On Tue, 22 May 2001, Gunther Schadow wrote:

> Damn! And I must have gotten the first edition that doesn't contain
> any of the specific stuff I need to know about my VAX 6000. I was
> actually quite disappointed about the VARM as it contained little
> more than the VAX Architecture Handbook. Basically it comes down to
> an annotated list of machine instructions. Fine, I really would have
> liked to learn about all the different generations of the busses
> Unibus, QBus, XMI and BI, etc. But nothing to be found. Now you say
> I just should have gotten a later edition?
>
> Anyone wants to trade? My copy is cosmetically quite nice, and I
> don't care about that at all here.

Well, even the later version of the VARM doesn't cover the busses or
things like that.  It's specifically covers the features of the VAX
processor and then dedicates two chapters to the vector processor.

Appendix B (pp 531-545) of the book gives some very brief descriptions of
all VAX systems produced up until 1989-1990.  This edition was printed in
1990.  The VAX 6000-400 is mentioned though, with the following paragraph
worth noting:

  This VAX system is the first one to offer integrated vector processing.
  A single board vector coprocessor can be attached to a scalar processor.
  Up to 2 scalar/vector pairs are supported.  The vector coprocessor runs
  at a 45ns cycle and achieves peak megaflops rates of 45 megaflops in
  double precision, and 90 megaflops in single precision.  The vector
  coprocessor has it's own 1MB cache.

So I imagine with the maximum of two scalar/vector processors, you'd be
doing roughly 90 Mflops double precision and 180 Mflops single precision.
Not too bad.

The VAX 9000 on the other hand can have up to 4 scalar/vector processors
with the scalar processors running at 30-40 VUPs each and the vector
processors peaking at 125 Mflops each.  Vroooom.

-brian.