Strictly speaking this system is nor new, nor under development, nor will it be commercially available. In fact, it is an existing system at the University of Tsukuba that has all characteristics of the Hitachi SR2201, except that it has twice as much processors as the maximum available at the SR2201, viz. 2048. We mention the system because it shows the close involvement of university application groups (in this case the physics department) in the development of the future systems of a vendor. This kind of cooperation is not limited only to Hitachi: similar examples were the development of the NEC SX-2 in cooperation with the Institute of Laser Engineering at the University of Osaka, and of the Fujitsu VPP500 with the Japanese National Aerospace Laboratory.
As said, the CP-PACS is almost identical to the SR2201 but the number of processors is two times larger: 2048. The machine features in \cite{Don96} with a measured performance of 368.2 \gfl for the solution of a 103,680 order linear system. The theoretical peak performance is 614 Gflop/s. It is a fair guess that Hitachi will pursue this line of architecture further: the 3-D crossbar in the SR2201 and the CP-PACS is a very successful component and also the pseudo-vectorprocessing introduced with this generation of systems is quite promissing. When this would be combined with the PA-RISC 8000 processors for the nodes a quite interesting system would result.