Second is BlueGene/L in DOE/NNSA/LLNL, USA. Third is Intrepid BlueGene/P solution from DOE/ANL. Fourth is Ranger SunBlade x6420 at TACC. Fifth is Cray Jaguar XT4 QuadCore at DOE/ORNL. Remarkably, the systems at no. 1, 3, 4 and 5 are new. Seventh is Encanto SGI Altix ICE 8200 at New Mexico Computing Applications Center.

9 and 10 are French systems: Blue Gene/P at IDRIS and SGI Altix ICE 8200 EX at Total Exploration Production, the first commercial system in the top 10.

For the first time the TOP500 is tracking power consumption which is new, according to Dr. Strohmaier. In terms op power consumption the three first systems achieve the following values:

• 1. = 2.35
• 2. = 2.33
• 3. = 1.26

Changes in the TOP500 are not restricted to the top 10. Quad-core is new in the list. More than 50 percent are using quad-core processors. We also face a high replacement rate. As Dr. Strohmaier put it: "Quad-core means the acceleration of Moore's Law". Harpertown E54xx Xeon Intel is the dominant chip. Xeon 53xx Clovertown and Xeon 51xx Woodcrest are doing fine too.

Eleven years ago the teraflop barrier was broken. In 2008 the petaflop barrier has been smashed. Can we expect a breakdown of the exaflop barrier by 2019 and of the zettaflop barrier by 2030?

As for the performance rate by vendors the following list can be retrieved:

• IBM: 48%
• HP: 22%
• Cray: 7%
• SGI: 6%
• Dell: 6%
• Others: 8%

In terms of systems per country the results look like this:

• USA: 259
• UK: 53
• Germany: 44
• France: 33
• Japan: 22
• China: 13
• Sweden: 9
• Russia: 9

Bell's Law states that important classes of computer architectures come in cycles of about 10 years. It takes about a decade for each phase: early research, early adoption and maturation, and prime usage. The Vector/SIMD systems were popular around 1993 to 1995. Then Custom Scalar systems led the dance till 2003.

The following hype was the Commodity Cluster between 2004-2008. Right now, embedded/accelerated systems are coming up.

Now let's have a look at these much talked about power consumption data. Dr. Strohmaier explained that they have been planning this for years. The TOP500 results are independent from the Green500 but the TOP500 team is trying to learn from the Green500 mistakes. The method is to collect power consumption for Linpack as workload by including all essential parts of a system (processor, memory, etc.), exluding features related to machine room (most disk, UPS etc.) and taking into account the full system or a part of it large enough to include all shared components (fans, power supplies, ...). Then comes the real difficult part: analyse this data carefully!

In order to rank the objects by "size" one needs extensive properties, including weight or volume and Rmax. A larger system should have a larger Rmax. The ration of 2 extensive properties is an intensive one: weight/volume = density. Performance/power consumption = Power_efficiency. Dr. Strohmaier emphasized that one cannot 'rank' objects with densities by size: density does not tell anything about the size of an object. A piece of lead is not heavier or larger than one piece of wood, so to speak.

Therefore, one should not produce a list based on independent systems based on density. Otherwise, it will always sort smaller systems before larger ones. There are absolute power levels for power consumption. Power efficiency is related to processors.

If we look at power efficiencies of systems with different processors, we can observe that 8+1 cores are most power efficient, followed by QC embedded - quad core processors. The QC embedded and DC embedded processors are inserted into the IBM BlueGene types. The most power efficient systems consist of quad core processors. Dual core is less efficient.

Blade systems are more power efficient than BlueGene systems. The IBM blade system uses low power L5420.

The SGI Altix ICE 8200 has an efficient Linpack and the HP cluster platform 3000 BL 2x220 is an efficient double density blade.

All in all there are three different approaches to measure power consumption but you can combine them, concluded Dr. Strohmaier.

More information is available at the website www.top500.org