Machine type: Vectorprocessor.
Models: VP2100/10, VP2100/20, VP2200/10, VP2200/20, VP2400/10, VP2400/20, VP2600/10, VP2600/20.
Operating system: VSP/S (IBM MVS compatible) or UXP/M (a Unix variant acquired from Amdahl).
Compilers: Vectorising Fortran 77/VP both under VSP and UXP, C.
Note: and are given for the largest configuration.
The speed differences between the different models stem from replication of the multi-functional multipy/add pipes in the models 2100--2600. The VP2100 has one pipe set, the VP2200 two, the VP2400 four, and the VP2600 eight.
The multiply-add pipes as employed in the VP2000 series is of a type that is seen in an increasing number of vectorprocessors. It is capable of a combined multiply-add operation in which case 2 results per clock cycle are delivered. The same pipe also does single adds and multiplies. A separate pipe is used for the floating division operation. The clock speed of the scalar processors is two times lower than that of the VPU. With regard to Amdahl's Law which asserts that the slowest component of the system will be the decisive factor this design choice seems strange. For highly vectorisable programs the effect is small, however.
Just as in the former Fujitsu VP series (VPx00) the vector registers are not fixed in length, but can be reconfigured according to the vectorlength encountered. This may avoid or minimise stripmining thus increasing the performance.
In the VP2000 systems each vectorprocessor can be acommodated with 2 scalar processors (in which case the suffix /20 is used. The 1-scalar-processor models have /10 as a suffix). Multiple scalar processors should give a considerable better throughput when many users access the system because of less frequent context switching. A factor of 1.5 is quoted by Fujitsu.