This time last year, there were only three Intel-based machines in the Top 500 listing, and one of them was the ASCI Red parallel supercomputer that dominated the ranking for so long a few years ago.

Supers based on the 64-bit Itanium processor from Intel are also – finally – making a good showing on the Top 500 list, with 18 Itanium supers in the list, up from one six months ago. Machines based on Intel processors comprised 119 of the 500 machines listed.

Intel is, as you might imagine, ecstatic, and thinks that the shift in the rankings among the Top 500 indicate a major shift in HPC computing away from parallel RISC architectures and vector supercomputers and towards clusters of Intel-based machines, particularly those running Linux.

This is something that many RISC/Unix and vector super suppliers will argue with, contending that what HPC customers in the broader market buy is different from what the biggest organizations do. But Rick Herrmann, manager of Intel’s HPC segment, says the situation is otherwise.

I think that what is happening at the top of the HPC pyramid is very representative of what is going on throughout the HPC market, he said, adding that there is a general trend toward the lower-cost computing environments in HPC, just as there is in every facet of the computing business.

Hewlett Packard Co and IBM Corp, the two dominant vendors in terms of sales in the HPC space, were unusually quiet about the Top 500 listing this time around, but neither organization is a slouch in the HPC space and they are by and large the two vendors who will benefit from the shift to Lintel iron for HPC.

SGI Inc, with its Altix line of Itanium-based 64-way shared memory supers looks like it has a winner, too, in terms of architecture and performance; now all it has to do is take off in the marketplace. Six of the Altix machine made their debut on this installment of the list.

HP had 159 systems in the Top 500 ranking with an aggregate of 90.2 teraflops of number-crunching power, while IBM had 158 machines with 130.9 teraflops of power. NEC Corp, which still holds the top spot with its Earth Simulator massively parallel vector supercomputer, has 43.9 teraflops in the Top 500 base in 14 machines.

The Earth Simulator super, located at an eponymous supercomputing center in Yokohama, Japan, comprises 35.9 teraflops of that 43.9 teraflops. Earth Simulator is still the top machine in the list, a position it took when the machine was announced this time last year. It will hold that ranking until IBM, HP, and Cray Inc roll out larger installations in the coming years.

The number two machine in the ranking is HP’s ASCI Q AlphaServer cluster at the Los Alamos National Laboratory in the US; this cluster is currently rated at 13.9 teraflops on the Linpack benchmark that is used to rank the machines.

The number three machine, which knocks out some other ASCI boxes built by the US government, is a Lintel cluster built by Linux Networx using Quadrics system interconnect. This Linux cluster is based on 2,304 Pentium 4 Xeon DP processors that have an aggregate Linpack rating of 7.6 teraflops. This Lintel cluster is located at Lawrence Livermore National Laboratory in the US as well.

Fujitsu Siemens is in the top ten of the Top 500 list with a cluster of its PrimePower 2500 Solaris-compatible machines, which is located at the National Aerospace Laboratory of Japan in Tokyo. That machine has 2,304 processors of the Sparc64-GP chips made by Fujitsu, which run at 1.3GHz and is rated at 5.4 teraflops. The other interesting machine in the list is a cluster 1,540 1GHz Itanium 2 processors built by HP for the Pacific Northwest National Lab, which is rated at 4.9 teraflops.

SGI had 54 machine in the Top 500 ranking this time around, with 23.2 teraflops of aggregate power, Cray had 26 machine with 15.1 teraflops of power, and Dell had 15 machines with 10.7 teraflops of power. The advent of Linux clusters has pushed smaller RISC/Unix machines that dominated the lower hundreds of the rankings this time last year entirely off the list. Sun, for instance, only had nine machines make this list with an aggregate of 3.8 teraflops of aggregate floating point power.

The entire Top 500 list represented 374.6 teraflops of processing power in 247,125 processors. Academic institutions had 116 of the machines, research institutions had 130 machines, companies in various industries had 202 machines. The remainder were split between government agencies, vendors, and classified government projects.

Only 41 of the machines in the list were based on vector processors, comprising 59.7 teraflops of power. Scalar machines, predominantly massively parallel clusters but also big shared memory machines like SGI’s Altix servers, accounted for 459 machines and 314.9 teraflops.

Intel architecture servers boasted 119 machines and 89.6 teraflops of power across all those machines, bringing it roughly even with IBM’s Power architecture, which accounted for 102 machine and 95.8 teraflops. HP, with its PA-RISC and AlphaServer lines, has 167 machines based on its architectures (although it is not always the primary vendor, apparently) and 91.3 teraflops of power.

Source: Computerwire