Sonnet PPC PCI Card

[arnljot]

PIPE DREAM MODE ON:

The Sonnet PCI PPC Card is now in stock: see here at about $60 for a 400mhz G4

If one had the skills. What licenses are needed (WarpUp) to make a driver?

Or would one have to do it by rev engeneering the current WarpUp release? Or are there WarpUp docs available which says how to make a new PPC HAL for WarpUp?

[alexh]

I think you'll find they are not technically compatible with any PCI controller (G-Rex, Prometheus, Mediator).

This product was to be the Shark PPC from Elbox (rebranded by Elbox of course) that in the end never came out, almost certainly because of a technical limitation they could not overcome.

I am surprised they are back in stock, last time this subject was brought up, some Amiga user bought Sonnet's entire back stock!

[arnljot]

@alexh
I know about the sonnet + shark relations. Thats how I first noticed the card as someone pointed out that they are twins, infact spitting images of eachother. Maybe it was you

So now I have two questions:

1) What can make the card incompatible with Classic amiga PCI boards

2) (Still) What resources and legal stuff is needed to make a WarpUp HAL for such a card, granted that a hardware superhero looked at it

[ddniUK]

plug one of these into a Power Macintosh® 7200, 7215, 8200 then grab the Moana ISO and run os4.1


EASY LIFE!!!  :lol:

PS. If I could sell my entire back stock to Amiga users, then I would restock and sell some more  :-P

[arnljot]

@ddniUK

LOL, I know I invited to "Pipe dream mode", but that just... hehe

@alexh
I would guest that for os3.9 they hit legal issues. Maybe no license for the WarpUp HAL.

If it was for OS4, then surely it was legal and not technical. The technical issues must have been related to a coldfire versions 68k handling, no?

[da9000]

The first resource I as a programmer would need is the technical docs (registers, memory map layout, any other info on PCI transactions, etc.) of the Sonnet card. That will be absolutely necessary to get it up and running.

I personally doubt AlexH is right on this one (although he usually is on technical matters), because if the card is a PCI compliant card and the Amiga PCI boards are PCI compliant (how can they not be, they are PCI bus boards afterall), I can't see a physical/electronic reason it wouldn't work. It [the reason] must all in the software and documentation.

A possibility is also that Sonnet supplied a binary-only library that Mac programmers had to use to access the card. If that library doesn't abide by any ABI [Application Binary Interface](other than Mac), then you're in a world of hell.

Want to start something? Talk to Sonnet, ask them for developer documentation. See if they have any NDA or libraries that you have to use. NEVER mention Amiga, but say it's needed for custom embedded systems that are themselves under NDA and you can't talk to them about it :-)

[alexh]

arnljot wrote:
1) What can make the card incompatible with Classic amiga PCI boards

I am making an assumption that there was a technical problem Elbox couldn't overcome without a new Mediator. The economics if they could have made it work speak for themselves, the markup was to have been from $59 to $400! It would have been enormous for Elbox, they will have done everything they could to make it work!

I am not 100% "au fait de" with PCI topology, but I think the Mediator is a Zorro III to PCI bridge and in order for there to be another CPU on a PCI bridge, the Mediator has to be a Zorro III DMA master to allow the PPC to be able to initiate access to Amiga RAM. I am 99% sure that no Amiga PCI bridge cards are DMA masters. Even if they were I doubt very much if the PPC could access anything else other than RAM, I doubt it could access the registers in the custom chips for example.

But why would this matter? Surely it could work the other way around? The PPC could run happily just transferring data between other PCI cards? I can see that might work. But to interact with the host Amiga it would need some shared memory in the PPC's PCI space that the 680x0 could post to, to give commands etc.

Well I bet such a scenario, while perhaps possible under the PCI specification, was never tried before Elbox released Mediator. Asynchronous DMA transfers between two PCI "masters"? Maybe that did not work? Collisions etc.

All pure speculation, unless we get someone in Elbox to talk to us... just pure speculation, and probably bad technology on my part.

Michael Boehmer of E3B would be the best man to consult, he knows this stuff inside out having played with both Zorro III DMA and the Prometheus DMA.

[arnljot]

@da9000 & alexh

Very interesting. I see what kind of limitations one could have like alexh points out. But like he also says, one could invert the roles, so that the 68k moves data for processing on the sonnet. This to overcome an assumed limitation where the G4 on the sonnet doesn´t have direct access to amiga hardware registers.

---
What I am prepared to bring to the table:
1) Contact Sonnet and see what support can be gotten from them.

2) Contact Elbox directly and simply ask.

3) Offer a bounty:
- I have a spare Amiga4000D whith a 060@50mhz holding a 180mhz 604e ppc. I would outfit this with a PCI graphics card, mediator and sonnet g4 card to anyone who has the skills, time and motivation to attempt this.

The bounty will be offered if we can first establish that this might be done. If #1 or #2 fails. Then the endevaour is over.

Id like to hear what bloodline, hans, alexh and piru thinks about this. Platon42, mboehmer, da9000 and other of the old boys too of course!

[alexh]

da9000 wrote:
because if the the Amiga PCI boards are PCI compliant (how can they not be, they are PCI bus boards afterall)

But that is just it, I doubt they were fully tested to the ends of the PCI spec. They worked with the cards that they needed to. Lets ship it. I mean look at Prometheus PCI bridge. It turned out that it didn't work with the vast majority of sound cards because of some unimplemented portion of the PCI specification.

da9000 wrote:
A possibility is also that Sonnet supplied a binary-only library that Mac programmers had to use to access the card. If that library doesn't abide by any ABI [Application Binary Interface](other than Mac), then you're in a world of hell.

Do you not think that if they had dangled money (or at least the promise of money) on a stick Sonnet would have not supplied the source code if they could? Especially so many years after the "end of life" of this product? An opportunity for them to shift all their unsold stock and make a healthy profit?

The licensing issue is a possibility, after all Amiga Inc. initially wanted to sell AmigaOnes. But after the death of the AmigaOne, Hyperion needed as many PPC enabled Amiga platforms as they could to help boost AOS4 sales, you'd think they'd have done everything in their power to help Elbox... no?

[arnljot]

@alexh
I was thinking about the AInc vs. HVOF case, not realising that it´s only from april 2007.

Maybe it was like you first said. That the DMA issues and a rework of the DMA card wasn´t doable cost wise, and that they neither could defend a $400 markup.

I remember once reading that the SharkPPC needed a hardware upgrade to the Mediator. But can¨t seem to find the source now. Also, I think that the Shark PPC was a OS4 only product, maybe they didn´t see it viable to market it for Classic WarpUP customers.

That´s my angle. I want to find out if there is something the os 3.x community can do here.

[alexh]

arnljot wrote:
I remember once reading that the SharkPPC needed a hardware upgrade to the Mediator. But can¨t seem to find the source now.

I cannot say I ever remember seeing that.

arnljot wrote:
Also, I think that the Shark PPC was a OS4 only product, maybe they didn´t see it viable to market it for Classic WarpUP customers.

I did read that Elbox said they did not want to release the Shark PPC until OS4 was shipping. But when I read that... I immediately thought "they cannot get it to work". Otherwise why hold off? It's not like they were manufacturing them.

[da9000]

@AlexH:

You raise some good points about DMA mastering and I didn't know (as I've not owned a PCI bus board) that none could do DMA mastering. And I didn't see any "bad technology speak" on your part. I'd agree that it'd be nice to hear from Michael Boehmer on this matter.

Also the economics speculations make sense (needing a new Mediator, etc.)

I can also believe the possibility that they just wanted to ship without being fully compliant, due to the limited set of hardware that was supported (and apparently, would be supported), thus not being 100% PCI compliant.

As for dangling money infront of Sonnet, I believe that they probably didn't have enough to get Sonnet moved by it. Rememeber that in the big old organizations (well, I guess Sonnet ain't that big, but bigger than Elbox or Matay) it's more of a chore to do something than not to(*), unless they make a certain large percentage of money. So my guess is that yeah, Sonnet would give away source etc., if they had huge stock and they'd think it'd be worth it. Of course, anyone outside A.org and the Amiga scene knows the Amiga is "dead tech", therefore their immediate answer would be "no, it's dead tech".

As for licencing and stuff, not sure what the relationship is between Hyperion and Elbox. It seems that they're not good because if I was in their shoes (few and dying companies in a dying land), I'd consolidate or cooperate immediately.


* To add to this, it's like having a choice between 1 engineer going through old backups to find the Sonnet PPC source code, vs. fixing bugs in their newer products. Assuming this engineer is being paid a measly $80k, then they'd have to wager how many hours it would take him (more like days) to get this stuff together and update any docs needed, etc., etc., and any support, vs. who much value they'd be getting by making their latest accelerators better and faster for the hundreds of thousands of installed Mac G4s and G5s. If I was a decisionmaker at Sonnet, I'd say: work on new code. It's the only logical way, unfortunately.

[arnljot]

Hmm

This from the dragon FAQ might give a hint:

Dragon FAQ wrote:
Q: Will using SharkPPC card be possible with DRAGON?
A: Yes. SharkPPC can be used in the PCI slot of the DRAGON board as well as in the PCI slot of the MEDIATOR board. What's more, due to the enhanced DRAGON's DMA capabilities, the SharkPPC card in the DRAGON PCI slot has direct access to all resources of the Amiga 1200 motherboard as well as to DRAGON's DDR memory and all cards in the DRAGON's AGP and PCI slots.
As we informed many times, SharkPPC production will not start before the final version of AmigaOS4.0 is released.

[alexh]

Interesting stuff.

The Dragon is another thing where I am sure Elbox just couldn't get the technology right. Again pure speculation, but you just get they feel that it was a product that the masses would have wanted and would have paid for. The only reason not to release must have been the performance of the coldfire (with it's associated emulation) being well below that of a 50MHz 060 for unmodified code and so it was not worth going into production (also I think the particular coldfire chip they chose for the prototype went out of production before they could start!)

Again pure speculation, but I do remember Elbox posting performance stats which were considerably less than stellar.

When you consider how much money they must have invested in that program, something very strange must have happened.

[arnljot]

@alexh

Do you think we can find a capable programmer who would be interessted in the bounty I suggested?

I think that if such a person could get it to work under 3.9, then perhaps some of the experiences could be passed on to 4.x

It´t not a big thing if it could be pulled off. Just a nice little extension to the Amiga