2938 Algorithmic Processor

The Gallery is searching for documentation or artifacts concerning the 2938 Algorithmic Processor (sometimes called an array processor attachment). The only information we have found consists of tidbits from Gallery readers, such as Wim Brul (pictured above), and IBM Systems Journal, Vol.8, No.2 (1969, without pictures).

The few leads we have found include the following.

Wim Brul sent us the picture at the top if this page and said:

Hello Curator,

When I searched the internet for information about the IBM 2938 Array Processor, I came across the gallery of old iron where you asked for pictures. Attached you will find a picture of an IBM System/370 Model 145 connected to an IBM 2938 Array Processor. The picture shows On Site support at the IBM Poughkeepsie Plant in August 1975 by Wim Brul (2nd from right). The system was shipped to an Indian Oil Exploration Site.

The 3145 side of the 2938 Array Processor Attachment Feature was designed by me in 1971 when I worked for the IBM Special Engineering Department in Uithoorn, the Netherlands. It replaces the system board for block channels 3/4.

You must know that I was only involved in the 3145 side of the attachment. Here some more info:

IBM 2938 Array Processor attached to IBM System/370 Model 145.

This project started out with a document of understanding between the developers responsible for both sides of the interface. This included interface signal specification and timings. After that the design and development was done independently from each other.

The hardware modification on the 3145 side consisted of a system board replacement of block channels 3/4 with a special designed system board populated by newly designed logic cards that interface the control and data busses to the central processing unit and main storage. There was also a microcode change necessary to access to the 2938 at channel 3.

The integration test was performed at the IBM Development Laboratory in Uithoorn, the Netherlands. There appeared to be a timing problem. This was resolved by a minor off spec change to the microcode of the 3145.

Best Regards,
Wim Brul

George Paul dropped us a note to say that he designed the FFT algorithm for the 2938. He believes there is a picture of a 2938 on the cover of one of the IBM annual reports in the early 1970's (but we've been unable to locate it).