If there's one thing you can tell by looking at the newest list of the most powerful (read fastest) supercomputers in the world, it's that some real change is taking place in the HPC field. Five of the top 10 systems are new to the list, and over 200 smaller systems from last November's rankings have been eclipsed by faster ones.

As expected, IBM kept its lead position with the BlueGene/L System, a joint development of IBM and DOE's National Nuclear Security Administration (NNSA). The system is installed at DOE's Lawrence Livermore National Laboratory in Livermore, CA. This same system also nabbed the top spot last November, but has since doubled in size and reached a new record Linpack benchmark performance of 136.8 TFlop/s.

IBM's leadership further increased due to six of its systems now ranking in the top 10. Five of these systems use the Blue Gene design. The new No. two listed system is an IBM Blue Gene system with the same architecture but smaller in size than the No. one BlueGene/L at LLNL. It was recently installed at IBM's Thomas J.Watson Research Center in Yorktown, N.Y. and reached 91.2 TFlop/s.

“The latest list, particularly if you look at the top 10, clearly illustrates the dynamic nature of supercomputing today. In just one year, we have seen a dramatic turnover from a ranking topped by the Earth Simulator followed by a number of clusters and two prototypes of IBM's Blue Gene,” said Erich Strohmaier, one of the founding editors of the list and a computer scientist at Lawrence Berkeley National Laboratory. “Today, we see that Blue Gene has gained the first two positions and occupies five of the top 10 slots.”

Some analysts agree that IBM has solidified its lead as a clear HPC powerhouse.

“IBM's general strategy of delivering solutions based on a wide variety of in- house and OEM technologies also succeeds at the top end of the IT food chain,” said Charles King, principal analyst at Pund-IT Research. “While a number of competitors argue vociferously against the wisdom of the company's efforts, IBM's latest Top500 success offers something for them to chew on along with their humble pie,” he continued.

SGI's Columbia system at the NASA Ames Research Center in Mountain View, CA dropped one spot to No. three. Columbia clocked in at 51.87 TFlop/s. The NEC-built Earth Simulator, which has a Linpack benchmark performance of 35.86 TFlop/s and had held the No. one position for five consecutive TOP500 lists before being replaced by BlueGene/L last November, is now shown as No. four.

Intel also had a good showing on the list; 333 of the world's top 500 systems on the list have Intel processors inside. Five years ago only four systems on the list were Intel-based.

According to King, HP had the weakest showing since before its acquisition of Compaq, though after IBM, HP sells the bulk of systems at all performance levels of the TOP500. The company's major management changes and replacement of its RISC platforms may have contributed to its poor showing. However, customers are slowly adopting HP's Itanium-based products, making the company a potential force in the future.

Analysts are optimistic about the trends that seem to be appearing with the 25th list.

“From the standpoint of IT evolution, this newest list demonstrates impressive gains over the past half year, with entry-level systems now offering nearly 1/3 better performance than those just six months ago,” said Charles King. “In addition, faster systems replaced half of the top ten and over 40 percent of last November's list. This breathtaking performance boost offers a good deal to consider at the stratospheric limits of computing, but it also underlines another fact: the evolution of supercomputing as an increasingly pervasive entity across a broadening range of commercial applications,” he stated.

“This new list also demonstrates that clustered systems are continuing to replace traditional monolithic supercomputing systems. As this trend accelerates, and every indication suggests it will, the influence of supercomputing will expand across an ever widening range of commercial and consumer markets.”

Linux analyst Stacey Quandt, Quandt Analytics, seemed to agree. “The overall advancements in semiconductor and interconnect technology in the supercomputer realm will in time reap benefits for customers running 'real-world' applications.

The semi-annual TOP500 list of supercomputers is the work of Hans Meuer of the University of Mannheim, Erich Strohmaier and Horst Simon of the U.S. Department of Energy's National Energy Research Scientific Computing Center, and Jack Dongarra of the University of Tennessee.

Despite the bragging rights it provides, the Linpack benchmark and TOP500 list are generally thought to be insufficient indicators of actual application performance.

“The TOP500 is a pretty narrow perspective,” said Richard Partridge, VP of Enterprise Servers, Ideas International. “We've been arguing about Linpack for decades now. Clearly Linpack is a theoretical peak kind of metric. But it is easy to measure and has been measured for decades. Therefore, it has the great advantage of continuity — being able to bridge between the champions of old and today's high-end systems,” he said.

Partridge went on to say that the HPC community will need to look at more than just list placement. He suggested that the HPC Challenge might be a more comprehensive procedure to judge a system's true performance.

Though Linpack may not be the best way to judge a supercomputer's true capability, it is easy to conduct and comprehendible to more people in the field, rather than just experts.

“I don't think there's anybody who champions Linpack as being superb, but I don't think that anybody's come up with a simple, one numerical kind of number that can replace Linpack,” said Partridge. “I'm not sure that I would worry about who is number three and number 13 and number 23, as much as…what kind of trends do you see?”

The semi-annual list, regardless of all its possible shortcomings, cannot be ignored. Here are some other trends that stand out from the rankings.

• As predicted several years ago by the research team behind the TOP500 listing, only systems exceeding the 1 TFlop/s mark on the Linpack were qualified to enter the list this time. The system in No. 500 spot reached 1.166 TFlop/s.

• The entry level for the TOP10 exceeds 15 TFlop/s and the entry point for the top 100 moved from 2.026 TFlop/s to 3.412 TFlop/s.

• Entry level for the TOP500 is now 1.166 TFlop/s, compared to 850.6 GFlop/s six months ago. The last system on the list would have been listed at position 299 in the last TOP500 just six months ago. This exemplifies the continuous rapid turnover of the TOP500.

• The last system (#500) in June 2005 has about the same compute power as ALL 500 systems combined, when the list was first created 13 years ago in June1993.

• Total combined performance of all 500 systems on the list is now 1.69 PFlop/s (“petaflops” or thousand “teraflops”), compared to 1.127 PFlop/s six months ago.

Other trends of interest:

• A total of 333 systems are now using Intel processors. Six months ago there were 320 Intel-based systems on the list and one year ago only 287. The second most-commonly used processors are the IBM Power processors (77 systems), ahead of Hewlett-Packard's PA Risc processors (36) and AMD processors (25).

• There are 304 systems now labeled as clusters, making this the most common architecture in the TOP500.

• At present, IBM and Hewlett-Packard sell the bulk of systems at all performance levels of the TOP500. IBM remains the clear leader in the TOP500 list with 51.8 percent of systems and 57.9 percent of installed performance. HP is second with 26.2 percent of systems and 13.3 percent of performance and SGI is third with 5 percent of systems and 7.45 percent of performance. No other manufacturer is able to capture more than 5% in any category.

• The U.S is clearly the leading consumer of HPC systems with 294 of the 500 systems installed there (up from 267 six months ago). A new geographical trend, which started a few years ago, now emerges more clearly. The number of systems in Asian countries other than Japan is rising quite steadily. In this latest list, Japan is listed with 23 systems and all other Asian countries combined have an additional 58 systems. However Europe is still ahead of Asia, with 114 systems installed.

• China is home to 19 of the systems in Asia — up from 17 systems six months ago.

• In Europe, Germany claimed the No. 1 spot from UK again, with 40 systems compared to 32. Six months ago, UK was in the lead with 42 compared to Germany's 35 systems.