Antonio Carlini <arcarlini(a)iee.org> wrote:
I think the origins of the GPX go back to the QDSS
Q-bus
board set as used in the VAXstation II (the "Dragon" chipset).
I know. I meant that the GPX used in VS3100s originated in VS2K.
My recollection is that the same GPX board was indeed
used on
the KA410 and KA42s. That was the VS40X-MA.
I thought so too. But what about SPX? Can it go into a pre-M76 VS3100? And what
about VS2K SPX? (It would of course be very silly in practice, but I'm talking
in principle.)
I *think* that in the KA42 systems, they converted
from CDAL
to EDAL (or whatever) as necessary.
"As necessary"? I thought that except for memory KA42 is 100% EDAL. At least to
system software KA42 is basically a faster KA410. (Well, there are some tweaks,
like the ugly hack to make LANCE address 32 MB instead of 16. On KA410 they
simply got lucky that their system memory size just happened to be what LANCE
can address. Then I recall that the NetBSD folks had discovered that for SCSI
the DMA byte count had to be set off by one between the two or something like
that. I don't remember the details, but the question can probably be settled by
comparing the KA42 SCSI driver in Ultrix sources against the KA410 st driver
and the KA410 TM.)
So I thought that KA42 had one big CDAL-to-EDAL bridge upfront and the rest of
the system except memory was EDAL. But I could be wrong, maybe different
subsystems have their own independent connections to CDAL.
But if KA42 indeed has one big CDAL-to-EDAL bridge upfront, the million dollar
question becomes: why did the VS4000 M90 dev team toil to design their own
CDAL-to-EDAL bridge (CEAC) if there already was one? The only plausible
explanation I could come up with is that perhaps on KA42 the CDAL-to-EDAL
bridge was inseparably integrated with the memory controller.
BTW, I have never found any references to a technical manual for VS3100 (any
model) or for the corresponding early MV3100 models. It looks like one never
existed. Do you have any more info?
The KA43 was a Rigel chip
shoe-horned into a CVAX system, so they did something to
convert the Rigel bus (RDAL?) to CDAL and then left as much
of the rest of the box alone (I don't have the KA43 stuff
to hand so I may be misremembering the exact details here).
Does KA43 have memory on CDAL or on RDAL? I once had one in my hands and when I
looked on the board to see what chips it had, I found the P-chip and the F-
chip, but not the G-chip. The G-chip (used on KA670) is a Rigel memory
controller and an RDAL-to-CDAL bridge combined. (The VAX 4500 team got the NCA
idea from it.) I then thought that having no real need for CDAL KA43 went
directly from RDAL to EDAL, perhaps combining the RDAL-to-EDAL bridge with an
RDAL memory controller. But I guess they could have also made a chip like the G
but without the memory controller and then plopped the KA42 memory and I/O on
CDAL. That could explain why KA670 pulls 8.0 VUPs and KA43 only 7.6.
But you are right in that the gap between KA42 and KA43 is much smaller than
between KA410 and KA42.
In the VS4000-90, the NCA is (IIRC) an NDAL-to-CDAL
bridge.
Then the EDAL hangs off the CDAL to give access to some of the
internal options that presumably were leveraged from earlier
CVAX designs (KA42 etc.). But the graphics hang off the CDAL.
As you note, these are *not* the standard SPX etc. but designs
that presumably had already been modified from the original
SPX
Yup.
to remove the EDAQL interface and use CDAL instead
Yeah, maybe that was the change. (Was that a typo or was EDAQL a chip
converting EDAL to SPX's internal bus?)
I'm guessing, I
cannot find the article I'm sure I've read that describes
the development of the VXT2000).
http://www.research.compaq.com/wrl/DECarchives/DTJ/DTJ402/DTJ402SC.TXT
But it talks about the X aspects of it and says nothing about VXT2000 hardware.
MS