As far as the programming is concerned we see in PERCS an x10 programming language, a continuous programme optimization and various programming tools and productivity metrics, Elnozahy said. The system has near memory processing and there is a processor vector morph, support for collectives and different switching structures. At the level of the technology, Mr. Elnozahy mentioned the Silicon carrier and LDSL circuits. There is also a high-speed circuit for fast communications and an advanced packaging for cooling. IBM deploys an architecture that is compatible for the commercial as well as the HPC space.

The supercomputer community is facing a range of technology hurdles. Mr. Elnozahy pointed out that without innovation nothing will be possible.

The PERCS productivity environment is trying to move away from the two level memory: local and global. It offers a performance automization.

The X10 programming model is a new programming model and an experimental language for scalar concurrency. It builds on Java and targets both HPCS applications and commercial server-side applications.

As current issues, the speaker cited software as a big part of the problem. You cannot scale to PetaFlop level if you do not adapt the software. Sustaining innovation is difficult if the community does not embrace it, Mr. Elnozahy went on to say. A huge gap exists between the current practice and the vision of HPCS.

Mr. Elnozahy had some "tough" questions to the inquisition himself. Should we allocate all the new cycles to performance or trade some of these cycles for better productivity? What about the ease of use? And what about more reliability? Should we accept algorithms that are inefficient in the small but scalable in the large? Will the community embrace new tools and a new programming language?

In 2010 peta-scale systems will be used to run and vendors are currently designing unusual hardware features.

There are a lot of technologies in performance and potentially, there is productivity in general. The current practices are at odd.

Inquisitor Ad Emmen wanted to know whether the design is taking into account the needs of the users and he asked what the targeted user needs are in the HPC community. Mr. Elnozahy answered that the analysing applications had taken a lot of time. The general complaint is that the number of processors is not sufficient but in fact it is the cache and its inefficient use and the latency that are important.

An example is that conventional wisdom is looking at solutions but it wants to fix the latency.