fire V890 10/05 (8, 16, 16) 1.5 US-IV+ 117,986 4 29,497

Benchmark Description

SPECjbb2005 (Java Business Benchmark) measures the performance of a Java implemented application tier (server-side Java). The benchmark is based on the order processing in a wholesale supplier application. The performance of the user tier and the database tier are not measured in this test. The metrics given are number of SPECjbb2005 bops (Business Operations per Second) and SPECjbb2005 bops/JVM (bops per JVM instance).

Disclosure Statement:

SPECjbb2005 Sun Fire V890 (8 chip, 16 cores 16 threads) 244846 SPECjbb2005 bops, 30606 SPECjbb2005 bops/JVM, Sun Fire V890 (8 chip, 16 cores, 16 threads) 117986 SPECjbb2005 bops, 29497 SPECjbb2005 bops/JVM, IBM System p 570 (4.7 GHz) running AIX 5L V5.3 175,474 SPECjbb2005 bops, 87,737 SPECjbb2005 bops/JVM, 2 chips, 4 cores, 8 threads, IBM System p 570 (4.7 GHz) running AIX 5L V5.3, 691,975 SPECjbb2005 bops, 86,497 SPECjbb2005 bops/JVM, 8 chips, 16 cores, 32 threads, SPEC, SPECjbb reg tm of Standard Performance Evaluation Corporation. Results as of 06/20/2007 on www.spec.org.

Results Summary
Results
Sun Fire V890: 244846 SPECjbb2005 bops
30606 SPECjbb2005 bops/JVM
Reference Date: June 20, 2007
Systems: Sun Fire V890, 64 GB
Total Number Processors: 8
Processor/GHz of Server: US-IV+ 2.1 GHz
Operating System: Solaris 10 6/06
JVM: Java HotSpot(TM) 32-Bit Server, Version 1.6.0_02

See Also

SPECjbb2005 Benchmark Reports

IBM Consolidation Press Release

[9] Comments
Like this post? del.icio.us | furl | slashdot | technorati | digg
Posted at 07:00AM Jun 20, 2007 in IBM Follies
Tags:
Comments:

One chip, eight cores on the v890? Overly speedy cut and paste with T1 results perhaps?

Posted by rick jones on June 20, 2007 at 08:52 AM PDT #

Whoops, fixed the typo.

Posted by BM Seer on June 20, 2007 at 09:37 AM PDT #

and 8-way Xeon is 28% faster
Prices:
8-way Opteron SF X4600 - $43K
8-way SPARC SF V890 - more than $300K
8-way Power6 p570 - $2,1M (2x faster, price from tpc-c report)
16-way SPRAC E6900 - more than $1M

Posted by Triffids on June 20, 2007 at 09:56 AM PDT #

If you post prices, you must also post exact Memory Configurations&Mhz, processor GHz, #chips/#cores, and where you got the pricing. and way=what in your terms?

Posted by BM Seer on June 20, 2007 at 10:33 AM PDT #

Opteron SF X4600 - $43K, 16 cores, 32Gb (www.sun.com)
SPARC SF V890 - $279K 16 cores 1.8 GHz UltraSPARC IV+, 128Gb (www.sun.com)
Power6 p570 - $2,1M, 16 cores, 768Gb (2x faster, price from tpc-c report)
SPARC E6900 - more than $1M, 32 cores, 128Gb
Let's calculate systems price with 128Gb
Opteron: + $31K (+96Gb ) total cost: $74K (8 GB (2 x 4 GB DIMMs) DDR2-667 Registered ECC Memory [Add $ 2,595.00 each])
SF V890: 8x2.4Ghz cores will cost more than $300K
IBM power total cost $2.1M, Activation of 256 GB DDR2 POWER6 Memory cost $1,163,520 256GB Memory (8x32GB) DDR2 POWER6 memory cost $291,378
so total memory cost = $1,455M Conclusions:
1.16-core Power6 system with 128Gb is 2x faster and 2x costly than equal v890 and will cost $887K.
2. 16-core Power6 is cheaper than 32-core SUN's E6900 and provide more power :) 3. Sparc&Power6 can't beat 16-core Xeon prices

Posted by 217.198.232.133 on June 20, 2007 at 11:21 AM PDT #

Memory size is a HUGE factor in system cost. The biggest.

Posted by BM Seer on June 20, 2007 at 12:11 PM PDT #

Note: The systems pricing above are list prices and have different discount structures. Usually AMD/Intel systems have razor thin margins so discounting is minimal but on the high end Sun systems, discounting is much much higher, and probably higher than the new Power6 system which is trying to compete against AMD/Intel on price and Sun high end on performance and RAS. Whats funny though is trying to compare a p570 against a E6900! Why not include an IBM Mainframe to the comparison? Thats like trying to compare prices and performance of a Race car vs a Tank. They might have similar prices and different performance (depends on the surface) but clearly not designed for the same task.

Posted by Phil on June 21, 2007 at 01:49 AM PDT #

>Whats funny though is trying to compare a p570 against a E6900!

BM Seer has no problems in such compare :)
http://blogs.sun.com/bmseer/tags/e6900

P.S. sun.com is down ?

Posted by Triffids on June 21, 2007 at 02:21 AM PDT #

I wonder why would anyone want to buy those expensive boxen to run Java? Why not just fill the space with lots of Clovertowns/Opterons plus some content switches (load balancers) if you run huge website? If you buy such a big iron like V890 and way up to E25K, you probably don't run PostgreSQL or MySQL on it, do you? Then don't forget to factor in some Oracle license cost. Then you could see that per-core performance makes a huge difference.

Posted by Mike on June 23, 2007 at 04:15 AM PDT #

Post a Comment:
Comments are closed for this entry.