A waaaay back machine
Outstanding reminder of my past days at Purdue University. We had a state-wide timeshare network running on IBM-7094 hardware that ran as a batch processing front end to the CDC machine.
On one experiment in our Physics Department, I consulted with the graduate student and his professor because they were burning all of their NSF grant money up on the IBM SYS/360 they used. We recoded the electron density probability study (large matrices and double precision floating point) in FORTRAN on the CDC and it ran like greased lighting -- 4 hour runs reduced to 15 minutes.
You could actually run from the 8 word instruction stack on the processor. If you hand-optimized your code, it was possible to make an inner-product matrix multiplication loop run inside the stack. Of course, this also balanced performance on the two ALU's in the processor.
If I am not mistaken, the initial description of the CPU and PPUs was incorrect in the fact that the PPUs were actually 15-bit computers instead of the referenced 12-bit.