Will the cell take over the world????

Yeah yeah....cell this this cell that


CAN IT OVERCLOCK?

Lol...AMD will probably bring out a CPU called "Trunks" :p
 
I love how EVERYONE in this thread speculates on the next consoles and the cell cpu... wait 2 months till PS3 and the next nintendo consoles to be shown. While I don't know what form the Revolution (Nintendos Next Console) will be shown in (video or actual unit running), Sony has guarenteed that at E3 the PS3 will be there, no videos but that actual unit. Prior to E3 they will unviel it, so untill then keep speculating whats more powerful and how much this and that will suck because untill then we are in the dark.
 
The Cell is a promising new architectural twist, but it faces a problem similar to Intel's Itanium. A lot of the optimization work will have to be done by hand until the compilers mature. If the software isn't written properly to take advantage of the incredible amount of potential SIMD throughput that the Cell has on paper it won't matter one whit how good it sounds in theory.

I'm looking forward to seeing what happens.
 
lets face it guys the ps3 isnt gonna be a friggin super computer if it was then theres no way anyone would be able to afford it
 
cell_491 said:
lets face it guys the ps3 isnt gonna be a friggin super computer if it was then theres no way anyone would be able to afford it

Razor and razor blades, that's how the game is played. They make money on the games. Remember how XBox made Microsoft lose 100$ on every console sold?

Besides, PS3 will just be a marketing gimmick as far as Toshiba and IBM are concerned. Get the name out on the street, have people talk about it, then boast it in your own products.

And Sony is willing to take a hit as long as PS3 is opening the doors to Cell and Blu-Ray. Now if it creams the XBox2 (which remains to be seen, Microsoft is a formidable opponent) on the way it's just gravy.


I'm not saying that Cell will have the power of a top 10 supercomputer, but it might get pretty close.
Edit: in terms of ratio, not 1 cell beating a cluster ;)
 
cell_491 said:
lets face it guys the ps3 isnt gonna be a friggin super computer if it was then theres no way anyone would be able to afford it
Cheap supercomputing exists.

The thing you have to remember is that most desktop CPU's are designed to be a "jack of all trades" - they're designed to process almost anything you throw at them, wether it be video, word processing, 2D/3D graphics, physics, etc... and process it. Lots of decision making, not much calculation - they're not particularly good at doing specialized tasks. x86 architecture CPU's were *terrible* at audio/video encoding until MMX/SSE came out, and they're still relatively bad at it. Dissipating 50 watts of power just to do realtime MPEG4 encoding is embarassing.

I have a Blackfin BF533 DSP evaluation board on my desk, which can encode realtime 320kbps MP3 at a clock speed of 16MHz - and the chip runs at a max clock frequency of 750MHz while dissipating less than a watt. At full speed, it can encode/decode multiple MPEG4 audio/video streams. In video/audio encoding this chip can blow a P4EE out of the water, and I can buy them for $10.

The only thing is, the DSP chip would make an awful desktop CPU - it's simply not designed for any sort of multiuser, multitasking system. It's a specialized processor - it's used in cellphone base station applications, set-top boxes (your PVR box probably has a Blackfin in it), digital video cameras, and other applications where the goal is to process and move the same kind of data over and over again.

If you put a DSP chip in a computer (say, on a PCI card) then applications like DivX, LAME, photoshop and so forth could offload almost all their processing onto the DSP card and work a lot faster. In fact, you can already buy these cards - they're not that common outside of the scientific field however.

Effectively, the Cell is a CPU with a bunch of DSP cores integrated into it. It will absolutely fucking scream at 3D rendering, intense physics or light calculations, video encoding/decoding, audio rendering and mixing, and so on. This is what the playstation 3 will be doing a lot of, and the Cell will kick ass all over the place in this application.

But things like word processing and web serving won't be any different - your computer's CPU does all that kind of stuff, and a vector processor like a DSP would be useless. So for everyday computing, a Cell is a waste of silicon.
 
gee said:
Effectively, the Cell is a CPU with a bunch of DSP cores integrated into it. It will absolutely fucking scream at 3D rendering, intense physics or light calculations, video encoding/decoding, audio rendering and mixing, and so on. This is what the playstation 3 will be doing a lot of, and the Cell will kick ass all over the place in this application.

But things like word processing and web serving won't be any different - your computer's CPU does all that kind of stuff, and a vector processor like a DSP would be useless. So for everyday computing, a Cell is a waste of silicon.

Someone gets it.. though they are not really dsp's, more like single precision vector only ppc's.. But you are exactly right. It will be great at what it does well, but it will not take over the world.

Just like you do not use a motorcycle to move cargo, or a 18 wheeler to offroad, a Cell will not make a good General Purpose comptuer. It will be a great console, and a great scientific workstaion (for weather modeling etc) but not a very good computer in general.

==>Lazn
 
Hand-picked? They aren't going to have to use a standard PC C/C++ compiler... If you design games for a game console, you use specialized stuff for them. I imagine sony will provide it, but, even if not, well, then someone will design it first most likely. I mean, can you imagine trying to optomize every function that should be? Some will be missed for certain, and the possibility of it being an important one is painfully high. Because, they won't know for certain just by looking at the results...

Anyway, what I'd like to see people doing is less looking at what it can do at it's best, and forget what it is at it's worst, what is it in real conditions? I don't know how much of that can actually be guessed at before we have one to compare directly with, but, it seems to me like a little bit of it can be imagined. We already know how 3D games work basically. It's a lot more than just graphics, like someone said. There will be a lot of branching dependencies in a real game, which will, purportedly, limit it's 8 APUs to being almost just one CPU whenever it runs into them. However, I will be the first to admit that these games are more and more graphical as well to the extent that we have found our video cards being too far above everything the games can throw at them so that our CPU becomes the limit (I'll tell you one thing, I'm rather enjoying playing even games like Doom 3 and Far Cry with the highest possible quality settings...) Thing is, those games do require that you have a fairly decent CPU because they are doing a lot of calculations beyond just the video (especially FC I suppose, Doom 3 seems too much a contained environment.)

Now, realistically speaking there are a lot of games that get away with miniturized environments. Such as Soul Calibur. Thanks to those small environments, these games get to really shine through in the graphics department. I mean, just look at how incredible SC looked on the Dreamcast even. But, then we get to more complex games. Some have large complex areas where you can go around and do what you want (and aren't we all just sick of too much linearity?) Can it handle that? It's going to have to be doing some basic graphics at least, sound, ai, physics, modifying variables such as how much net time has passed (why do games keep up with this? It always makes me feel rushed) and so on throghout a big list that admitedly will depend on the game in question. Well, maybe it is possible only to tell by a direct comparison, but, how do you get it? So many console games are made for one console only, only a few of the most popular get ported back and forth typically.

EDIT: Bah, I stepped away for a bit and didn't finish, then come back and it seems during the time it took me to do all that someone else posted a fair bit. Well, for one thing, when WE can get those MPEG4 realtime encoder cards that use 1W supposedly (come on, however little power it needs, 1W? My video uses more than that and it's a videocard made to do videocard things so it's not being used as a CPU or something silly like that) for a mere $10, let me know. I'm willing to bet they don't sell for that cheap here in the real world... Thing is that those things are rather specially designed for one use and that's about all they are good for. When MPEG5 comes out in some future date, will it do that better than the CPU? Or will the cpu have to take the brunt of the work while trying to adapt some of the data to work on that thing to get a 1% increase? As for encoding MP3s, well, my outdated Barton does that in seconds, and I insist on using high quality vbr rather than wasting storage space on wasted bits like that. (PS, I've read that 256Kbps is true cd quality, not 320. Everyone just assumes 320 is the best that can be done simply because it's the highest most encoders can handle, but, the law of diminishing returns applies here too.)
 
"It will absolutely fucking scream at 3D rendering, intense physics or light calculations".

Isn't that what we all build high-end gaming rigs to do?

Who cares if it's not any faster running Outlook or IE. You don't need
a ton of horse power for that stuff anyway.
 
Maynerd Goldstein said:
"It will absolutely fucking scream at 3D rendering, intense physics or light calculations".

Isn't that what we all build high-end gaming rigs to do?

Who cares if it's not any faster running Outlook or IE. You don't need
a ton of horse power for that stuff anyway.
Yep... but what i'm saying is that comparing the Cell to a P4/A64 for light/physics/graphics/etc, the processing capabilities of the Cell will make the competing x86 processor look absolutely silly for these types of processing.

On a sidenote, i'm personally expecing the playstation3 to suck. I own a playstation 2 and there's only a few games on it that I like, but I've got a stack of about 50 NES games that I'm still partly addicted to. ;) Graphics/processing/etc can't replace gameplay.
 
Nazo said:
EDIT: Bah, I stepped away for a bit and didn't finish, then come back and it seems during the time it took me to do all that someone else posted a fair bit. Well, for one thing, when WE can get those MPEG4 realtime encoder cards that use 1W supposedly (come on, however little power it needs, 1W? My video uses more than that and it's a videocard made to do videocard things so it's not being used as a CPU or something silly like that) for a mere $10, let me know. I'm willing to bet they don't sell for that cheap here in the real world... Thing is that those things are rather specially designed for one use and that's about all they are good for. When MPEG5 comes out in some future date, will it do that better than the CPU? Or will the cpu have to take the brunt of the work while trying to adapt some of the data to work on that thing to get a 1% increase? As for encoding MP3s, well, my outdated Barton does that in seconds, and I insist on using high quality vbr rather than wasting storage space on wasted bits like that. (PS, I've read that 256Kbps is true cd quality, not 320. Everyone just assumes 320 is the best that can be done simply because it's the highest most encoders can handle, but, the law of diminishing returns applies here too.)
Ignorance sure is bliss, isn't it Nazo? Perhaps you would do best to leave your judgements of DSPs and FPGAs to yourself. FPGAs are highly adaptable and customizable, and extremely powerful. You may not know it, but FPGAs and DSPs are already used in high end soundcards like the RME Audio cards. The possibilities don't end there. Instead of expressing doubt about something you can't even comprehend, maybe you should just let gee do the talking and see your way out of this thread.
 
Nazo said:
(PS, I've read that 256Kbps is true cd quality, not 320.
Haha, I just wanted to quote this for posterity.

Where did you read that, the National Enquirer?

Come on down to the Computer Audio forum and we'll set you straight.
 
xonik said:
Ignorance sure is bliss, isn't it Nazo? Perhaps you would do best to leave your judgements of DSPs and FPGAs to yourself. FPGAs are highly adaptable and customizable, and extremely powerful. You may not know it, but FPGAs and DSPs are already used in high end soundcards like the RME Audio cards. The possibilities don't end there. Instead of expressing doubt about something you can't even comprehend, maybe you should just let gee do the talking and see your way out of this thread.
I was gonna reply to this earlier, thanks for the clarification. Nazo tried to describe a DSP, but instead described an ASIC.

I've got a PVR350 in my HTPC at home, which does MPEG2 encoding onboard. It doesn't do MPEG4, and there's no way the chip could ever do MPEG4 or even remotely accelerate it; the chip just isn't designed for that, because MPEG2 is fixed in the silicon. Again, it's an ASIC.

DSPs are infinitely reprogrammable to do whatever you want them to do - they're just like a CPU, except their instruction set and architecture is geared towards specific forms of processing. So can the Blackfin on my desk do MPEG5? of course - the code just has to be written for it. Wether it can be done in realtime is another question; considering the MPEG5 standard isn't out, you can't say for sure. But you also can't definitely say that an Athlon64 will do MPEG5 realtime. When MPEG5 is out, we'll know for sure.

But I can't see things like iDCT algorithms and motion estimation (convolution) going anywhere - iDCT is used in jpeg, mpeg1, mpeg2, mpeg4, windows media... a properly selected DSP will *always* kick a general purpose CPU's ass doing this kind of work, and I expect this to continue with MPEG5.

That is, until modern CPU's start integrating better media processing capabilities into themselves. Not just extra instructions - a media coprocessor of some sort would speed things up incredibly... oh wait, what was the purpose of the Cell again? ;)
 
EX-FUCKING-ACTLY...lol if computers are going to move to the next level video and media actions are going to eventually need to be integrated into the main cpu wether it be a fcpga or a "cell" of dsps
 
xonik said:
Ignorance sure is bliss, isn't it Nazo? Perhaps you would do best to leave your judgements of DSPs and FPGAs to yourself. FPGAs are highly adaptable and customizable, and extremely powerful. You may not know it, but FPGAs and DSPs are already used in high end soundcards like the RME Audio cards. The possibilities don't end there. Instead of expressing doubt about something you can't even comprehend, maybe you should just let gee do the talking and see your way out of this thread.

Ye, thanks for clearing that up xonik.

That guy sure rants on alot. Nazo maybe you should consider taking xonik advice and keeping your judgements to yourself.

Once again xonik thanks for clearing up someone's misinformation.

;)
 
Keeping my judgements to myself? Last I checked, this thread was about opinion... Correct me if I'm wrong here...

I've already answered each of the points, but, I've grown a little tired of repeating myself, and my original question remains unanswered, so, just have fun with your little discussion of DSPs. Frankly, I must point out that the cell isn't a DSP anyway, but, I still find it quite hard to believe a chip made for sound processing can do all this stuff it wasn't even made for in realtime.

BTW, PhyberOptik, I forgot to answer your little question earleir about OCing. It seems to me like this device would overclock a tad bit but probably would be more sensitive due to the multiple cores. Maybe not, but, that's the impression I get. Then again, if it's half of what it claims to be, a small OC would make a larger difference.
 
Nazo said:
Keeping my judgements to myself? Last I checked, this thread was about opinion... Correct me if I'm wrong here...

I've already answered each of the points, but, I've grown a little tired of repeating myself, and my original question remains unanswered, so, just have fun with your little discussion of DSPs. Frankly, I must point out that the cell isn't a DSP anyway, but, I still find it quite hard to believe a chip made for sound processing can do all this stuff it wasn't even made for in realtime.
I find it quite hard to believe that you know what you're talking about :p

Ever thought that vector processors and DSPs might be the same thing? eg:

vector processor = specialized processor which is designed to execute the same operation on each value in a large array of data... these run fast through parallelism of instructions, simultaneous fetch/store/execute capabilities, built-in hardware loop capabilities, etc.
DSP = specialized processor which is designed to execute the same operation on a stream of data. You're hard pressed to find a DSP chip that can't do at least two multiply/add operations while simultaneously fetching and storing from memory, or a DSP that doesn't have hardware loop capabilities.

Mind you, "vector processor" is more of an academic term, and DSP is more of an industry term. You can't just go out and buy a vector processor.

And stop assuming that DSPs are designed only for sound.
 
But that's the only example I gave, which sums up the amount of knowledge he has on this topic :p
 
Nazo said:
Keeping my judgements to myself? Last I checked, this thread was about opinion... Correct me if I'm wrong here...

I've already answered each of the points, but, I've grown a little tired of repeating myself, and my original question remains unanswered, so, just have fun with your little discussion of DSPs. Frankly, I must point out that the cell isn't a DSP anyway, but, I still find it quite hard to believe a chip made for sound processing can do all this stuff it wasn't even made for in realtime.

BTW, PhyberOptik, I forgot to answer your little question earleir about OCing. It seems to me like this device would overclock a tad bit but probably would be more sensitive due to the multiple cores. Maybe not, but, that's the impression I get. Then again, if it's half of what it claims to be, a small OC would make a larger difference.

http://www.epanorama.net/links/dsp.html

"DSPs are special microprocessors designed to execute repetitive math-intensive algorithms. Today many embedded applications require both types of processors. Digital Signal Processors have approximately the same level ofintegration, the same clock frequencies as general purpose microprocessors. Pn signal processing tasks DSPs overtake general purpose processors from 2 to 3 order in speed. This is because of architectural differences. "

==>Lazn
 
Back
Top