Commodore Gemini dual 030 card and Buster magic

[PurpleMelbourne]

In the Commodore labs they made a dual 68030 card called Gemini.
Its elegantly simple how it works.

_____________________________
............Amiga 2000 motherboard..................
.............68000.....Agnus...................
.....................\..../............................
.....................Buster.........................
........................|...............................
.............A2000 CPU slot...................
_____________________________
..............Gemini twin CPU card....................
........................|...............................
....68000 to 68030 conversion PAL....
........................|...............................
..............Arbitration PAL...................
.........../..........................\................
....Fat Buster...............Fat Buster......
...../.....\....................../.....\.............
..030.--.Ramsey.........030.--.Ramsey..
..............\..............................\..........
........4MB ZIPs.................4MB ZIPs....

Each 030 has its own local memory and can access the other CPU's memory as expansion memory.

Fat Buster has some magic inside which dates back from Dave Haynie creating the Buster chip to evolve the B2000 from the original A2000 design.
So some extra feature there lets Buster and Fat Buster make a co processor type system.

I think the same thing can be done with three PCI Bridge chips.
Each CPU has a PCI Bridge chip both in master mode, and there is then an extra PCI to PCI bridge gender bender between them which is in slave mode on both sides.

Without much change to the Amiga OS, one immediate use for a co processor system would be to give ShapeShifter a CPU all of its own.

At the time of development the problem was that programable logic was too slow. But the current stuff is down to 10ns delay with 5ns in and out of the chip, plus another 5ns through the logic across the chip.
Which is only four cycles at 50Mhz. With this we can do what was proven at Commodore, but not functional without properly manufactured ASIC chips. We can do it with cheap CPLD's, or FPGA's if we want to add advanced features like Level 2 cache for the memory.

With IDT QSpan PCI bridge chip on a pair of 040's or 060's joined with a Texas Instruments PCI2050 we can probably do the same thing via a PCI bus using these off the shelf chips.

Its often brought up that these old CPU's can't do SMP, but that's wrong.
Sony had a dual 030 server called the SonyNEWS which ran a variety of BSD UNIX on it.
http://aminet.net/package/util/misc/Executive in 1997 imported BSD schedular functionality to Amiga without requiring the Amiga source code to do it.
BSD supports multiple 68k cores, and we could use that functionality. There wont be any splitting a single program across two CPU's, but we'd be able to multitask the system better under heavy conditions, and newly written software could take advantage of this ability.

_____________________________
............Amiga 2000 motherboard..................
.............68000.....Agnus...................
.....................\..../............................
.....................Buster.........................
........................|...............................
.............A2000 CPU slot...................
_____________________________
..........PCI master / PCI slave twin 68040..........
........................|...............................
...68000 to 68040 conversion CPLD...
........................|...............................
..............Primary mc68040...............
........................|...............................
..............QSpan PCI bridge................
...................PCI master.........................
........................|...............................
....................PCI slave..........................
...............Texas PCI2050...................
....................PCI slave..........................
........................|...............................
...................PCI master.........................
..............QSpan PCI bridge................
........................|...............................
..............Secondary mc68040...........

If the Texas Instruments PCI2050 were reproduced in FPGA to make a double slave bus, then it could join two master PCI bridges to reproduce the original Fat Buster type bridge.

_____________________________
............Amiga 2000 motherboard..................
.............68000.....Agnus...................
.....................\..../............................
.....................Buster.........................
........................|...............................
.............A2000 CPU slot...................
_____________________________
............PCI Buster twin CPU card...................
........................|...............................
.......PCI Gemini Buster CPLD............
........PCI slave.................PCI slave.............
.........../..........................\................
..........|.............................|...............
.......PCI master..................PCI master.........
......x-QSpan................y-QSpan........
..........|.............................|...............
....X-MC68040.............Y-MC68040....

[britelite]

Its often brought up that these old CPU's can't do SMP, but that's wrong.

Really? I can't recall anyone claiming anything like this, only that the Amiga hardware and OS doesn't support it.

[Fastdruid]

Originally Posted by PurpleMelbourne

Without much change to the Amiga OS, one immediate use for a co processor system would be to give ShapeShifter a CPU all of its own.

I'm not sure there, the notes on Gemini say that each core was running a modified copy of AmigaOS. I suppose from the timescales it would probably have been 3.0.

That's not really SMP, possibly more akin to a Beowulf cluster. It was only ever an experiment rather than a product so its really hard to tell. You can imagine how useful it could have been though, for example I'm sure those using the A4000 for Babylon 5 would have loved the ability to fit a pair of Gemini type cards in each A4000 and cut rendering time.

It's also arguably a NUMA architecture as each processor has "local" memory but can access any other memory as well.

Originally Posted by PurpleMelbourne

Its often brought up that these old CPU's can't do SMP, but that's wrong.

Less can't and more just didn't.

SMP goes back to the 60's and there were significant advancements in the 80's (arguably the Inmos Transputer was 20 years ahead of its time) but when each new processor was massively faster/better than the one before there wasn't any use outside the "server" market until Intel started hitting the limits of ever increasing speed and branched out into more cores.

That said, I found it slightly surprising that SUN didn't go multi-processor earlier, I think it was 1991 or so, long after they dropped the 68k.

Originally Posted by britelite

Really? I can't recall anyone claiming anything like this, only that the Amiga hardware and OS doesn't support it.

Yep, the Amiga does AMP not SMP, although it does do multi-processors it's a co-processor system. Ignoring gemini as well as copper etc there are even actual hardware examples of using more than one CPU, the A2000b/B2000 supports using the original 68k as a co-processor with the priority going to the CPU in the CPU slot (due to Buster and it giving priority to Bus Request from the CPU slot, AKA /CBR)
. I don't think anything ever _used_ that ability, but it was there.

[Kin Hell]

Originally Posted by Fastdruid

<snip>


SMP goes back to the 60's and there were significant advancements in the 80's (arguably the Inmos Transputer was 20 years ahead of its time) but when each new processor was massively faster/better than the one before there wasn't any use outside the "server" market until Intel started hitting the limits of ever increasing speed and branched out into more cores.

<snip>

Nice to hear someone else partially Rave about the Inmos Transputer Chip.

It's off topic really, but some Company based in the Midlands was looking at using those Transputer Chips in Parallel with Amiga's. Stick how many on a Zorro based card & even into an External box for some seriously Faaaast Render times. Never saw an outcome to it but an old friend running I.T. as a Unix guy for Inmos, we had some serious Boners for that hardware concept in it's day.

The Joys of being a 3D Ray Tracing nut!

[PurpleMelbourne]

Also how's this model for adding a WarpUpPPC PCI expansion card?
I think it should work to allow both CPU's to run simultaneously in asynchronous mode. If I remember correctly, the problem with the CyberStorm system I think is they need an extra Bridge in their to distance the two CPU's from each other as the PCI bus locks the CPU's to be one as master and one as slave while they both will only work as master. As it is they must shut one down for the other to operate alone.

A Gemini card could have the PCI bridge on their to allow the WarpUP. Between two 060's and a PPC, you'd be able to do quite a lot of processing and all in parallel.


There are a lot of lonely PPC's out there which could do some useful co processor work. Maybe as a Linux Bridgeboard running JAVA, Python or something else on the side?

_____________________________
............Amiga 2000 motherboard..................
.............68000.....Agnus...................
.....................\..../............................
.....................Buster.........................
........................|...............................
.............A2000 CPU slot....................
_____________________________
............68040 with Warup PPC via PCI.............
........................|...............................
........FPGA SDRAM with L2 cache.......
........................|..........\.....................
..................MC68040....SDRAM.............
........................|................................
.............QSpan PCI bridge.................
.............PCI-1 master.....PCI-0 master............
........................|................\..............
....................PCI slave..............\............
..........Texas PCI2050.........PCI RTG......
....................PCI slave............................
........................|................................
...................PCI master...........................
.............QSpan PCI bridge..................
........................|................................
...............mpc8260 PPC......................
........................|.................................
...............PPC Local SDRAM.......................

[Kin Hell]

Originally Posted by PurpleMelbourne

<snip>

There are a lot of lonely PPC's out there which could do some useful co processor work. <snip>

When the PPC Cards were conceived, that was the Idea. It never happened though & then only 7 years later, OS4 showed is embarrassed face as an Epic Fail running on a PPC @ 233Mhz under Emulation.

[grelbfarlk]

Maybe look at the Sonnetlib thread, we have been doing this for years now, with the caveat that we have to go through the slow Zorro-3 bus via PIO on Mediator (and maybe someday with Prometheus). What would be great is to extend the 040 bus from the 040 slot in whatever Accelerator you have to a fast PCI bridge, bypassing slow Buster and hopefully being able to DMA to the PCI space (quickly).

The other option is add some larger FPGAs to the Prometheus design so we could at least have DMA to the PCI bus, it would still not be very fast but it could open the door to some things which are not currently possible.

[Juz400]

Originally Posted by Kin Hell

Nice to hear someone else partially Rave about the Inmos Transputer Chip.

It's off topic really, but some Company based in the Midlands was looking at using those Transputer Chips in Parallel with Amiga's. Stick how many on a Zorro based card & even into an External box for some seriously Faaaast Render times. Never saw an outcome to it but an old friend running I.T. as a Unix guy for Inmos, we had some serious Boners for that hardware concept in it's day.

The Joys of being a 3D Ray Tracing nut!

I do remember something back in the day with maybe a Fluro green case and going by the name `Raptor`
I dont remember anything more than a little writeup in a magazine..


I did wonder if the Dragon coldfire card would be something along these lines (RIP)
https://www.vesalia.de/e_dragon.htm
You have all the good stuff like PCI, Networking, DDR ram, list goes on.

Did things go wrong when they started trying to get the Amiga OS running on it?
Why not use it as its `own` system with a Zorro-3 bus?
The coldfire could handle all the PCI DMA to its own RAM and service the Zorro3 bus and whatever the Amiga wants through the bus.

There should be plenty of free processor time left in which to pass `tasks` over to it as a Co-Processor.
No running or trying to get AmigaOS running on it.

[eXeler0]

Originally Posted by Kin Hell

Nice to hear someone else partially Rave about the Inmos Transputer Chip.

It's off topic really, but some Company based in the Midlands was looking at using those Transputer Chips in Parallel with Amiga's. Stick how many on a Zorro based card & even into an External box for some seriously Faaaast Render times. Never saw an outcome to it but an old friend running I.T. as a Unix guy for Inmos, we had some serious Boners for that hardware concept in it's day.

The Joys of being a 3D Ray Tracing nut!

The Atari ATW800 sold less than a few hundred units. I guess the "general market" wasn't ready. Software support for multi processor hardware wasn't in a good place back then.

[Kin Hell]

Originally Posted by eXeler0

The Atari ATW800 sold less than a few hundred units. I guess the "general market" wasn't ready. Software support for multi processor hardware wasn't in a good place back then.

Now I didn't know that.
Why on earth did or would anyone want a Transputer version of Atari TOS?

*cough* I shouldn't laugh, but Atari just were not in the same league as the Amiga at any time I remember.

Also & iirc, the Inmos Transputer Chip was well known for it's use in the "Rapier Missile" Guidance System.

I remember when Jurassic Park was released back in '93, all the wire frame animations were Rendered on Amiga's because nothing could render so smoothly. I understood all the Textures were rendered on Raptor Boxes in Real Time but all that could be he say, she say....

[PurpleMelbourne]

Originally Posted by Juz400

I did wonder if the Dragon coldfire card would be something along these lines (RIP)
https://www.vesalia.de/e_dragon.htm
You have all the good stuff like PCI, Networking, DDR ram, list goes on.

The trouble with the cold dire is they dropped enough instructions that it's not 68K compatible anymore.

It's a pretty good CPU, but now also languishing.
I've got a friend who does racecard computers with them, and there are a lot of annoying bugs making some things not work.

As for PCI on Amiga, once I started looking into PCI a year ago I found there are options we could already be using.

After these little projects will come an ITX or ATX Amiga PCI motherboard with 060 all the way, not 030 design!
To have the full throughput, the PCI needs to go straight to the CPU.
Will anyone be interested in a PCI version of PicassoIV?

[Fastdruid]

Originally Posted by PurpleMelbourne

Will anyone be interested in a PCI version of PicassoIV?

IIRC isn't AGP a subset of PCI anyway? If you're going to go to the effort why not go the whole hog and do an AGP PicassoIV?

[PurpleMelbourne]

AGP is a DDR version of 66MHz PCI yes.
But the Cirrus Logic GD5446 has a PCI interface. So it's PCI unless I figure out how an FPGA can read the ram direct to bypass the PCI bus.

I should probably not actually refer to it as Picasso IV, but people know what it is.
Alpine is the name of the chip series... so mabe AlpineIV?

Getting a little off the A2630 topic though

[Kin Hell]

Originally Posted by PurpleMelbourne

<snip>

Getting a little off the A2630 topic though

Yeah fella, sorry to be one of the culprits.

[Hedeon]

Originally Posted by PurpleMelbourne

Getting a little off the A2630 topic though

Try a search on A2630 in this topic and see what the results are ;-)

[Fastdruid]

Originally Posted by PurpleMelbourne

Getting a little off the A2630 topic though

That was the other thread...this is dual CPU's and buster magic.

Anyway, interesting though the Atari is (it didn't run TOS btw) going back to Amiga and multi CPU (and also Transputer)

This page has a bunch of interesting info on it including a comment by Dave Haynie that in summary is "it was too expensive for not enough performance".
http://www.amigahistory.plus.com/pro...ransputer.html

Would have been fantastic for render farm type duties if it had been cheaper but actually having the production version of Gemini Zorro card with dual 030/040 would probably have been more cost effective.

[AmigaHope]

How on earth would this thing maintain cache coherency? I guess 030 caches are only 256 bytes but it'd still be an issue for IPC...

[grelbfarlk]

Originally Posted by PurpleMelbourne

Will anyone be interested in a PCI version of PicassoIV?

Why would we want a Cirrus Logic 5446 PCI card, when there are lots of those available for about $10-20? Even VillageTronic that made the PicassoIV made these in PCI flavor as well.

Also while we have Voodoo 3, 4, ATI Radeons, Permedia 2, all of which are superior to the Cirrus Logic 54xx series. There are also PCI graphics cards from the newer generations, the model numbers escape me at the moment but I want to say I've seen PCI versions of the Radeon 5450 HD series, X700, X300...

And if we ignore the pointless part of redesigning the wheel for low volume and high price, there's still the problem of drivers. If you want to build a new graphics card, just write some P96 drivers for one of the many PCI graphics cards we don't have support for now, I've heard on good authority this is really easy and should take about 15 minutes™

[Juz400]

Originally Posted by PurpleMelbourne

The trouble with the cold fire is they dropped enough instructions that it's not 68K compatible anymore.

It's a pretty good CPU, but now also languishing.
I've got a friend who does racecard computers with them, and there are a lot of annoying bugs making some things not work.

I think you missunderstand my usage case for the CPU,
The Amiga should not see it as a CPU, its the translation layer between Amiga and the new peripherals and RAM
They `share` Zorro3 bus, one or the other has Master as required, the Coldfire should be fast enough that the Amiga does not wait for for it.
It would not matter that has missing instructions as it is NOT running the Amiga OS, It will run its own limited `OS` to watch and respond to which ever BUS wants to move data.
With CPU on both sides of the Zorro we should be able to fine tune the Coldfire OS push Zorro3 bus to its limit.

The price range is from £30 to £60 per chip, DDR Ram is peanuts from ebay making the Idea pretty wallet friendly
Coldfire with Ram slots on each PCI slot?
The `Interface` Coldfire could identify a CPU on each PCI slot and `upload` a small OS to each to again watch the BUS and work on data modules sent to spare/idle slots.

Mmmmmm..
Anyway, we have missed the Boat as Mouser et all recomend these not be used for new products as they are nearing EOL

[Gorf]

Originally Posted by PurpleMelbourne

In the Commodore labs they made a dual 68030 card called Gemini.
Its elegantly simple how it works.

As far as I know the Gemini was a Zorro III card and meant to be used in a A3000. At least that's what all webpages claim about this card ...
And it makes more sense too, doesn't it?

So the setup would look more like this:


______________________________________
............................Amiga 3000 motherboard.................
.....................68030.....68882.....Ramsey.........
.....................................\..../............................
..................................Fat Buster.......................
........................................|...............................
.............................A3000 Zorro III...................
______________________________________
..............................Gemini twin CPU card....................
.........................../..........................\................
.Fat Buster (reverse mode)...Fat Buster (reverse mode).
...................../.....\....................../.....\.............
.........030..68882..Ramsey.........030..68882..Ramsey..
..............................\..............................\..........
....................4MB ZIPs....................4MB ZIPs....


So the Busters take only care of the transaction from 030-bus to ZorroIII and back again.