By now, most of you will have seen the announcement of the T5 and M5 systems. I don't intend to repeat any of this, but I would like to share a few early thoughts. Keep in mind, those thoughts are mine alone, not Oracle's.
It was rather obvious during the Launch Event that we will enjoy the competition with IBM even more than before. I will not join the battle of words here, but leave you with a very nice summary (of the first few skirmishes) found on Forbes. It is worth 2 minutes of reading - I find it very interesting how IBM seems to be loosing interest in performance...
Since much of the attention we are getting is based on performance claims, I thought it would be nice to have a short and clearly arranged overview of the more commonly used benchmark results that were posted. I will not compare the results to any other systems here, but leave this as a very entertaining exercise to you ;-)
- TPC-C - a new world record for non-clustered systems
Official result listing at tpc.org
Oracle's BestPerf benchmark summary
It's always been a disk benchmark, really. But removing the IO bottleneck makes it show CPU power...
It would be interesting to see if IBM will publish any non-clustered results to counter this... - SPECjEnterprise2010 - a new world record
Official result listing at spec.org
Oracle's BestPerf benchmark summary
T4 was already faster than Power7, and now T5 improved this by 43% with the same number of cores... - SPECjbb2013 - a new world record
Official results submitted to spec.org, you will find them there soon.
Oracle's BestPerf benchmark summary
No IBM submission yet, but T5 does quite well against x86 CPUs too... - SAP SD 2-Tier - a new world record for 8 socket systems with T5-8, and a new overall world record with the M5-32
Official result listing at sap.com
Oracle's BestPerf benchmark summary
You need to be careful with SAP SD benchmarks, as the results are only comparable within the same benchmark version. - SPECint_rate2006 - new world records for systems with one and eight sockets
Official results submitted to spec.org, you will find them there soon.
Oracle's BestPerf benchmark summary
IBM's always been asking for SPECint. It's not really a full server benchmark with IO and all. But there you go...
There are more performance publications, especially on the BestPerf blog. Some of these are interesting because they compare T5 to x86 CPUs, something I recommend doing if you don't shy away from reconsidering your view of the world from time to time. But the ones I listed here are more likely to be accepted as "independent" benchmarks than some others. Now, we all know that benchmarking is a leap-frogging game, I wonder who will jump next? (We've leap-frogged our own systems a couple times, too...) And to finish this entry off, I'd like to remind you that performance is only one part of the equation. What usually matters just as much, if not more, is price performance. In the benchmarking game, we can usually only compare list prices - have a go at that! To quote Larry here: “You can go faster, but only if you are willing to pay 80% less than what IBM charges.”
Competition is an interesting thing, don't you think?