Hi Bolton, that sounds like a market. We can see it now mounted on your mountain bike...

We think we have to take a more general approach now. Use the Pegasos II as a starting point and then produce a series of reference designs that will push the platform into the market farther, eventually reaching the point were what you want is possible. Of course, you could take the "KRASHAN" approach and devise something totally radical...

If you put the power supply above the back mud flaps, and...

R&B

Quote:
So, I get the impression that there are at least two strategic directions that are being evaluated (and it's not bad, since they are not mutually exclusive): one is aimed at corporate/institutional (in a loose meaning) customers, and here Linux and OSS are the key, combined with low power, high flexibility and the open architecture - if only you can add a lower cost to the equation, you've got a winner, IMHO. But the solution seems actractive nevertheless, even at the actual pricing. This is a businness opportunity worth trying, because it's actually "easier".
The second target could be "Joe user", who is shopping for particular electronic devices: now, as Tronman pointed, there *is* demand for a good laptop and/or handheld device. As well as other devices, like digital television STBs. Here the key is the user interface: you can sell to "Joe user" if the interface is simple yet responsive. Here, IMHO, MOS comes to play - never seen anything more responsive. If you could scale both the HW and the SW (and I think this should be one of the points in having the two developments carried out in house, like Genesi does), you could sell a (small) family of devices to the consumer market, and actually you could have the chance to succeed. But to do this, I'm afraid you need some *big* partner involved.
And you shouldn't underestimate how appealling the MOS/Pegasos couple is to the most "educated" users out there as a desktop computer: the problem is that most people don't know it, because usually the feedback from those who get the opportunity to see it "live" for the first time is very good.

BTW, bbrv, I sent you a mail with a couple of those "thoughts", and "something" else I'd like to have some kind of feedback, and a pm here on MorphZone, too. Since you are keeping an eye on this thread, I hope you'll take a look soonish...

Kind regards,
Andrea

Hi Andrea, we have the email, but have been very busy over the last few days/weeks. We will see what we can do for your presentation, but no promises...

We are close to having either of two very "big" partners, but it won't be announced until we are sure which one it is. One week we think it will go one way, the next week the other.

In the meanwhile, we are busy establishing Pilot Programs, which has kept us quite busy. Also, keep an eye on these STATS, as you may be surprised at what you see happening there soon...

R&B

big partners....i like the sound of that

@ guruman

At least a couple of more devices seems to be planned:

http://www.genesi.lu/olp.php?slide=8

Quote:

(OK, someone mentioned "The Stats" and made me go Off Topic! Not my fault people! )

I have been working for 117 freakin' days to collect my 10,212 blocks, and for what I know, "bplan-gmbh" should pass me tomorrow! In only 21 days!! And I have been running the whole time!!!

And it seems like it has peaked at 1,247 blocks in a single day!
http://stats.distributed.net/participant/phistory.php?project_id=8&id=437432

What is this, a "Big Blue" SuperPeg? And is this really MorphOS running on that cluster?

BTW, where are "bplan-gmbh" on -** THE TEAM **-?
http://stats.distributed.net/team/tmember.php?project_id=8&team=30562

We need you!

@ takemehomegrandma,

I think that was 1200+ over two days...

on topic:

I think one of the most powerful advertisers is Stats.
especially if they are standardized. Hopefully in the next
couple of months we will learn more about this super Mos networking
extravaganza...

@bbrv

I think you are looking in the right direction linux is the present
as for the future... we will see.

katos1

@Tarbos

>I have AthlonXP 2500+, Radeon9800Pro and Dual-Channel DDR400 RAM that should make it run circles around PS2/NGC/Xbox - btw can you play in 2048x1536 with your gameconsole?

You can play games in 3600x2800, 10000x8000 or 100000x80000, but does that make your game any better than on my tv-res PS2/XBox?

Nope, it does not.

If better hardware capabilities wouldn't help to create better games, why will there be Xbox2, a Gamecube follow-up, Playstation 3 and so on?

I saw Morrowind on PC and Xbox and on PC you are able to see more details, have better textures, shorter load times, better possibility to save games, a large number of hotkeys (spells, weapons...) thanks to the keyboard, internet gaming etc. - and no flickering screen.
So yes, I think it makes games better than tv-res PS2/Xbox.
Not to say it's no fun to play on TV, I connected my A1000 to the bigscreen and played Lotus Turbo Challenge 2 with friends, International Karate + etc.

Quote:

Actually the nintendo president is pushing the idea that future consoles will not depend on ever better graphics, I'm inclined to agree, the Cell processor in the PS3 may be powerful enough for real time ray tracing, you can't get any better than that.

The best console graphics are from the XBox yet all it's managed is joint 2nd place (it's sales figures are very close to the gamecube). Sony doesn't have the best graphics but does have 70 million sales.

Quote:

Thats a very good point, what's more they also cost less. Would be an interesting choice for a server as a pair of low end chips will give very good performance, especially for server type applications.

>I saw Morrowind on PC and Xbox and on PC you are able to see more details, have better textures, shorter load times, better possibility to save games, a large number of hotkeys (spells, weapons...) thanks to the keyboard, internet gaming etc. - and no flickering screen.
So yes, I think it makes games better than tv-res PS2/Xbox.


You actually believe that we will have so much better games when XBox2, PS3 or yet more powerfull PC's arrive, than we have now?
Why don't we have the ultimate supa dupa hypa game yet for pc then??? We sure have got the hardware to create one.. have had for a long long time?

Games quality haven't progressed that much in past years. 3d engines have, games have not.

>by bbrv on 2004/6/28 20:50:34

>Also, keep an eye on these STATS, as you may be surprised at what you see happening there soon...

What is the surprise - Bplan stopped crunching RC5?

lol ... I visit the bplan stats every few days
since this announcement ... guess now that amiwest
seems not to be the release of 1.5 I can stop hoping

A watched pot never boils ...

The presentation is ok, though some of the "feature" sheets are a bit cluttered but that lies a bit in the nature of such feature sheets of course Don't really think all those arrows are needed. Also the background with those buildings should be lighter so the sheets are better readable. Other than that it looks fine.

Tarbos wrote...
Quote:

Exactly! Doesn't even matter if MOS will be 64 bit, as long as it works. Those other architectural improvements will speed it up greatly already.

bbrv wrote...
Quote:

Do I understand correctly that you cancelled the G5 peg 3 boards (because of heat or supply problems) then? I thought that the peg 3 was already announced:

on 19 jul 2003 bbrv wrote on morphos-news:
Quote:
And on that last presentation I see "Future versions of PegasosPPC will closely mirror next generation CPU releases". So I hope I was wrong in the above assumption (although Moore's law will be slowed down because of etching problems in the future with SiON and other higher-k materials than SiO2 needed for the gates, so we'll be golden for a while with a G5).

I already saved up for when it comes out, so I really hope the plans are not abandoned IBM has solved their 90 nm manufacturing problems, while the 90 nm G5's run cooler than the 130 nm ones, the difference between 2 lithography generations isn't so dramatic as before.
So perhaps the peg3 could use an (older, cheaper) 130 nm PPC 970 (made in older IBM factories - they probably don't want to throw all those "old" wafersteppers and etching tools away!) which is underclocked to like 1.4 GHz. The architectural advancements alone are quite large over the G4, underclocking would mean you wouldn't need an extreme cooling solution.
Although Apple now wants to move their imac line to G5's which may introduce shortages again

For the people talking about the cell processor which will be used in the PS3, this could be interesting in the (far) future of Genesi as well...

Indeed one chip (and they are scaleable) can "theoretically" crank out matrix calculations 20 times faster than the GeforceFX 5800/5900/5950 (NV30/35..). 1 Tflop = 1000 Gflops for 1 cell chip vs 51 for NV30/35 vs 8 for Itanium2. That's huge! Oh and you don't need a graphics card anymore because of that, so no more driver nightmares either (though writing the 3D driver for the architecture may be quite some work). It's also much more flexible since they are not limited to graphics only and you can completely change the code paths.

Architecturally it basically consists of 4 simple powerpc cores with many coprocessors (128, but this is variable depending on the precise implementation). These coprocessors have NO cache thus making their silicon footprint very small. Of course this means all pipes inside (between coprocessors and processors etc.) and to the processor (certain rambus high bandwidth things are licensed in PS3) need to be really fast or the design not to implement cache on them would be fatal to performance. If done well, streams (like... all multimedia, especially 3D matrix calculations) can be processed really really fast. If one of the next generation pegasos (or the set top boxes) needs to stay competitive perhaps it is a good idea to licence these cell chips in the future plans. As I understood IBM/Sony may sell these for "PC boards" so perhaps they will licence them as well. Since powerpc chips are already used in the pegasos, the OS would be easy to port as well. It would be a golden opportunity to outdo the PC industry imo (that is, depending on the terms of Sony/IBM of course for the licensing). Oh and of course a small kernel with tasks like networking outside of the kernel (like I understood MOS has, but unlike linux) will be better suited for multiple procs than the case with lots of stuff in the kernel, so I'm sure MOS would run great on a cell chip! (Usually people put a lot of stuff in the kernel to avoid taskswitching overhead between different processes, but putting less stuff in the kernel can make it easier letting OS tasks run on different processors since they run in userspace and the kernel only does (async) messenging). Oh and above all, these Cell's will be "very cheap", especially for their performance. Pegasos with one (or more, since they are scaleable!) cell processor in it would be a dream machine reminding of the revolutionary things the Amiga did back in the day!
For more info see: http://www.ps3land.com/CELL.ppt
http://arstechnica.com/news/posts/1084391000.html
http://www.ps3insider.com/modules.php?name=Content&pa=showpage&pid=3
I hope that will become part of the future strategy of the Peg

Azalin - AthlonFX53 2.6 GHz "4000+" 1 Gb SuSE 64 bits UAEAmiga!


[ Edited by azalin on 2004/7/18 20:47 ]

Quote:

You don't need a graphics driver as such as the Cell would be doing everything in software, you just need to port Mesa to the Cell - that however may be a *lot* of work.

Quote:

It's not quite clear if they are PowerPC, nothing has been announced but given IBM are doing part of the design work I'd guess its fairly likely.

Quote:

Doubt it, they're probably pretty limited and the lack of cache would cripple any OS or general purpose app. Cells will be very good at DSP like taks but awful on other tasks - if you can get them to run at all. But that said theres a lot of DSP like tasks out there (video, audio, 3D etc.).

Quote:

You're forgetting - MorphOS is not (currently) Mutliprocessor.

Quote:

Have you been reading OSNews
The whole series is about a "dream machine" and a dream OS on top.
Which reminds me I have to finish off Part 3...

Quote:
Interesting links, the PPT in particular crunches the Sony patent down to something more readable.

Quote:
Yep exactly: you would only have to implement a 3D API for 1 "architecture" which probably stays more or less the same in future generations (i.e. more cells get added, but the instruction set stays more or less the same), instead of writing drivers for changing architectures from several competing companies (ATI, NVidia) which are likely very different. The big advantage would be that since it is basically a "general purpose" processor(s) with DSP like stuff on it, the information on the instruction set is probably easier (or cheaper) to get than information needed to write drivers for Nvidia and ATI products, who keep the information on their chips (and especially drivers/algorithms) very closely guarded. (For linux they offer mostly closed source drivers, though for some older stuff this isn't so.) Of course this does not necessarily have to be so and Sony may be as secretive as Nvidia/ATI. It's just an assumption on my part. Indeed porting (or rather... hand coding in assembler) something like MESA/GL, especially one that runs efficiently on cacheless coprocessors, would be a lot of hard work.

Hmm in that .ppt presentation I linked (which opens fine with openoffice btw ) they said it was a PowerPC core in one of the pictures, and the processors did have a cache (probably not much though) but the coprocessors did not. Then again I don't know exactly where those people got their info from (thought from the patents filed, and there "PowerPC" could have just been an example). These powerpc cores probably would underperform quite badly because they are nowhere near as advanced than "normal" processors (otherwise they would just be as big in silicon footprint). So I think you are indeed right, if a cell processor is used there might also have to be a "normal" processor to handle the usual jobs (but I wonder how you would do stuff like the memory buses then - that must be a nightmare to design - or give the general purpose and the CELL chips seperate memory of course, but then you'd just have a video card). And true MOS is not multiprocessor .

And yeah nice article on OSNews, good job! Didn't know you wrote those lol. Glad you like the link, the patent itself was quite hard to read indeed.

p.s. Sorry for being a bit off-topic, I hope this part is still interesting


[ Edited by azalin on 2004/7/20 14:31 ]

A Cell subsystem could probably deliver kickarse performance when connected to the system by PCI-Express 16X with a giant up _and_ downstream bandwidth.

>Indeed porting (or rather... hand coding in assembler) something like MESA/GL, especially one that runs efficiently on cacheless coprocessors, would be a lot of hard work.

How much cache do the gfxchips have?

It doesn't specify how much cache exactly. The APU have 128 kb of sram storage though and there are 128 registers of 128 bits so plenty of registers to play with.

Quote:
Memory interface is very wide (many banks) and gets 50 to 100 Gb/s, that's insanely high. As opposed to 2.1 GBps, for the current 133 MHz FSB DDR memory bus, 6.4 for a memory bus like used in the PPC970 and 4 Gb/s for a bus as the PCI-express 16x you mentioned. So the bandwidth of a shared memory architecture would definitely not be enough.

Some more information on the memory interface is here: memory and bus interfaces. These are (theoretically) about an order of magnitude faster than current implementation of buses and memory interfaces, that's also the reason the processor can do with less cache.

If this would be integrated in a Peg, the main processor and the cell chip can not have the same memory. The "graphics card" with cell chip on it, should have its own memory on the card.
p.s. Perhaps we should open a new thread for this part of the discussion
p.p.s. running MOS on a PS3 would be awesome, they are looking for an OS I think

Addition:
It seems the biggest benefit of the PS3 architecture is that they use the high speed interconnects and memory interface licenced from Rambus, thereby are less dependent on caching so they can increase logic on the chips instead. These interfaces are linked above. However, since a graphics card can also use this kind of interface, Nvidia and ATi will be able to use the same techniques to offer an order of magnitude speed increase because of these extreme high bandwidth memory increases. Since these GPU's can be parallellised extremely easily, their future cards will probably rival the PS3's projected performance as well. And of course these cards might fit in a Pegasos as well (that is, if they still make AGP versions by the time those cards come out). So no cell chip needed at all. Infact a recent article I just found (in german) confirms my suspicion. So let's just hope ATi and Nvidia will give enough info to Genesi for writing drivers for their newest hardware and maybe these cards can be used on the peg. Note that this type of memory could be used in future iterations of the Pegasos as well for main memory. It will come out next year at its earliest but it is not known how expensive it is. Basically the memory runs at 800 MHz but there can also be 8 (!) signals per clock as opposed to 2 for DDR.

(edit: removed some confusing lines from my post to make it more clear.)
(edit: added some more on the memory interface.)


[ Edited by azalin on 2004/7/25 16:45 ]

Quote:

First one to get more info on cell, post it here. I'm particularly interested in what type of DSP like work it can do.

Any news about all this?