It is not the strategy of HP to build specialised supercomputers. HP high-end computers use a highly competitive node infrastructure and software, have high bandwidth and low latency. This makes them suitable for HPC.
This proves to be a right choice. HP is now number one in the TOP500 with 168 systems. This number is a combination of PA-RISC-based Superdomes and Alpha-based clusters. Superdome has been used successfully in commercial areas for quite some time. But we also see that Superdomes are now replacing vector machines in industrial CAE-environments, says Frank Baetke.
BMW is an example of a company that is consolidating a lot of technical computing on servers from the Superdome family. From the eight machines that BMW has in the TOP500, more than any other individual computing site on the list, 7 are HP Superdome machines.
Within the coming years, the two server product lines, Alpha-based and PA-RISC-based will both see next generation Alpha and PA-RISC processors respectively - but finally everything will be consolidated on Itanium-based platforms.
Because of the collaboration between HP and Intel Itanium has inherited some PA-RISC features easing the migration. The new Itanium2 processor - to be announced shortly - features dramatically higher performance then the first generation Itanium processor. Actually, the largest Linux-based Supercomputer installation so far, PNNL (Pacific Northwest National Laboratory) will be based on nodes equipped with the new Itanium processors.
The current Superdome systems can also be field-upgraded to Itanium systems - and indeed some contracts have been signed that insist on this upgrade. As Frank Baetke poitened out, HP-UX is the first professional Unix that is fully ported to Itanium including the full set of tools and compilers. So customers can upgrade without changing the operating environment.
Commenting the TOP500, Frank Baetke said, he considered the number one machine, the Earth Simulator, an interesting experimental machine. In comparison, HP has three Alpha-bassed machines in the TOP10. They are based on technology that can also be used in large volume mainsstream machines.
The pre-merger strategies for TOP500 type of HPC machines fit nicely together. Compaq was working its way top-down. HP was working bottom-up.
HP has a Grid strategy organised in its Gridlab which follows two lines:
- Support the research community
- A professional approach