Well however they spin it, bottom line is even Sony's internal people can't really make up their minds on what the PS3 is supposed to be. I can imagine it being the "centrepiece" of your home entertainment but your PC?

If network capabilities were all that defined a PC I'd agree with you - but Sony this year has brought on a lot of bad publicity simply by not being able to keep their mouths shut. If theirs truly is a better console - just shut up and deliver the goods, that's all we're saying.

Let us not forget that PS3 does compute and that Cell is a pretty good looking beast. So, technically speaking PS3 is a "computer".

But what PS3 is not, although Sony exec disagrees, is as a credible PC contender. Point 1 being that linux has never truly taken over as a Windows replacement. Also, the PS3 looks too flimsy as an office desktop replacement and is certainly not mobile enough to be carried around.

I do however think that the PS3 would make a nice entry level linux server and as Cell's linux development platform. But they have to contend with millions of obsolete PC hardwares running free open-source linux.

If history were to taught us something it is that developers dislike change and love compatibility. Itanium didn't take off because people are comfortable with the original Intel Architecture and wants their legacy codes to just work. AMD64 fills that need in a huge way, prompting Intel to rethinks its strategy, bringing forward EM64T and the compatible XEON platform. Sony should learn a few thing from Intel's blunder.

Code shops are already busy porting PPC codes to support Mac's Intel platform natively, I sincerely doubt that they would welcome Sony dubious forray into general purpose computing.

This post has been edited by ray_: Jun 8 2006, 03:17 PM

Thanks for reminding me of that. Yes, the PS3 DOES "compute".

I agree with your points - trying to take on not only the home entertainment market but also desktop computing? No matter how many Linux geeks are salivating over the possibility of a Cell-server (and I'm interested to see how it goes) point is Sony is widening the gap between their installed fanbase with every little news bite that comes out, and that can't be good.

So right now I'm wondering: what the hell do I get for my USD 600? A games console, a home entertainment hub that plays bluray, a PC, or all of them? And why shouldn't just get a PC that's ultimately modular?

Games consoles should play games. Everything else is just dressing.

It's interesting how Sony exec wants to associate the PS3 as a PC rather than a console when all the analyst agreed that PC market has pete-out and matured and has very little growth margin; with brutal competition on various fronts (chips, OEMs, graphics). Whereas the game market looks like the next tech bubble yet to burst.

Yes, Game consoles should be playing games only I think they can add features like online but playing movies erm no thanks. Others chuck inside making it more expensive only........

Itanium has got to be one of the biggest blunders in tech history. Who comes out with a shoddy chip with no software support. Sheesh. And it was more a replacement for the HP PA-RISC chip rather than replacing the Intel Architecture.

Cell processor is seriously lacking on branching logic, its a high speed DSP like processor which can operate on SIMD (video/audio) very effeciently but not on game logic. It has only 1 general purpose processor and hence it made a lousy 'computer'.

You can just hack up the old Xbox to install the linux and use it as a computer, there is nothing new here.

By the way anyone see the article here? http://www.theinquirer.net/?article=32171
Things look real bad for PS3 as what i'm suspected the main bus is going to be very very conjested and slow down the whole console for sprites and objects intensive titiles.

This post has been edited by chtan: Jun 13 2006, 02:09 PM

It isn't clear enough on the article whether Cell access (r/w) to the main memory or local memory on the chart relates to the PPE or the SPE. I was rather suprised after reading the embedded.com article on programming the Cell. (Here)

I've always thought that the PPE or RSX would initiate the transfer of data/code to the SPE via DMA since PPE/RSX would know how best to distribute SIMDs' processing load. But it seems that SPE has a mind of its own (or at least a fixed and predetermined task), fetching code/data at its own wimp via the DMA.

Let's assume that "Cell" here refers to PPE access to local store, and assuming that PPE has visibility to SPE's DMAs, I would boldly hypothesize that the abysmal read access figure (16MB/s) makes perfect sense. As there is really no need for PPE to read from the local store, since computation results would ideally be pushed by the SPE via the DMA to external memory.

I sincerely doubt that IBM would screw themselves up like this. They are probably just toying with the minds of doomsayer and folks who had too much coffee and are severely jet-lagged.

This post has been edited by ray_: Jun 15 2006, 02:51 AM

Hehe i think everyone (inquirer too as usual LOL) missed the fact that the 16MB/s refers to the Cell read bandwidth from RSX local memory (that 256MB of GDDR3). It could be a bug, but i would not be too surprised if it wasn't a bug either. I think in the overwhelming majority of cases, Cell would be sending info to RSX and not the other way around. The write bandwidth is 4gb/s, which looks fine to me.

Nice try inquirer. LOL.

edit: after actually reading the article, i'm even more certain it's not a bug:

makes a lot of sense to me... GPUs are very bandwidth hungry and the RSX with its 128bit memory bus is almost certainly bandwidth starved, so you wouldn't want to interrupt it by reading from the local memory anyway.

This post has been edited by ikanayam: Jun 15 2006, 08:01 AM

Ah...that made more sense.

You've pretty much nailed it. I don't see a need for Cell to read texture or vector information that were sent to the RSX by Cell in the first place. If Cell really do need these information, it would (at least a discerning programmer would) have placed it into external memory to be worked on later. There should be a few odd cases where Cell does need to read from RSX (status update?), that would be accomplished by having the RSX push data into the external memory and asserting an IRQ (you could poll this as well) informing Cell that the data is served.

This post has been edited by ray_: Jun 15 2006, 10:20 AM

yup, flow of data purposely like that, PS3 is a game machine anyway, anyone know how x360 did with theirs?

Xbox360's memory is unified, so no such limitation applies.

Also the point about developer's lukewarm reception toward the PS3 is expected. XBOX 360 employs symmetrical MP while the PS3 is asymmetrical. It takes more to write optimized code for the PS3 than the XBOX 360.

I'm feeling a sense of deja-vu. I've heard this before when PS2 emotion engine were initially launched. Developer were dissing out complaints about how hard it is to program for PS2. I remember someone commented on this misperception of the emotion engine complexity. And his comment was that these sloppy developer needs to do more reading and understand the architecture before slamming those semicolons and hashes on their keyboard.

Personally, I think Cell is a beautiful beast, it might be hard to program compared to symmetrical MP system, but nothing is really complex after you've put your mind to it. My advice to budding Cell developer is.... read. Give it 1 year, if the PS3 takes off at all, we would be in for some serious eye candy.

This post has been edited by ray_: Jun 15 2006, 11:47 AM

Well, even the great CRAMAK has said that the current console CPUs are not really that good. Theoretical figures look nice but IIRC on game code he said that an x86 OOO machine would probably do better. Right now both the consoles have many measly threads.

This post has been edited by ikanayam: Jun 15 2006, 11:50 AM

Don't forget the lesson about over-complex Saturn architecture that spelled its doom.
Here is another great detail about cell processor:-
http://www.blachford.info/computer/Cell/Cell0_v2.html

This post has been edited by chtan: Jun 15 2006, 11:54 AM

Cramak's point of reference is the PC. Console would never replace the PC when it comes to raw performance, no matter how much Sony exec spins it. PC would still be on the cutting edge of gaming.

This post has been edited by ray_: Jun 15 2006, 11:57 AM

The epitaph of the PS3 will not read "RIP: Overly complex architecture FTW", trust me. Publishing houses will root for the "Cool" kid in town. The epitapth will read "RIP: Overly uncompetitive price point FTW".

Yeah, but what he's saying is they could have just gone intel/AMD dual core or something like that, and it would have probably been faster and easier to code for too.

Cell and Xenon are streaming processors. Game code isn't like 48 mpeg2 streams lol.

I think the reason for this is very much cost related. Intel/AMD wouldn't want to give a premium so that the console manufacturer could sell their console at a loss and regains it through the sale of games which Intel/AMD will not profit from. Plunging into the console market would severely strain Intel/AMD's inventory, which could be sold at a higher profit margin in traditional PC market.

And their embedded solution (Celeron, embedded IA) aren't competitive enough in terms of performance and power consumption compared to the likes of Cell and Xenon.

This post has been edited by ray_: Jun 15 2006, 01:22 PM

Well IBM/nvidia/ati isn't selling at a loss. It's microsoft that's selling at a loss. And Cell and Xenon aren't exactly power efficient. Xenon consumes 85W in its current form, and Cell probably consumes more. A Yonah/Conroe based cpu would probably kill them in terms of performance/watt. If you're talking about manufacturing capacity, intel has plenty. That's one of the reasons why Apple switched (in addition to the performance/watt)

It may be that IBM is the only one that is willing to customize their chips that much.

This post has been edited by ikanayam: Jun 15 2006, 01:35 PM