UltraSPARC T1 Screams Running Java

Sun has announced the new Sun Fire T1000 and T2000 servers today along with SPECjbb2005 benchmark results on these systems. What makes these results so special? They run the UltraSPARC T1 processor with 8 cores and 32 threads on a single chip. The performance of the UltraSPARC T1 systems easily surpasses performance on all other 1U, 2U, or 4U Systems. These results also leverage the high performance features in the newly released J2SE 5.0._06. Take a look at the the chart below. The Sun T2000 surpasses all other competition in the 2U and 4U space. The 1U Sun Fire T1000 leads the 1U results. How are these results comparable? Its simple, compare the raw throughput SPECjbb2005 bops score. One may ask: "How can you compare a 8 core / 32 thread box to a 4 core / 8 thread Power 5+?". Its easy. Chip and core counts are steadily becoming irrelavent. What really matters is how much work (throughput) a system can achieve, how much is that system going to cost to run, and how much lab space, power, and cooling will this system require. Looking at the above results with this in mind clearly shows why Sun UltraSPARC T1 systems are separate from the pack. Sun Fire UltraSPARC T1 much, much less expensive to run than is competitors. How about those Cool Threads! Here's the details on the configurations compared above: How about scalability? Here's a good example of how the Sun Fire T2000 and the UltraSPARC T1 processor scales from 1 to 32 threads. Each SPECjbb2005 is a new thread. Throughput steadily increases as new threads are added, peaking at 32. Fine print SPEC disclosure: SPECjbb2005 Sun Fire T1000 (1 chip, 8 core, 32 threads) 51,540 bops, 12,885 bops/JVM submitted for review, Sun Fire T2000 (1 chip, 8 core, 32 threads) 63,378 bops, 15,845 bops/JVM submitted for review, IBM eServer p5 520 (2 chips, 2 cores, 4 thread) 32,820 bops, 32,820 bops/JVM, AMD Tyan white box (2 chips, 4 cores, 4 thread) 44,574 bops, 44,574 bops/JVM, IBM eServer p5 550 (4 chips, 4 cores, 4 thread) 61,789 bops, 61,789 bops/JVM . SPEC™ and the benchmark name SPECjbb2005™ are trademarks of the Standard Performance Evaluation Corporation. Competitive benchmark results stated above reflect results published on www.spec.org as of November 30, 2005. For the latest SPECjbb2005 benchmark results, visit http://www.spec.org/osg/jbb2005.

[ T: http://technorati.com/tag/NiagaraCMT ]


The Opteron 280 is labeled 2 core in the bar chart, but doesn't mention that there are two of those chips in the system.

Posted by Tom Hawtin on December 06, 2005 at 07:23 AM EST #

Thanks for catching the error, I'll update the chart.

Posted by dagastine on December 06, 2005 at 12:13 PM EST #

hi there,

it will be nice if someone can post a comparative approx. prices for the above dell/ibm/sun boxes. That will let the customers decide how well it will work for them.

Thank you,


Posted by anjan bacchu on December 06, 2005 at 01:52 PM EST #

I second the request for the price/jbbmark info. In particular, I'd be interested in a comparsision with the Sun X4200 with the same disks and memory size. The T2000 looks like a truly impressive engineering achievment, but at the announced prices, it isn't clear that it represents a better value than the X series for most customers. Quite likely the T2000 is a better deal over the long haul due to reduced power consumption, but that it is a hard sell. It certainly didn't work out well for Transmeta. It would be really unfortunate if the Niagara systems failed simply because they weren't priced properly

Posted by Robert Weiler on December 09, 2005 at 06:35 AM EST #

Trackback: Oracle ist auf 8-Core CPUs billiger als auf 4-Core CPUs (german)

Posted by Bernd Eckenfels on December 09, 2005 at 01:09 PM EST #

The fineprint should contain the RAM sizes for the various systems. I also think we need to see single-instance JVM numbers or have the competitors also distribute Heaps on 4 address spaces. Which of the VMs run 32 or 64 bit?

Posted by Bernd Eckenfels on December 09, 2005 at 01:14 PM EST #

To followup on myself: the T2000 was running with T1 1.2GHz Solaris 10 64bit 32GB RAM, 4x32bit Server VM (256mb pages, 2506mb heap the SC1425 was running with Xeon DP 3.6GHz single core Windows 2003 32bit 8GB RAM 1x32bit JRockit VM (1800mb heap) Unfortunatelly this memory differences (1.8GB vs. 12GB) are totally uncompareable... BTW: I can't find any published T1000 results? Gruss Bernd

Posted by Bernd Eckenfels on December 09, 2005 at 01:53 PM EST #

Its all about throughput Bernd. 4 instances was chosen for the T2000 to highlight multi-threaded and mult-process scalability. The Dell SC1425 and Jrockit could reach the throughput of the T2000, nor could it come close to supporting the number of concurrent threads. Single JVM results running on Sun Fire T2000 and T1000 will be coming shortly. Current results are ~5% behind the multi-JVM result. Feel free to do the math yourself. Yes the memory configurations are much different. But so is the supported thread count. As more threads are run in the SPECjbb2000 benchmark, more memory is required. The SC1425 would through a OOM error with a 1800mb heap when running 64 concurrent threads with SPECjbb2005. Of course these results are comparable, BEA has stated that competitive benchmarks are all about putting your best number out. The reason BEA chose 1800mb is because the platform they target makes its very difficult to perform well with a larger heap. The Sun Fire T1000 results are still in review at SPEC. You'll see them on http://www.spec.org in 2 weeks.

Posted by dagastine on December 09, 2005 at 02:45 PM EST #

Sorry about the typo, I meant SPECjbb2005 in all cases in my last post.

Posted by dagastine on December 09, 2005 at 02:47 PM EST #

Dave, Couple of questions for you... 1. Did you do any work on 6-cores? If so, would you share the results? 2. Will you be revisiting this work for Niagara 2?

Posted by Steve Feldman on February 17, 2007 at 07:57 AM EST #

Post a Comment:
  • HTML Syntax: NOT allowed



« April 2014