First he pointed out Cray is also succesful in Japan, for instance with Toyota, which was unthinkable a number of years ago. Also the Indian market is increasingly becoming important for the company.
The overall Technical Road Map of Cray is heading towards what the company calls "Adaptive Supercomputing". Different types of applications can run on a system with different types of processors, memory, interconnects, etc., that are automatically reconfigured with a uniform interface to the user. Part of the development is done under the DARPA HPSC programme.
The system will contain scalar processors, vector processors based on "Black Widow" and multithreaded processing "Eldorado" based on the MTA processor.
As a first step, an intermediate system called "Ranier" will be available in 2007. By then also the Black Widow and the Eldorado systems will be available for customers. Ranier will then be developed in a full "Adaptive Supercomputing" architecture that is code-named Cascade.
Cray is still targeting the high-end HPC market. This does not mean they must be on the first position in the TOP500 - although Steve Scott would not mind - because the machines now on top are specialized systems. But in general machines in the +1 million euro range are of interest to Cray.
With the Cray3D they also gained experience in the mid range market and how to work with distributors. This was not a big success, but gave Cray a lot of experience not only in addressing this type of markets, but also with applying FPGA's.
Steve Scott does not see Microsoft replace Linux as the operating system for HPC systems. What Microsoft could do is help with developing parallel languages and tools. With multicore processers, parallel processing will become important on almost all computers, even the desk tops. The current parallel tools, like MPI, are not really suited to address that.
According to Scott, the main problem is not so much that there are no good parallel tools and language ideas and concepts, but that there is not one widely accepted standard. It is a cultural issue, not a technical one. The DARPA HPCS programme is addressing this, and most big vendors participate. Cray's contribution is the Chapel parallel programming language. The idea is that the venndors in the end agree on one language they will all support.