The recent thread about the PowerPC 970 evaluation boards made by momentum rekindeled a flame in my heart, a while ago on the AmigaOne mailing list someone was moaning about the price and spec of teh A1 and flashing their cash saying they would pay someone to develop a beter one, of course as soon as I made a serious offer he disappeared. After this I forgot all about the idea and went back to my day job for Xion Internet Services - designing embeded systems using PowerPC 405 and 400 series processors.
As you can probably tell I am an electrical engineer, I work with PowerPC chips regularly, recently I managed to persuade XionIS to look into expanding their developemnt to include a G3/G4 server system which we could both replace your existing x86 Linux box's and sell. Athough a decision has yet to be made it looks like it should get the go ahead, we have been accepted by Marvell and signed the NDA, we have also been in dicsuccion with Mai, so watch this space. As yet we have not aproached anyone to get AmigaOS 4 running on it as this is not what it is designed for and we are still at a concept phase and are still evaluating chipset suppliers (Also I have no idea who to contact - Amgia inc? KMOS? Hyperion?) If you want any more info on the system or know who to get in touch with to get AmigaOS 4 running on it email me on edward.dore@xionis.com
I also asked about a G5 system, but becauses of the cost of the development boards and the liklyhood of Apple taking up the entire supply of G5 shipcs until IBm can get their 90nm manufacturing process fully working with decent yeilds it was turned down for "the forseable future". Time to get to the point, my question is this.....
Would people be willing to donate to a fund to develop a G5 AmigaOne board and port the OS to it? My idea is quite simple, setup a not for profit organistaion and sell the boards at cost, at elast to individauls, possibly charge companies a small cost for large orders/batches. Possibly even get IBM involved to cut deals on chips and chipsets or cover some initial costs of manufacturing and advertising. I'm sure they would be interested in it if you could get Linux on it, which shouldn't be too hard for a competent kernel hacker!

Excuse me if this sounds like random mumblings, it is 1 o'clock in the moarning and my typing is lousy at the best of times. Also, please understand that nothing I have said here is an official statment of XionIS etc.

Edward

@digitaldisaster

Well, it sounds like a nice concept... I think that Hyperion are the ones to speak with in terms of OS4 being licensed and ported. Good luck with this project.

@Herewegoagain

Thank you I was woried that all I was going to get from this was flamed
If anyone is interested then please post your opinions here or email me. I will try to set up a web site soon (Anyone know somehwere good that cna host it for free - I'm thinking Amiga sites). A combenation of this and the project reality case would make one amazing AmigaOS 4 launch system, although there is almost no way that either the case or the motherboard could be finished in time (Unless OS4 slips to next year)

I have no real figures on cost but if you look at the Teron evaluation systems (750FX) they cost about 4 times what the G3 AmigaOne's cost, so take the $6000 970 evaluation board and you get $1500 (~£805). Now I know this is nowhere near concrete but it's better than any other "estiamtes" I have seen

Edward

@digitaldisaster

I think it sounds like a good ideal. And if I remeber correctly amiga.org was offering to host amiga related sites for free at one time. I don't know if the offer still stands, but you can check. (Some one correct me if I was wrong about the hosting)

Plaz

@digitaldisaster

Flaming doesn't lead to anything that we can use.

Personally I think the project would be rather interesting since I am into electronics too, though I have never fiddled with PPC, but instead Z80, HC11 and PIC.

However, is this really the time to compete?

As it is, we are struggling to get prices down, and our only way is to bring volume up on the boards that are at hand.

I want to think that Eyetech don't charge more than they have to.

To let the market somewhat stabilize first is what I think seems best even though I know it would take some time to finish a new mainboard.

My 2 krowns

@digitaldisaster

I have been having thoughts about what I would do if I had to create a computer from scratch.

At the moment G5 appears quite power-hungry - I don't know if this is the way I would want to go. Looking at the amount of low-power support out there I was having some pretty wild thoughts. Why not a modular G4 design?

I was thinking of something radically modular, connected only by external power and PCI-Express bus. I know the chipsets aren't there yet, especially for PowerPC but one has to dream... :)

Perhaps a very basic A500-style system could have the following:
- External power brick (or internal power supply)
- Core single-processor module
This is where the BIOS, memory and bus controller live
- Basic i/o module (includes hard drive)
- Basic graphics module

There could be a master A500-style case to hold these 'raw' modules.

Each of these modules could be a fairly skimpy circuit board. The trick is they all talk over PCI-Express and each module is SMALL and LOW POWER. No more great honking huge PCI connectors. No heatsinks. No fans. If you want a radically fast graphics card then that particular module would have to do its own cooling separately from the rest of the system. If you want to upgrade your processor you replace that module. Sound familiar?

A more sophisticated system could be built up from:
- External brick or beefier power module
- Core single or dual-processor modules
(perhaps more than one in a system - they would have to
negotiate over the PCI-Express bus - lots'o'fun for the OS)
- Separate I/O module with lots of fancy stuff like Firewire,
USB, 6-channel audio, mouse (two?), keyboard, infrared, etc...
- Separate high-end graphics module
- Separate high-end disk i/o module (RAID, caching, mirroring, etc.)

Perhaps some of these modules would simply have a PCI-Express connector internally and you could plug in off-the-shelf cards?

If you wanted 4 video displays you could plug in 4 display modules.

Want lots and lots of I/O connectors? Plug in multiple modules.

These could be 'stacked' inside custom (injection-moulded?) micro-pizza boxes perhaps only 4" x 6" x 1" so you would build up a tower. You could also design a rack-mount enclosure to contain n modules as required.

Heck make the modules small and low-power enough and it isn't too far-fetched to progress to a laptop design.

Nifty add-on modules could include:
- Back-up battery system
- Solid-state storage (multiple slots for flash media)

This is a lot of design work and probably not worth doing at all but I figured it could be kicked around by smarter folks than I.


There's a big brain-dump. Have fun ripping it to shreds...

@digitaldisaster

Amigaworld offers the hosting service just contact Davey.

Mark

_________________
IceStar Media Ltd.

@digitaldisaster

I would be interested in both high end and low-end models.
(miniA1 could serve the low-end need)

Also... perhaps nanoA1 with embedded PPC CPU would also be interesting...
I could use some 10"*10" size modules..... and some might be interested in building Amiga (rendering/compiling) clusters with those...

I would recommend contacting KMOS, Hyperion and then Eyetech. ( in that order )

Perhaps your board and A1 could share same CPU slot specs and "custom slot" specs? Perhaps you could start with the development of A1 compatible CPU module... A1 users would love to upgrade their CPU to something like 1Ghz G3 with 1Mb L2 or 1.6Ghz G4.

Perhaps Eyetech would be interested in distributing (also) your board?

Other valuable partners could be Individual Computers and Elbox.

Also, how about letting Mac CPU card's to work on your motherboard?

Especially if KMOS & co do not pay any attention, head towards AROS as the development phase OS....


and..... one idea, even if it will never make profit ... simple G3 card for Amiga1200...


UPDATE: Contact also David Haynie, I'm sure he has sack full of advices.
Also, check this out: http://www.amigadev.net/index.php?subpage=developer&function=docs ( The PIOS One System Specification - Graciously supplied by Dave Haynie .... etc....)

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@gregthecanuck

I have also had some ideas of models built out of modules.
Stackable (btw. AmigaTechnologies Walker was similar) or by using dumb PCI-X backplane. Perhaps the PCI-X backplane could be some industry standard backplane?

- CPU on the A1 compatible megarray connector
- CPU connector possibly on a PCI-X card with DDR RAM (to make the backplane "totally" passive and G5 upgrade possible as well as quad data rate memory)
- GPU on a PCI-X card
- sound and I/O on a PCI(-X) card (separately or combined, initial super-IO + sound card from Individual computers)
- free PCI-X slot(s) and PCI slots
etc...

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@Plaz and Mark

Thanks for the info, I will contact both Davey and amiga.org, although I wouldllike to keep it here on AmigaWorld as this is where I sugested it and I generally prefer these forums (hence why I hang out here)

@Wed
This is the perfect time to compete, competition bring lower prices and more choice both of which bring more people to the platform (Look at the apple clones for example). This would be a brilliant incentive for Eyetech to risk a bit of cash and shift production into high gear.
At the moment these boards would be more expensive than the A1-XE and uA1 but also offfer much more processing power, higher clock speeds and a wider, faster bus. These would not compete with the XE, or even the XC (If it exists, I haven't actually seen anything from eyetech on it, although they are amlost certainly working on something to succeed the XE)

@gregthecanuck
The 90nm G5 consumes 45w per chip @ 2.0ghz IIRC, compare that to a P4 Prescott which consumes ~103w at 3.2Ghz (i think, it may be 3.0 ghz or 3.6ghz)
A while ago, before I did any major work on real world systems, when I was just a dreamer, such a system would have been my idea of perfect. Back then I wanted to design a whole new processor to use such a system with all new busses etc. and no legacy whatsoever, make my own custom chipsets, build them into my own purely SIMD (Think AltiVec) processor which would have mutliple cores and run in limmitlessly scalable SMP environments. I soon realised that it would be a mamoth effort for anyone to acheive such a design, just look at Intel's Itanium.
Such a design would be possible but would take so long to create, a lot of ASIC's would ahve to be designed and made, which would cost a not so small fortune. When I have learnt VHDL or Verilog (Haven't decided which yet) then amybe we could mess arround producing prototy[es of such a system, but I doubt it would ever reach a production stage, not that it amtters, it would be fun to do! One way to cheat would eb to license a lot of IP cores, for example USB, AGP, PCI-Ex etc.ad then just build your own modular system, but for that you need money, anyone know a good venture capitalist?

@KimmoK
A mini-itx G5 board is a possibility as Apple have said they are coming out with a G5 iMac soon (IIRC the iMac base is slightly smaller than the mini-itx form factor but someone would have to measure one fore me), but such a board would be very dificult to make and woulld only be on the cards after an ATX or micro/flex-ATX board has been released.
I have been in touch with may about an Apple to A1 CPU slot adapter but I can't get the nessacery info without buying a board so that's another $3,900 that I would need. I will probably use the Apple CPU interface spec for the G3/G4 developed by XionIS. The G5 almost certainly has to have it's own CPU inferface as it would need more power and more pins, I haven't seen the insides of an Apple G5, if someone could send me a photo of the CPU slot I would be very grateful.
I spoke to Jens a while ago and he doens't want to *make* a board to compete with the A1 but I will ask both him and Elbox if they would stock such a board. If Eyetech were allowed to rebadge a board as their own they may sell i.
Someone from Hyperion has allready been in touch with me, I will reply as soon as I have had breakfast.
A toyed with the idea of a PowerPC upgrade for classic systems (1200's mainly). My idea was to use a Marvell Discovery III northbridge, a Marvell SATAII chip, an NEC USB 2.0 chip and a Texas Instruments Firewire 800 chip to make a tapdoor expansion for the A1200. The problem is I would need to either design an interface between the PCI-X slots on the Discovery III and the Tradpdoor slot. The best way to do this would be in VHDL or Verilog and an FPGA. I don't have any knowledge of VHDL or Verilog so I would be learning from scratch, whcihc I would be willing to do but would take time. In preperation for this I just got my hands on the Amiga Hardare Reference Book, all I need now is a PCI-X spec book such as http://www.amazon.co.uk/exec/obidos/ASIN/0201726823/ref=wl_s_3/ref=cm_mp_wli_/026-9823326-6748410?coliid=I3OVUKR07TSH24
I also thought about building a couple of Zorro slots in for legacy upgrades and then why not throw a AGP slot in using the same method Genesi did, a PCI-X to AGP converter, again I would be willing to try it but woukld need an AGP spec book such as http://www.amazon.co.uk/exec/obidos/ASIN/0201700697/ref=wl_s_3/ref=cm_mp_wli_/026-9823326-6748410?coliid=I1AQRO9P37A5O3

@KimmoK

PCI-X would be nowhere near fast enough to handel that sort of data, currently most people use PCI-X 133 which can do just over 1Gigabyte per second. PCI-X 533 can do ~4.3Gigabytes per second PCI-Express would be a better choice, allowing up to 128Gbps per connector. Alternativly a custom connector using multiple 32bit hypertransport lanes, each with a maximum speed of 11.2gbps per lane.

Edit: These figures are wrong, I must have been hlaf asleep when I worked them out, pelkase see page 7 for accurate figures of several bus types.

@digitaldisaster

Are you planning one or two projects, Ie one 1200 upgrade + one regular board, or just one of them?

And if you are planning on using a 970, why not get a quote on IBM's own 970 northbridge (CPC925)?

@digitaldisaster

>I spoke to Jens a while ago and he doens't want to *make* a board to compete with the A1 but I will ask both him and Elbox if they would stock such a board. If Eyetech were allowed to rebadge a board as their own they may sell i.

I think the naming needs to be sorted with KMOS. Eyetech would not need to rebadge anything all that is needed is the will to distribute your boards.

About the high-end and G5 CPU.... I think the most clever way would be to use the "dumb PCI-X backplane" idea. You could use readily available components and change the CPU and RAM when things mature. And later all high end 3D GFX chips will be available for PCI-X anyway.

Also that could blow the HW market wide open. Various manufacturers of CPU cards, I/O cards, GFX...

hmmm... perhaps the AOS4 dongle would need to be on the backplane.

>A toyed with the idea of a PowerPC upgrade for classic systems ...

David Haynie and Jens should be capable to aid you a little bit (Jens masters FPGA, Haynie knows guys who developed AmiJoeG3 etc...).

IMHO: A1200 PPC accelerator needs to be very simple to avoid huge costs.

I suggest three development steps:
A) 1Ghz G3 with 512k L2 and 256Mb of SDRAM soldered on the board (at this point you would only need Hyperion to write the HAL for your PPC card)
B) Later you could make model with mobile Radeon GFX chip. (or jump directly to C))

And may I remind you again at this point, most likely sales would never cover your R&D cost. It would need to be done "for your own fun". Perhaps Elbox or someone else would be interested in a production run of 2000 units, but that's the maximum.

C) Build a replacement for the A1200 motherboard itself !!!! (mobile Radeon with 64-128Mb GFX RAM, preferably AIW, 3D audio, 2 RAM slots, USB2.0, Firewire, etc... and custom connector or faster PCMCIA connector)
(at this point you would need the licence from KMOS)

(Amigans would love that kind of roadmap...)

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@digitaldisaster

"PCI-X would be nowhere near fast enough to handel that sort of data, currently most people use PCI-X 133 which can do just over 1Gigabyte per second. PCI-X 533 can do ~4.3Gigabytes per second PCI-Express would be a better choice, allowing up to 128Gbps per connector."

I do not remember PCI-X speeds out of my memory. But things like 4.3 Gigabytes per second from CPU to GPU should be enough for some time. On the PCI-X idea the G5 would have QDR RAM on it's own (local HT) bus.

But sure, I would not complain about using something faster, but still a standard solution.

"Alternativly a custom connector using multiple 32bit hypertransport lanes, each with a maximum speed of 11.2gbps per lane."

Custom connector should be used only for the CPU.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@Eric_S

At the moment I am planning a signle system, the standalone G5 motherboard, an idea that I had had was for the A1200 upgrade, at the moment that is simply beyond my abilities and also I don't ahve the time to work on both at once so the one that would be more beneficial and have the bigger market is the obvious one to go for, the advantageof the G5 is that not only Amiga users but people that want a PowerPC or G5 system would buy it and run Linux or BSD etc. on it

@KimmoK
By rebadge I mean they could sell it as an Eyetech A1, IIRC this is what they used to do with several upgrades and add-ons.
As I allready mentioned, PCI-X as it stands is simpl not fast enough, it is also not designed with processors in mind, I don't knwo the details as I ahve never ahd the misfortune to design a PCI device but I suspect there are a lot of bits the would not provide an optimal bus for a processor.
The SATA, USB and FireWire would all be optional extras like SCSI on classic accelerators, a custom connector jacked into the PCI bus wouldallow people like Jens to make their own upgrades as well. If it is done for fun the only developemnt costs that I can see are the chips and other parts then getting the board etched and assembled. If you are lucky/good you get it right first time and don't have to gop through the whole constuction processa gain, that is when it starts to get costly.
A replacment motherboard, well I supose you could drag the custom chipsets out of retiremnet, possibly hire some of the old staff to fisnish the AAA chipset, then you would have to get Freescale to fab new 68060's or design a system using the coldfire and modify the custom chipsets to use that then emulate the nessacery instructions in software mode. But what would be the point, they are so old they no longer have the advantage that they had in their day, a G5 could emulate them and a 68060 better than they can perform.

@gregthecanuck

Haha, I love the idea of stacking boxes to make a tower!!!

Here's what I would do:
One small board with CPU, basic IO, PCI-Express controller and memory, this is the motherboard.
Selectable size backplane cases, for instance single/dual slot 180 degrees (A500 style case with motherboard, graphics and sound card) or 3-8 slots 90 degrees (A4000/A4000T style jobbie, but plugin modules are plugged into the backplane which in turn is plugged into the motherboard).





The boxes, PSU and the backplane would come with the tower, and you could have cabling from each box into the drive bay area in the front (SCSI, SATA and sound cables.

There would be a door on the side of the drive bay area, and drives would be mounted in cassetes which would be removed by pulling them out of the front. Adding a new controller and a drive would mean removing the drive bay side door, putting card and cabling in box, slotting box into backplane, positioning cable so it will reach drive, slotting drive in from the front and connecting it to PSU and controller.

This would be so cool!!!

I'm gonna design something around an existing PC here just to prove the concept, even though a passive PCI-Express backplane would be endlessly cooler. Maybe I can get hold of an SBC and a backplane cheaply and do it with normal PCI/ISA cards. I even have an old Whistle Interjet ISDN-dialer that can be modded into something like this.

Hold that thought!!

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@KimmoK

The new generation of PCI-Ex graphics cards that I would love to see on the Amiga have a transfer speed of 64gbps on a 16x PCI-Ex slot, sudely 4.3gig jsut seems so slow.
Putting the RAM on the card would certianly remove a lot of the trafic from the processor to the RAM, but there is still a lot of other trafic to flood that bus if it is the only way to the processor, also there are no PCI-Ex chipsets for PowerPC, again, we start having to design our own ASIC's

@digitaldisaster

" PCI-X as it stands is simpl not fast enough, it is also not designed with processors in mind, I don't knwo the details as I ahve never ahd the misfortune to design a PCI device but I suspect there are a lot of bits the would not provide an optimal bus for a processor."

But I meant that the CPU would have all it needs on it's own card. PCI-X would only connect to GPU, keyboard, firewire etc. For example BlizzardPPC cards have all the CPU needs on the CPU card. A1200 HW does not stand as the bottlenec of the CPU.

"A replacment motherboard, well I supose you could drag the custom chipsets out of retiremnet, possibly hire some of the old staff to fisnish the AAA chipset, then you would have to get Freescale to fab new 68060's or design a system using the coldfire and modify the custom chipsets to use that then emulate the nessacery instructions in software mode. But what would be the point, they are so old they no longer have the advantage that they had in their day, "

Absolutely right. And that was not what I meant.

I meant with the step C) that you would replace all legacy stuff of the A1200, except perhaps the A1200 keyboard connector. No AGA, no Paula, nothing to stand on the way as the bottlenec.

"a G5 could emulate them and a 68060 better than they can perform."

Right. And with AOS4 you would not need to emulate anything. (uness someone is running 68 apps, identical situation with A1 HW)

CONTINUATION:

"but there is still a lot of other trafic to flood that bus if it is the only way to the processor,"

Care to elaborate?

Because I think:
- the GPU-3DRAM needs the highest bandwidth. (does not affect PCI-X traffic at all)
- CPU-QDR RAM needs the second highest bandwidth (does not affect PCI-X...)
- CPU-CPU (SMP) needs the third highest bandwidth (does not affect PCI-X if there is single multi-CPU card)
- then comes the bandwidth of system RAM - GFX card RAM (this would affect PCI-X)
- the rest is peanuts and insignifficant when compared to any of the above

Unless you can tell something that I've missed.

It will take several years before AmigaSW put any stress to AGP1x or 2x speeds on the "system RAM to GFX card RAM" -bus and as long as there is "enough" GFX card memory, the system RAM - GFX card RAM traffic load stays pretty low.


But ofcourse if we would be building a HW that wipes the floor with everything else .... (not realistic)

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@olegil

Wouldn't it be more realistic to only concentrate on the motherboard/CPU-module, becaus this idea of yours (while nice in some ways) wouldn't be able to use a standard case, no?

@digitaldisaster

Well there is the Articia I and Is (...if they ever materialise)