Does Cell Have Any Other Advantages Over XCPU Other Than FLOPS?

[MfA]

Large FPGAs are expensive and have atrociuous power consumption.

 

[Kryton]

Obviously I'm no expect on FPGAs, but it seems to me that these chips offer good value and flexibilty up to a certain complexity of program, but beyond that you hit limits, and I guess they lack the developer tools of traditional language based programming. How easy would it be for me to write a 3x3 convolution matrix given knowledge of how to do this in C++? Are my existing skills portable? What's the performance limit of FPGA's at the moment and how is that likely to progress?

Yes they do have issues. Hence they all have embedded ARM cores that allow you to do such things, want edge detection stick it on the ARM to operate on the FPGA output.

This is premade technology that has been about far longer with existing programs written. You use this exact reason to argue for Cell when you forget that Cell will require someone to pay for all these changes and Sony/Toshiba won't lump the burden so it will be pushed onto the pricetag. Then the cost of changing from ARM code to Cell code will be the area that puts most off. (Perhaps with things like Jazilla on ARM if Sony endorse Java on Cell people will transition but I doubt it)

FPGAs + ARM have this market tied up much in the way Windows has the desktop OS market tied. No one wants to jump to Good Ship Linux if they can't take their software and existing practises with them.

 

[Kryton]

Large FPGAs are expensive and have atrociuous power consumption.

So are Power PC chips. If you say 'that isn't needed they will just use the SPU' then why not just use an ARM chip for which you won't need to reinvent the wheel in your software.

 

[LunchBox]

i think sony's main goal is to make things cheaper for them to manufacture different products through cell...

since cell is mainly being produced for the PS3...

they'll be making a lot of cells...

so any functional cells that didn't make the cut will be used accordingly...

eventhough a cheaper altnative can be made let's sayfor their blu-rayplayers or for their HD TVs,

they'll end up using cell anyways since its essentially "free" to them...

("free" meaning they'll throw them away if not used, since they're the chips that didn't hit the cut-off spec for PS3 or their parallel computing thing.)

 

[Kryton]

i think sony's main goal is to make things cheaper for them to manufacture different products through cell...

since cell is mainly being produced for the PS3...

they'll be making a lot of cells...

so any functional cells that didn't make the cut will be used accordingly...

eventhough a cheaper altnative can be made let's sayfor their blu-rayplayers or for their HD TVs,

they'll end up using cell anyways since its essentially "free" to them...

("free" meaning they'll throw them away if not used, since they're the chips that didn't hit the cut-off spec for PS3 or their parallel computing thing.)

Didn't we know that if the PPC is faulty the entire chip must be binned? So, you need this mighty PPC sucking juice up which is fine in some CE applications but not all. The areas where you can get away with it are where FPGAs can be used, the areas you can't are where ARM has the market tied up.

 

[mckmas8808]

Didn't we know that if the PPC is faulty the entire chip must be binned? So, you need this mighty PPC sucking juice up which is fine in some CE applications but not all. The areas where you can get away with it are where FPGAs can be used, the areas you can't are where ARM has the market tied up.

So are you basically now saying that no company will end up using CELL chips in any CE products?

 

[Kryton]

So are you basically now saying that no company will end up using CELL chips in any CE products?

Well done. I have been saying since the start it only makes sense to Sony and Toshiba to use it in their own products to save them licensing from ARM and buying FPGAs. To everyone else it is another alternative and with such a momentum built up behind ARM and the FPGA market it is hard to see even Sony jumping in and stopping it.

 

[expletive]

i think sony's main goal is to make things cheaper for them to manufacture different products through cell...

since cell is mainly being produced for the PS3...

they'll be making a lot of cells...

so any functional cells that didn't make the cut will be used accordingly...

eventhough a cheaper altnative can be made let's sayfor their blu-rayplayers or for their HD TVs,

they'll end up using cell anyways since its essentially "free" to them...

("free" meaning they'll throw them away if not used, since they're the chips that didn't hit the cut-off spec for PS3 or their parallel computing thing.)

I agree that the idea of a scalable multi purpose Cell architecture could be a good thing for Sony. I htink what we're debating is if the Cell we provide any benefit to the consumer.

 

[mckmas8808]

Well done. I have been saying since the start it only makes sense to Sony and Toshiba to use it in their own products to save them licensing from ARM and buying FPGAs. To everyone else it is another alternative and with such a momentum built up behind ARM and the FPGA market it is hard to see even Sony jumping in and stopping it.

Okay I can respect that. But then can you answer this? What if Sony and Toshiba use the mini-Cell chips in their CE products and in a result start producing better CE products than the others (even though the price will be a little higher)? Could it then be possible that other companies will jump ship or do you think they will still use a lower price hardware that doesn't allow them to do certain things as well?

 

[Shifty Geezer]

Yes they do have issues. Hence they all have embedded ARM cores that allow you to do such things, want edge detection stick it on the ARM to operate on the FPGA output.

FPGAs + ARM have this market tied up much in the way Windows has the desktop OS market tied. No one wants to jump to Good Ship Linux if they can't take their software and existing practises with them.

Yes, but I think you're looking at the immediate market, whereas Cell is looking to the future. In 10 or 20 years time, are FPGAs going to be able to provide the performance and cost effectiveness of whatever's being done? Also regards the software...

This is premade technology that has been about far longer with existing programs written. You use this exact reason to argue for Cell when you forget that Cell will require someone to pay for all these changes and Sony/Toshiba won't lump the burden so it will be pushed onto the pricetag.

Cell is going to have the benefit of a presumably large enthusiast market generating code, and even if not the SDK and development of custom code should be no more than developing code for ARM. If the ARM component can run an existing Edge Detection, can it also be programed to solve an 4 dimensional nth order Infinite Regression of a Hooky-Coburt Polynomial Abstraction in realtime when such a routine is invented? I'm sure Cell is total overkill for current apps, but in ten years time it'll be cost effective, very well understood, and scalable to any need, whereas how will FPGA's be able to cope? ASIC's were great for their time, but are becoming more costly. FPGA's fill the role amicably, but have limits on their versatility and scalability. Cell will provide the versatility and performance of future applications and it'll have it's time until an alternative such as artifical neurons comes to replace it. The capacity for humanity to consume technology and processing power is seemingly insatiable.

 

[Kryton]

Okay I can respect that. But then can you answer this? What if Sony and Toshiba use the mini-Cell chips in their CE products and in a result start producing better CE products than the others (even though the price will be a little higher)? Could it then be possible that other companies will jump ship or do you think they will still use a lower price hardware that doesn't allow them to do certain things as well?

We are talking programmable devices here: they are all computationally equivalent (from a theoretical level). Speed may be the only issue so people will just push for the hardware they know to go faster perhaps at the sake of power consumption or at the sake of cost. I doubt they would jump ship though because of the vast investments they have already made in this architecture (for Sony and Toshiba they have made huge investment in Cell so this point is mute for them) such as designing compatiable bus objects etc.

 

[expletive]

Okay I can respect that. But then can you answer this? What if Sony and Toshiba use the mini-Cell chips in their CE products and in a result start producing better CE products than the others (even though the price will be a little higher)? Could it then be possible that other companies will jump ship or do you think they will still use a lower price hardware that doesn't allow them to do certain things as well?

Imo, if Sony and Toshiba start producing better CE products it wont be because theyre using a Cell processor.

Currently, some of the best video processing in the world is running on FPGAs with custom, proprietary software algortihms. Between all the companies that do this sort of thing, a lot of them are using the same FPGAs, the differentiating features are software, price, and functional design.

I'm not saying that a 'better CE device' is simply video processing, just using that as an example since its been somewhat of a poster-boy for this thread.

 

[mckmas8808]

10 years Shifty? What about 5 years from now? But you point is very valid and I agree. Cell is more the future and for future applications not current ones.

 

[Shifty Geezer]

Currently, some of the best video processing in the world is running on FPGAs with custom, proprietary software algortihms. Between all the companies that do this sort of thing, a lot of them are using the same FPGAs, the differentiating features are software, price, and functional design.

Though there haven't been any rival solutions though, have there? If you don't go with an FPGA then the price/performance of a conventional processor would be prohibitive I'd have thought.

@ mckmas8808 : Dunno about a timelime. I don't know how well FPGA's can scale for future operation, what level of 'advances' are going to made that'll want lots of processing power, etc. On the one hand things always take longer to become established then you'd think, but on the other technology races along and some things change before you know it.

 

[Bobbler]

I think some of you are getting ahead of yourself. Cell will not offer anything more than what is currently available (depending on what chips you use) -- it isn't as if Cell is a magical CPU that allows things physically impossible by something else, especially in the CE realm. On the flip side, some of you are assuming the other side is talking about using 2+ghz Cells in the CE devices, when it's likely not the case -- they'd use appropriately clocked Cells.

What Cell does offer is a cheap source of chips for Sony and Toshiba -- it also offers them marketing leverage. Cell Inside!

When people see PS3 and hear more about Cell from the countless misinformed articles that will arise surrounding it's launch they'll get this image in their head that Cell makes things better. Seeing Cell put into other CE devices will have a marketing benefit, along with being a cost effective solution for Sony and probably Toshiba. After a while, if Cell CE devices prove successful then you might just see other companies going after Cell stuff (because of marketing, not necessarily because a Cell will provide something an ARM + DSPs can't).

Cells going into CE devices will likely be clocked at a fraction of the speed of a Cell going into a PS3 (likely at a much lower voltage as well), probably only clocked enough to provide exactly what they need and not much more. Right now you can get fairly beefy ARM chips and a handful of DSPs for pretty cheap and be just as happy -- the reason you'd use a Cell is because of marketing benefit of having the word's "Cell inside" or "Cell embedded" or something on the outside of it. That's the only reason I can think of, and frankly, it's a pretty powerful reason (at least it quite possibly could be in a couple years) -- not because we need our TV's to mess with 40+ mpeg streams and crack encryption. I imagine a few hundred Mhz mini Cell (300-500) would provide more than enough for almost any CE device, and be very low power consumption (especially at 65-45nm). The limitation on CE devices currently isn't the processing power -- which some are forgetting -- it's, what do we do next?

Don't get me wrong, I rather like hearing about the Cell and I'm glad to see companies trying new things; I'm looking forward to my PS3 (more so than anything else this gen).

 

[Carl B]

I agree with the notion that in the near term, Cell is probably not going to take away from the dedicated chips used by other companies in the CE space. The almost guaranteed longevity of the architecture, however, will allow it to keep at trying to break into that market - sustained by Sony and Toshiba's own use of the chip within their respective products and of course the PS3. In the coming years, the Cell variations will become more varied and the code base for Cell will mature. Maybe then they'll be ready for adoption by the CE industry at large. Or, maybe not.

Either way I think it's been a good 'gamble' on the part of Sony. Cell in part has been a component of a larger strategy to wean themselves off of sourcing chips from outside the company, and of course the majority of their investment in Cell has come in the form of additional fab capacity. Whether Cell succeeds or fails, that fab capacity will allow them to accomplish that goal as it can just as readily be utilized for the production of in-house developed DSPs, FPGAs, or what have you.

But of course, Cell will not 'fail' by virtue of the PS3 alone. So the question then becomes, by what do we measure the various grades of 'success'?

I see Cell being a part of PS4, and the architecture as a whole sticking around for a while, so it'll be interesting to watch if and when Cell gains traction as the years go by.

 

[expletive]

Though there haven't been any rival solutions though, have there? If you don't go with an FPGA then the price/performance of a conventional processor would be prohibitive I'd have thought.

@ mckmas8808 : Dunno about a timelime. I don't know how well FPGA's can scale for future operation, what level of 'advances' are going to made that'll want lots of processing power, etc. On the one hand things always take longer to become established then you'd think, but on the other technology races along and some things change before you know it.

2 companies have recently introduced custom chips, gennum and realta. They dont sell finished products, only the chip and the associated algoritms with it. These technologies are at the higher end of the food chain in terms price and are marketed to be providing the very best in video processing, (doing things like noise reduction and advanced upsampling) but in a whole year theyll probably sell less than of these chips than Sony sells of any one of its TVs. Obivous reason being that theres a very small market of people who have the resources to drop $1-$7k on a stand-alone video processing solution that only they themselves can truly appreciate.

In terms of competing in the lower-cost space, theres lots of options. I'll go back to my cable box example. THe chip in there can basically do anything that we could want at the mass-market CE level and its been available for close to 2 years already. The cost of this chip has probably dropped a bit by now and its pricepoint will probably be replaced by an even more funcitonal one.

 

[Shifty Geezer]

it isn't as if Cell is a magical CPU that allows things physically impossible by something else, especially in the CE realm.

Only it is. Unless you can point me to another single-chip processor that can build a 3D scan of a human torso in 3 seconds . Obviously for existing apps there's other solutions, but there's some stuff Cell is just better at. Whether these will actually be of use in CE space is another matter.

What Cell does offer is a cheap source of chips for Sony and Toshiba -- it also offers them marketing leverage. Cell Inside!

I wouldn't call Cell cheap. Certainly not as cheap as existing CE solutions. Long term the price will drop but that won't be for a long time.

Quite the opposite to you, I think Cell offers a costly premium solution for high-end products. Sony and Toshiba will be able to use Cell as a single-chip solution at first, offering excessive numbers of features to make a point of having that versatility to hand. I can't see them using Cell to do the tasks of existing cheaper alternatives.

 

[ADEX]

Large FPGAs are expensive and have atrociuous power consumption.

So are Power PC chips.

You can get PowerPCs which go below 1 Watt.

If you say 'that isn't needed they will just use the SPU' then why not just use an ARM chip for which you won't need to reinvent the wheel in your software.

Now try finding an ARM which even gets close to a single SPUs performance.

Embedded systems are not like desktop systems, they obviously want to reuse code but there is no "legacy" to worry about.

ARM has most of the mobile phone market wrapped up but certainly not the entire CE sector. They don't seem to have had many design wins in the console sector...

Take a SetTopBox or cheapo DVD apart and you can find all sorts of different devices, all ASICs or off the shelf parts, you'll not find any FPGAs. There's all sorts of different processors in use, PowerPC, MIPs, Transputers and more.

IBM do a range of STB chips based on a low end PowerPC which includes all manner of I/O ports, gfx, MPEG2 audio / video decoding, decryption, Dolby hardware etc. They've very cheap in big volumes and you'll probably find some in those cheapo DVD players. They use around 2W, that was a few years ago...

Then we wanted HD and funky 3D GUIs, so ATI developed a whole new chip.

Then they wanted H.264 so a new batch of chips will be developed.

Now they're talking about VOIP, maybe even with video, the existing chips can decode this but can't encode so that's another new chip...

Each revision is costing 10's of millions to develop. With Cell you could do those changes in software.

STBs also require other chips e.g. demodulators, usually ASICs because no CPU or DSPs are powerful enough. If you can do that on the processor you can save money. CE is all about saving money.

Cell is too big, too expensive and too hot for that end of the market at the moment. Once they have done a die shrink and made the "mini Cell" or "micro Cell" they've talked about it'll be in the right range.

It'll turn up in high end stuff first and it'll probably be used in place of DSPs, FPGAs and ASICs in industrial stuff. Think mobile phone base stations or big printers. Cells have a big advantage over ASICs and FPGAs in that they are much easier to program and they'll be a lot cheaper. ASICs are cheap if you are producing millions of them, if not you have the problem of paying $15 million + in development costs. FPGAs cost less but they're harder to develop and you'll need a circuit board.

You can bet STI will be pushing Cell into CE, after all it was in part specifically designed for it.

 

[mckmas8808]

Take a SetTopBox or cheapo DVD apart and you can find all sorts of different devices, all ASICs or off the shelf parts, you'll not find any FPGAs. There's all sorts of different processors in use, PowerPC, MIPs, Transputers and more.

IBM do a range of STB chips based on a low end PowerPC which includes all manner of I/O ports, gfx, MPEG2 audio / video decoding, decryption, Dolby hardware etc. They've very cheap in big volumes and you'll probably find some in those cheapo DVD players. They use around 2W, that was a few years ago...

Then we wanted HD and funky 3D GUIs, so ATI developed a whole new chip.

Then they wanted H.264 so a new batch of chips will be developed.

Now they're talking about VOIP, maybe even with video, the existing chips can decode this but can't encode so that's another new chip...

Each revision is costing 10's of millions to develop. With Cell you could do those changes in software.

STBs also require other chips e.g. demodulators, usually ASICs because no CPU or DSPs are powerful enough. If you can do that on the processor you can save money. CE is all about saving money.

Cell is too big, too expensive and too hot for that end of the market at the moment. Once they have done a die shrink and made the "mini Cell" or "micro Cell" they've talked about it'll be in the right range.

It'll turn up in high end stuff first and it'll probably be used in place of DSPs, FPGAs and ASICs in industrial stuff. Think mobile phone base stations or big printers. Cells have a big advantage over ASICs and FPGAs in that they are much easier to program and they'll be a lot cheaper. ASICs are cheap if you are producing millions of them, if not you have the problem of paying $15 million + in development costs. FPGAs cost less but they're harder to develop and you'll need a circuit board.

You can bet STI will be pushing Cell into CE, after all it was in part specifically designed for it.

QFT. More people need to listen to this man.


*Bold are the parts that are most important*