I know that both the MorphOS community and the AmigaOS4 community haven't expressed interest in a PS3 port (primarily because the PS3's hypervisor prevents access to 3D hardware), but what if IBM were to change its policy of limiting access to Cell processors and allowed outside developers to work with the chip?
At 3.2 GHz with its additional 8 SPE processors the Cell could make a powerful core for a multimedia computer. One of its few limitations (besides its in-order execution) is the limited amount of XDR ram it can address. However, considering that the Efika runs fairly well with less memory a tight, compact OS (like MorphOS) would be well suited to it.
The main reason I brought this up is I've inquired directly with IBM, and they may not be as adverse to the idea of outside development as was once thought (but don't get your hopes up just yet).

Do tell, so IBM is responsible for the hypervisor in PS3? I wasn't aware other cell products were crippled too.

Tell us what (if anything) you hear back from IBM. What is the addressable memory limitation in the PS3 (what is the installed memory amount in the PS3)?

Playstation 3 has the Cell supporting 256MB of XDR and 256MB of graphics memory (which Linux maps as swap space).

With innovations like the compressed cache/swap modules (http://code.google.com/p/compcache) you could run a reasonable system in 256MB (let's assume you didn't have as capable a graphics processor as nVidia's). It already has more than the Efika :)

Most of the reason they use so "little" XDR is because XDR costs an absolute fortune, compared to GDDR3 (as used on the nVidia) or DDR3 memory.

IBM actually warned people off Cell development within Power.org since they had other plans for the development. They probably have that all worked out now, and would be happy for it to go to other places.

The basic problem then, as always, is that the Cell is a very powerful chip and therefore NOT the nicest on power consumption - or fan noise. That could be solved by reducing the clock speed. This would also bring the price down somewhat.

It would probably be a fairly reasonable idea to make a system based on Cell, using a chip sorted for less than 8 SPEs (maybe 6, this gives a system which has exactly the number of SPEs as any PS3 game is allowed to use). Any apps designed to work with Toshiba SpursEngine would be optimized for performance on only 4 SPEs - so maybe 4 is enough. Either way this would also bring the price down.

It would also be reasonable to keep the 256MB of XDR RAM and supplement it some other way. They sell and people DIY their own PS3 clusters where 256MB seems to be more than enough RAM. Again.. price stays low.

What might bite you is selling a "desktop" system which met the specs of the Playstation 3 (albeit without the nVidia graphics or Blu-Ray drive..). I doubt Sony would be too pleased to lose their DIY cluster market.. the PS3 being a "supercomputer" is one of their marketing points. Without the Blu-Ray drive and the graphics controller and all the console shenanigans it might even be the same price as a PS3..

What might also bite you is the system controllers available which allow attaching PC-standard peripherals (PCI Express etc. for graphics). You're also probably not going to get the PowerXCell 8i processor at an affordable price, so the original Cell is what you're looking at. All in all though, great idea, we'd love to do something with it/about it..



[ Edited by Neko on 2009/1/29 2:04 ]

@neko

are you telling me know that Genesi never considered the Cell to use in their products ?

And would it be a problem to place the sell on a megarray? So people can actually choose themselfs if they want a "green" machine or something more powerfull ?

[ Edited by Oepabakkes on 2009/1/29 12:04 ]

> both the MorphOS community and the AmigaOS4 community haven't
> expressed interest in a PS3 port

I wouldn't say so. There's a great part of the OS4 community that actually wants OS4 on PS3, leading to excesses like that. And there are also MorphOS users who want their OS on PS3.
It's just that the *developers* of OS4 or MorphOS respectively stated a non-interest in a version for PS3.

> what if IBM were to change its policy of limiting access to Cell
> processors and allowed outside developers to work with the chip?

IBM does *not have* such policy. Access to the Cell *is not* limited in any way. Outside developers are not only allowed to work with Cell, IBM are actually encouraging developers to work with Cell.

> One of its few limitations [...] is the limited amount of XDR ram
> it can address.

What's that limitation with the original Cell exactly? AFAIK the PowerXCell 8i can address up to 32 GiB DDR2-SDRAM.

> IBM [...] may not be as adverse to the idea of outside development
> as was once thought

The opposite seems more likely.

Quote:

http://www.powerdeveloper.org/forums/viewtopic.php?t=1516

> http://www.powerdeveloper.org/forums/viewtopic.php?t=1516

The SpursEngine is not a Cell as it lacks the PPE. So you cannot run an OS on it.

As of a few days ago I'd received this message:

"James,

While I understand your frustration, IBM does not currently offer these products for sale as stand-alone chips, and only offers them in the board level formfactors that I have mentioned previously.

Regards,

Rob Gibson
Business Development Executive
IBM Systems and Technology Group"

That was the end of a long exchange with Mr. Gibson over the possibility of purchasing Cell processors. At that point it seemed IBM was not interested in outside hardware development with cell processor. Yes they do encourage cell software development and they even provide a SDK. But you must run that on a PS3, an IBM cell based blade system, or a computer equipped with a cell based PCI-e card (from Mercury Systems or other vendors). At this point, hardware information on the cell processor is slim. Matt's right in stating that IBM has in some ways discouraged cell development (at least from a hardware standpoint).
So, why do I think we still might see IBM loosen up this policy? I contacted the President of the company and currently they have some of the technical staff (including at least one engineer on the cell team) looking at the idea. Hopefully we can convince them to allow a broader use of the cell.
BTW - I can see why Genesi hasn't looked at this yet. IBM hasn't been very forthcoming with technical data on the cell. Plus, even if we can manage to get them to allow this kind of development, its going to require a great deal of complicated design work. Unlike the 8641D from Freescale there's no ready evaluation platform (at least not one that's well documented). At this point I know of only one accessory chip (and Toshiba hasn't given me any information on that). A southbridge would not be a problem, but a northbridge would have to be specifically designed for this processor (I don't think there are existing designs that would work).
So, at this point, everything is more than a little "iffy". I probably shouldn't have even posted this, but the possibilities have me excited. Instead of a 1.5GHz G4 level machine we could have a 3.2GHz PPC bassed machine (with up to 8 SPE based co-processors). These co-processors would be ideal for multimedia streams. This, to me, looks at lot like how a modern day Amiga would be designed (with the additional processors freeing up the main processor - kinda sound like a Amiga). Anyway, I'll keep you posted.

[ Edited by Jim on 2009/1/30 2:18 ]

[ Edited by Jim on 2009/1/30 2:20 ]

Actually, while reviewing what technical information I have been able to get it looks like the cell has a rather unique bus (FlexIO) which will make the connection of companion chips considerably more complicated.

[ Edited by Jim on 2009/1/30 2:16 ]

> At that point it seemed IBM was not interested in outside hardware
> development with cell processor. Yes the do encourage cell software
> development and they even provide a SDK. But you must run that on a
> PS3, an IBM cell based blade system, or a computer equipped with a
> cell based PCI-e card (from Mercury Systems or other vendors).

Apparently, there have been some misunderstandings from my side. I thought "limiting access to Cell" to be meant programming-wise while you meant it distribution-wise and "allowed outside developers to work with the chip" to be meant software-wise while you meant it hardware-wise. Sorry for that. Golem fell for the same misunderstandings it seems.
But thanks for clarifying the actual situation around Cell and IBM's current -- and sad -- attitude. I also hope there will be change ahead.

Quote:

WHAAAAT? Jim, who are you?

Quote:

Now, that could be truly incredible: A bunch of geeks convincing the biggest computer company to change their mind, in exchange for nothing relevant, business wise.

Quote:

I doubt Genesi overlooks anything. Their main activity is business relations...

Quote:

Yes, they don't give away documentation and chips to computer hippies like us. But I'm sure that Sony, Toshiba and Mercury did get some attention from IBM. Size matters, simple as that.

Quote:

Excuse me, but I'm completely shocked.

Quote:

Well, you can buy a PlayStation3, it's for sure the cheapest development platform ever. Have you seen the figures for a development board for the measly MPC5121e? By the way, I can't find the one for the MPC8610, but sure it's more expensive.

What worries me most is that nobody is talking about an important matter here. The Cell processor is extremely complicated to program for. Essentially, it's what Kutaragi (Sony) asked IBM to do for their revolutionary PlayStation 3: A supercomputer on a chip. Very suited for lots of self-contained, low data volume tasks, but dare to write an operating system for it, that really takes advantage of all the processing power.

Quote:

Well, if you wanted us to drool, granted you've made it.

Quote:

I find IBM's Xenon chip (triple G5 core at 3200 Mhz!) more interesting, but it's even more restricted than Cell. We are talking about chips made specifically for certain customers, not with the aim of selling them to anyone, as manufacturers do with other discrete components. These pieces of technology carry as much business ties as technology inside.

I know what you mean about the Xenon. It is more impressive than the Cell, but Microsoft is never going to let us use that one. As far as contacting the President of IBM, just go to their website - they have his e-mail address posted there.
I also understand how difficult it can be dealing with Freescale. They want $4000 for the MPC8641D development board, and according to Maureen Lyman at Freescale the processor isn't even considered a production item yet (they have none in stock and none to sample). How many years ago did Genesi look at that platform? I think it was around 2006 when I saw the first postings about it. Plus, I'm beginning to doubt that Freescale really does intend to build an e700 core.
So, what have we got to look forward to in terms of Power PC processor advancements? Well there is always the long rumored IBM Power7 architecture. And there is AMCC's Titan core (while its only going to operate at 2GHz its a neat low power multi-core design). And then there are the Cell (and Cell derived processor like the PowerXCell 8i). Frankly, at this point I'm leaning toward that design.
After all, think about it, Bill over at Amiga Inc. has released specs on two designs that probably won't see the light of day. The high end design relies on a PA Semi processor that may not be available in the future and the low end design relies (once again) on a relatively slow Freescale Soc. We had better equipment from Genesi several years ago.
If we want competitive hardware we need to look at several upgrades (to our current components). First, we need more powerful processors (and at some point we need to consider multiple cores). We need modern expansion busses (like PCI-e) and accessories (like the upcoming USB 3.0 port standard). And of course one of the most important things is we need good operating systems and software.
I know I may push these ideas a little harder than most would, but I really believe we can attain these goals.

BTW - While it's mainly designed for video manipulation, look at what Sony intends to release soon.

http://pro.sony.com/bbsccms/ext/ZEGO/files/BCU-100_Whitepaper.pdf

Four times more XDR than in the PS3, 1G of DDR2, no restrictions on accessing the RSX, and a Southbridge that allows for the real time manipulation of video and audio streams.
Now THAT makes my mouth water. How'd you like to port Morphos or Amiga OS4 to that?

Take care
Jim

[ Edited by Jim on 2009/1/31 2:16 ]

[ Edited by Jim on 2009/1/31 2:18 ]

Quote:
If you want competitive hardware you need to go x86/pc road... Market drives innovation, not the other way around. PPC will always lag behind, in terms of speed, and expansion, simply because it's targeted at different markets.
And yes, Cell&Xenon are powerfull PPC variants, but closed ones and you can't expect IBM/Sony to open it up for you as they have nothing at all to win...

> according to Maureen Lyman at Freescale the processor isn't even
> considered a production item yet (they have none in stock and none
> to sample).

Interesting, as the CPU is considered "active" there.

> I'm beginning to doubt that Freescale really does intend to build
> an e700 core.

Really just beginning? Look here.

> what have we got to look forward to in terms of Power PC processor
> advancements? Well there is always the long rumored IBM Power6
> architecture.

No need to look forward to that, as it got released in June 2007(!) already. Maybe you meant POWER7, announced to be released in 2010.

Andreas, try to obtain an 8641D. They may list them, but they are not available.
And Leo is right. The STI partners aren't going to do anything that costs them a profit.
Plus, the design of a Cell based board is VERY complicated. A good example of this is the PDF published by IBM and Mercury engineers on the development of the Cell based PCI-e board.

http://www.sigrity.com/papers/2007/5-WP1GregEdlund.pdf

However, I'd rather not spend all my time working with slow G4 based Soc designs.
Freescale's e700 appears dead, AMCC's Titan isn't on schedule, and IBM still looks like the best bet for up to date designs.
Hey, it can't hurt to ask.
Jim

[ Edited by Jim on 2009/1/31 2:31 ]

[ Edited by Jim on 2009/1/31 2:34 ]

> try to obtain an 8641D. They may list them, but they are not
> available.

It's not that I don't believe you. It's rather I'm astonished they list a nonavailable CPU as "active". And what about the many embedded system providers listing MPC8641(D) (and/or MPC8640(D)) based products as available?

First, I received this:
"Jim -

Thank you we are not able to give you the MC8641VU1500KC we have no product this devices was just introduced. I am still trying to work on getting the other part number on order for you.

I will advise once that is done!!!"

Then, received this:
"Jim -

I am so very sorry I cannot order these samples though my internal system. I think once these devices are in a production mode samples will be available via our web site. I am going to send an email to the on-line sample team.

Please see note below:


------ --------- ---------- --------

** SAMPLE ORDERS NOT ALLOWED - CONSULT FACTORY **

23OCT08 EACH 1 - PCQ

EACH 1 - MPQ

BOX 120 - POQ"

The first message was a reference to the 8641, the second to the 8641D. While I'm sure that large purchases of these chips may be possible, it doesn't appear that they are considered to be in "production mode".
Of course, you might think that my request for the highest speed versions of these chips might have something to do with this response, but I specifically stated I could use slower versions.
I don't know Andreas, Freescale just doesn't seem to able to develop products as quickly as Motorola used to.
Jim

Jim, thanks for all this informations, I guess I should swallow some of the words I wote to you.
Could you please tell us something about you? Are you an engineer?

In the mid 80's until the early 90's I was the retail manager for Delmar Co.. We designed and sold Motorola 68000 based computers that used an ISA expansion bus. Our primary OS of choice was Microware's OS-9, although some of our customers used FLEX, Minix, and later Linux. Our first offering was called the System IV. Later designs (using the 68030) were called SystemVs. When Microware refocused on process control and embedded applications, I went back to school to learn more about electronic design.
Today, I do some consultation and I keep abreast with current developments, but I've come to the conclusion that it would be very hard (considering the frequencies that today's components operate at) for a small company with limited resources to design and build a modern motherboard.
That's one of the reasons I've been so impressed with bplan and Genesi. They succeeded where Amiga Inc didn't (I probably won't with their new proposals).
Engineer? No, more of a Systems Analyst. But I have worked directly with hardware and software designers and I'd like to think I have a good idea of what can be done by a small group of dedicated people.
If there was enough interest, we (as a community) might be able to get a small run of boards made, but right now I'm not sure Matt isn't right. Our market is probably too small to make a profit. If you think the SAM440EP is too high priced, well its probably a reflection on the size of that market.
Could it be done? Yes, but its going to more of a labor of love than a true commercial enterprise.
Anyway, lets see what happens.
Jim

OK, this may be my last posting on this topic. I've had a couple of weeks to review what information I could get on existing Cell applications. One thing really becomes apparent. This processor is HARD to implement. First, the use of XDR memory makes designing the memory circuits fairly difficult. If you review the PDF I've referenced before on the IBM/Mercury designed Cell based PCI-e board you'll note that when the memory circuits of that board were designed, Rambus was contracted to assist in the development. XDRs archetecture and the skew on the XDR memory bus increases the difficulty of implementing this type of memory. Its worth noting that while the Cell could potentially address 32 to 64 GB of memory, the largest current designs are limited to 1GB.
This of course does not take into account the PowerXCell 8i use of DDR2 memory, but that processor is far out of our reach economically.
Second, because of the complexity inherent in the design of Cell based boards, it does not seem possible to implement a Cell design with a simple four layer motherboard. Instead, most designs would likely be eight or more layers. This would greatly increase the complexity and cost of the design.
Further, as the Cell uses an expansion bus called FlexIO there are few options for interfacing peripheral circuits. Currently, the only southbridge available is a chip that Tosihiba and Sony refer to as the "Super Companion Chip". This chip is used on Toshiba's Cell Reference Set (and the Cell Reference Set 2) and on Sony's BCU-100 multimedia rack system. It may also be related to internal components in the PS3. Information related to this chip is simply not available.
When IBM, Mecury, and Fixstar have designed Cell related products they have designed their own glue circuitry to accompany the Cell. So, we have a processor that requires custom designed interface circuits (rather than the "off the shelf" components we've seen in the past).
None of this makes implementing a Cell design impossible, it just makes it much more difficult. It also may explain IBM reluctance to offer this product to the general public.
The last response I've received from IBM tends to reflect this. Ray Bryant is IBM's Director of Games and Power Solutions. Part of the message I received from him states these facts.

"a) There are no STI restrictions on IBM's right to sell BE chips for applications outside game consoles

b) IBM has not invested in the engineering work and documentation to support Cell BE as a standard, off-the-shelf component product that would be available for sale to anyone. That decision was based on the limited interest we saw in the market for high volume sales of this highly specialized multi-processor design.

c) IBM has supported usage of Cell BE chips in several custom board level applications for OEM clients, typically as part of a broader relationship between IBM and those clients beyond Cell chip sales

d) IBM continues to review specific client requests for access to Cell BE chips as loose components and makes case by case decisions based on the level of engineering support needed to assist the client in Cell based system design, the size of the business opportunity for IBM, and the strategic benefits to IBM of the project. We currently have several active client engagements where IBM has quoted sales of loose Cell BE chips for projects that met IBM's strategic business goals."

So, where does this leave us? IBM examines and qualifies each potential application for the Cell. While we could present a business case to IBM, they might reject the proposal if the application didn't have the potential for relatively high sales.
At his point, I don't think we can offer that opportunity to IBM so I am (at least for the time being) not going to pursue this further.
Jim

@ Jim

As you yourself pointed out, MorphOS 2.2 does not support multi-processing so the Cell processor is not very attractive to begin with.

I think the reason why a port to the PlayStation 3 has been discussed numerous times on MorphZone is that there are 20+ million owners out there who could potentially run MorphOS on hardware they already own. This would obviously not be true for custom Cell-based hardware.

> MorphOS 2.2 does not support multi-processing so the Cell processor
> is not very attractive to begin with.

Are you sure this is not only true for *symmetric* multi-processing, but also for *asymmetric* multi-processing (by using the SPEs)? Or do you refer to the 2-way SMT capable PPE rather than the SPEs?
The Friedens for example -- long before denying the reasonableness of OS4 port to PS3 -- once claimed that it would be technically possible to make use of the SPEs from within OS4 and OS4 programs and that they already thought out some rough technical concept on how that might be achievable. And AFAIK OS4 is actually exposed to the very same restrictions as MorphOS regarding multi-processing.

Actually the SPEs are not that similar to the PPE core. Their primary purpose is to process streams of calculations (greatly raising the processor computational abilities). I think its probably fairer to consider this to be a processor with eight tightly bound coprocessors on the same die (rather than a true multi-core processor). There really is nothing to prevent the SPEs from being used under Morphos while the PPE provides the bulk of the power to run the OS. As the PPE is the only true Power PC like element in this product (and there is only one on the Cell die), this would be where Morphos code would run anyway. The SPEs would just provide a great way to farm out concurrent calculations (they would make Morph a floating point beast).
Jim

> Actually the SPEs are not that similar to the PPE core .[...] its
> probably fairer to consider this to be a processor with eight
> tightly bound coprocessors on the same die (rather than a true
> multi-core processor).

Yes, I already know that obviously, that's why I considered the use of the SPEs to fall under ASMP, not SMP, in my answer to ASiegel.

> There really is nothing to prevent the SPEs from being used under Morphos

That's exactly what I implied by my questions and example to ASiegel.