Jump to content

Sonnet Quaddoubler project

Recommended Posts

Looking for a Quaddoubler at a decent price...now here is the caveat. I want to destroy it.......in the interest of reverse engineering one and then being able to get them at a decent price. Many cards have been reversed engineered in Amiga land and this would be a very beneficial one for the Amiga and Mac side


Link to post
Share on other sites

To cast the net a bit wider, the 50 MHz Newer Technology Quadra Overdrive upgrades were easier to find, and should work even better for any 68040 machine expecting a 25 MHz 68040. One was apparently sold on eBay as an Amiga 4000 upgrade.


I've never found the Sonnet version in the past decade and a half that I've been collecting old Macs.

Link to post
Share on other sites
  • 68kMLA Supporter

Seems they made a few different versions of the QuadDoubler that were rotated in different directions to accommodate different systems (according to Amiga Hardware Database):

> available in different pin orientations - the QuadDobler model number BSW2001004950 installs parallel over the A3640


Different part #s I've seen (mentioned here, and on Sonnet's page here and here):

BST-40F - Centris 610 (and presumably Quadra Workgroup Server 60/20)
BST-50FA (BSW2001004950) - Quadra 610 Workgroup Server 60/25, also Amiga A3640 & "X-Calibur" 

BST-50FB - 100/50Mhz Quadra 660AV, 700, 900, Centris 660AV

BST-50FC - Centris 650


Presumably by looking at Mac models you'd be able to determine how they're rotated to fit in the chassis.


As far as rarity - good luck. Saw one at beginning of last year, and one at beginning of this year, and that's it. I missed getting those, been trying to get my hands on one for a long time. 


Attached are some images I've collected of ones in the wild over the years, maybe they'll help your search: 


From an eBay listing 2018 Feb by taddy-izumi




From ??



From someone on reddit a few years back:





From Quadra 700 listing 2019 Jan:






This old thread now since lost in 68kmla transition had some images of an attempt to put one in a Performa 630. I believe this was @johnklos. Notice the addition of spacers, and the "spectrum engineering" label:






Mentioned by @nglevin, this is actually the Newer Technology Quadra Overdrive.


According to this post on AF, it came in a regular and PLUS version, the latter having L2 cache (!). 

Edited by nickpunt
Link to post
Share on other sites

The QuadDoubler / Quadra Overdrive only runs the processor at double speed when the bus relinquish signal is active. This way, literally everything that happens while the '040 is interacting with the motherboard happens at standard speed so there are no incompatibilities, and when no bus activity is going on (therefore the bus is relinquished), then the '040 is run at double speed. Stuff running in the caches will run faster, but memory intensive things won't.


italux: Contact me directly. I might have an extra I'm not using.

Link to post
Share on other sites

Makes sense, was wondering about interaction with the system bus and whether the 68040 might have asynchronous capabilities similar to math co-processors. I guess they work in a similar fashion on the CPU bus when in use. The CPU would be set up to sync with the FPU's independent clock?


If that's the only operation of the GAL aside from its clock multiplying duty, it seems to me that there wouldn't be a need for much reverse engineering at all. Formulas can be developed without a need to read the GAL if necessary and the layout would be straightforward. Mimic the support circuitry and build the bus.


Is a faster 68040 available in SMT for embedded applications? Designing for TQFP alone would decrease the size of the board significantly, especially if the support components could be moved to the sides of the CPU rather than putting them in that center section extension between PGA sockets. Sonnet's solution of putting everything but the PGA/CPU within the footprint of the PGA/Host is quite elegant, so that would be better yet.


edit: forgot to mention that I was thinking along the lines of reducing overall footprint to make the unit's host compatibility a wider range. The Sonnet component centralization trick would make designing cards for CPU placement at any point of the compass a snap wherever necessary.

Edited by Trash80toHP_Mini
the usual
Link to post
Share on other sites
  • 68kMLA Supporter

Changing the clock speed of the 68040 while it is running might be tricky.   It might also be as simple as changing the clock that's being input.  But some chips want configuration gymnastics when changing speed.

Link to post
Share on other sites

Looks like cloning the Sonnet board is the winner, does 2596 jibe with week 25, 1996?




That would likely make this the most recent, certainly the most efficient in terms of PCB footprint and KISS compliant version. Such might explain the elegance of its design. Whatever card shows up for GAL formula spelunking, reduction of component count to this level and using a modern equivalent of the GAL20RA10B seems the obvious choice.


Thanks to nickpunt's longstanding search efforts we have a fabulous collection of trace details to build a large portion of the schematic.










Looks like the detail is probably good enough to match up vias surrounding the GAL topside with traces on the solder side and determine which may be thruhole and which must be blind connections to sub-surface layers. Given a graphic built up from these samples with pin/signal assignments we'll be good to go for examining function and formula requirements before having any board in hand at all.


Link to post
Share on other sites
21 hours ago, trag said:

Changing the clock speed of the 68040 while it is running might be tricky.   It might also be as simple as changing the clock that's being input.  But some chips want configuration gymnastics when changing speed.

If we can't find the answer to that in the docs it might tested using stacked machine pin socket extension/riser layers with clock cutout and a breadboard set up for overclocking on the fly, no?

Link to post
Share on other sites
  • 7 months later...
  • 4 months later...
  • 1 year later...

Would love to revisit this.


First, figuring out a replacement GAL shouldn't be too hard. After all, it's just a matter of switching clocks between when the bus is and isn't used.


Second, I'd love to figure out a way to be able to configure clocks so that the motherboard can be run at one clock, and the CPU can be run at motherboard clock times 1.5 (33 MHz motherboard, 50 MHz CPU), or time 1.25 (40 MHz motherboard, 50 MHz CPU).


@italux: Are the files / notes you've made so far publicly available somewhere?

Link to post
Share on other sites
  • 68kMLA Supporter
Posted (edited)

I'd personally vote to put effort into recreating the version with 128k cache onboard. This should alleviate many of the conditions which otherwise downclock the CPU to motherboard speed, unless I misunderstand how the cache loads from/stores to RAM. 


Thanks to @Unknown_K for pictures of this one, which does operate at 1.5x (50MHz on Q950s) with the 128k cache.



Edited by jeremywork
Link to post
Share on other sites
  • 2 weeks later...

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Create New...