benchmarkingblog

Benchmarking and Systems Performance

T4 MIA

with 2 comments

I was recently looking to do a comparison of a couple of systems and figured the easiest data to look at would be SPEC CPU data. I mean it’s almost a given that even though these integer and floating point benchmarks can be, should I say, pretty boring (yawn), at least you’ll find what you are looking for because everyone publishes pretty much everything on these.

To my surprise I realized that I did not see the one system I really needed to look at – a SPARC T4 result. And then I started to laugh to myself (it was pretty late).

Of course. T4 is the one where you can find results for some really strange benchmarks but none of the regular ones where you can actually compare anything.

And then I remembered something else. I went out to my blog just to confirm it and there it was staring at me in the face. The #1 all time most popular search term that gets people to read my blog is just that — “sparc t4 benchmark.” Followed closely by “sparc t4.” “sparc t4 specint.” “sparc t4 benchmarks.” “specint t4.” “t4 specint.” And “sparc t4 performance.”

Pretty sad that for this type of basic data you can not find it on the SPEC site. You can not find it on the Gartner site. You can not even find it on the Oracle site. For this type of basic performance data you need to go to an IBM site to find out that this data just doesn’t even exist.

************************************************

SPEC, SPECint, SPECfp, SPECjbb, SPECweb, SPECjAppServer, SPECjEnterprise, SPECjvm, SPECvirt, SPECompM, SPECompL, SPECsfs, SPECpower, SPEC MPI and SPECpower_ssj are trademarks of the Standard Performance Evaluation Corporation (SPEC).

The postings on this site solely reflect the personal views of the author and do not necessarily represent the views, positions, strategies or opinions of IBM or IBM management.

technorati tags: , , ,,,,,,,,,,,

About these ads

Written by benchmarkingblog

February 13, 2013 at 9:51 am

Posted in SPARC T4, SPEC

Tagged with , ,

2 Responses

Subscribe to comments with RSS.

  1. I have the same “problem”. I have to do such technical work as capacity planning, without having the luxury of a T5 or T4 server in my hand. I rely on both specint rate benchmarks and specint benchmarks. Specint rate to see if the system can handle the overall workload, but also specint benchmarks to see if it is suitable for the pieces of software that need some speed. They represent only a small part of overall workload, but if they take 5 times longer than our requirements, I’ll have to reject the T5 or add servers from other platforms just to handle those workloads. This of course would complicate the whole software certification and release process. I am now being forced to take a worst-case approach, and divide the specint-rate results by the number of benchmark copies that was used. For the T5-8 server we’d get around 3500/896= ~4 specint-rate per thread. Maybe 4 is the correct number. Maybe it is 10? There is no way to know. If the real figure is 4, then T5 is not a candidate at all. Now, Oracle has explained on their site that the T4 and T5 cores can automatically figure out if a workload needs single-threaded speed, or multi-threaded throughput, and adapt accordingly, but I have never seen any figures published to verify this. By obfuscating their HW capabilities, and swiping the demand for benchmarks under the carpet, and instead relying on the kind of testimonial marketing that is used to sell sanitary products for ladies, they are not winning any marketing war, they are actually losing potential deals. 6 year old itanium benchmarks had specint rate per thread of around 10. Sparc T3-4 had just above 1 specint rate per thread. T4 is a no-show. Oracle’s Xeon-based X3-2 server did more than 600 specint rate with 32 threads, about 20 per thread. But it runs the wrong operating system.

    Bjorn

    April 26, 2013 at 4:17 pm

  2. Maybe they want to hide the truth. Sun systems are high priced and low performance, and have been so for at least a decade.

    Compare a M9000 8*8 core using SPECINT, and see that, performance wise, it is about the same as a 20 core intel system that costs 1/20 of the price.

    A T5220 system was worth 15000$ and, performance-wise, is equivalent to a 2000$ intel system.

    Sun latest entry-level server, the T4-1, is about 20000$, and i do not expect it to be performing better than a 3000$ intel server.

    Why should an enterprise buy a Sun system, that will cost more, perform less and won’t provide anything better than an Intel-based system ?
    Please, let Sun/Oracle hardware and OS die, and the world will be better.

    Herge

    December 10, 2013 at 5:58 pm


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: