SPARC T5-8 Server

You can have Larry’s slower machine but you have to be willing to beg Oracle for more discount on the software to make up for the added cost.

1920409Some simple back-of-the-napkin math speaks for itself, unless I am missing something. The new T5-8 machine achieved a peak SPECint rate of 29.3 per core across 128 total cores according to Oracle’s benchmark page. According to SPEC.org, Dell’s PowerEdge M820 (available last August) achieved 38.4 per core among 32 cores.

While I realize that dividing total throughput by number of cores is not entirely fair to either server, the calculations are justified by how Oracle licenses its software, i.e., by core factor. In this case, both T5 and Xeon chips are the same at .5 cores per Processor.

The damning conclusion here is that the same number of cores in a Dell/VMware solution offers 30% more throughput at half the server cost. I configured both machines via each supplier’s online store. Roughly speaking, T5-8s cost $250k and four Dell M820s total $140k.

Larry joked during the announcement that you can have the faster machine but you have to be willing to pay less for it. A more accurate statement is: You can have Larry’s slower machine but you have to be willing to beg Oracle for more discount on the software to make up for the added cost.

My assessment is that, besides having its head buried in the sand (or worse), Oracle is counting on existing customers that remain committed to SPARC Solaris to buy the new line of servers. Plenty of those IT shops exist, but not enough to sustain a profitable hardware business, and certainly not as profitably as software.

P.S. Oracle continues its egregious marketing claims against IBM despite IBM’s attempts to hold Oracle accountable for false advertising,  Case in point, a Power 750 Express achieved a SPECint rate of 54.4 per core across 32 cores back in October. While that isn’t double the T5 (and it sort of needs to be since the core factor is double at 1.0), PowerVM is more effective than Solaris Containers at cramming workload through at a higher utilization rate. Still, I chose Dell for this post because it costs so much less than both.


  1. Phil

    First, you are comparing systems on a benchmark *THAT DOESN’T RUN ORACLE SW*, so how could you claim per core performance comparisons for Oracle SW licensing when the benchmark is a RAW CPU only benchmark? Why not use one of the other 17 benchmarks that Oracle published on SPARC T5 to compare to the Dell server? https://blogs.oracle.com/BestPerf/

    But why are you comparing a SPARC T5-8 Enterprise class server to a 4-socket DELL Blade entry level server?
    Clearly VERY different class of systems so the comparison is irrelevant.

    The Closest Dell system would be either the PowerEdge R820 or PowerEdge R910 which are much more scaleable systems (and much more expensive!).

    If you look at benchmarks that run Oracle SW like SAP SD-2-Tier (For DB) or SPECjEnterprise2010, you’ll see that SPARC T5-8 for example on SPECjEnterprise2010, outperforms the R910 by almost 5x running Oracle DB and Weblogic! and on per core performance, SPARC T5-8 does 449 EJOPs/core vs 299 EJOPs/core on R910. Wow-that’s a 50% savings right there on licensing costs!

  2. Eric Guyer

    Thanks for your comment.

    Don’t get me wrong, I think Oracle’s TPC benchmark results are phenomenal.

    But to answer your questions: I prefer SPEC because it levels the playing field. Other benchmarks like TPC allow for expensive exceptions such as $1M in Flash Accelerator F40. The IT shops I work with have to live within financial constraints that are 10% the cost of Oracle’s TPC achievement. Said differently, 55 cents per tpmC is crazy expensive when the total cost is $4.7M and you don’t need 8.55M transactions per minute.

    As for the Dell, I’m comparing T5-8 to M820s because most IT shops do the same. The contest is between proprietary UNIX and commodity x86 hardware, not between Solaris, AIX and HP-UX. Those days are over and its odd that Oracle is taking the fight there.

    Of greater importance, however, is that your argument over license savings is a cannibalization of Oracle’s core business model, which is perpetual software support. Larry should be careful considering that faster servers require less software. If Oracle wants to compete in hardware (and I don’t think it should), at least do the smart thing and give the hardware away while discounting the software less. Remember, perpetual means forever, whereas hardware has to be resold every three to five years.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>