Will Microsoft trump CELL by using Proximity Communication in XB720 CPU?

[Bubonic]

Cell in XB3000 would do wonders for cross-platform development and make the idea of competing consoles pretty much obsolete.

Fiirst forgot to say hi to all you B3D guys. Long time viewer, first time poster.


Personally in my opinion, I would not like to see that. Non competing consoles equals bad for the consumer, no matter what way you look at it. I would actully like to see a Fusion based processor in the next xbox, or any of the consoles for that matter.

Easy peasy! PS3 is at 200+ GFlop now. Put four in PS4 clocked at 4 GHz, and you've broken 1 TFlop peak. You're look at more than 4x a power increase between console generations (unless you're Nintendo ) - as much as 10x.

Why on earth would you go out and put 4 CELL processors in a console?
How exactly would the PS4 benifit from that? How would you put four CELL processors to good use? Now tell me, would it not be more ideal to put multiple GPU's instead of CPU's? A closed system would benifit much more from GPU power than CPU, after all, we are talking about a game machine. I can't stop thinking about the possibilties of a multiple GPU ecosystem in a console.


Answer me this...
Which console would win if this were to happen?

A PS4 with 4 CELL blade CPU's and one RSX

or

A 360 with 4 Xenos GPU's, and one Xenon.



Keep in mind that we already have half teraflop GPU's. No telling what GPU's would climb to when we are in the year 2010.

 

[Rangers]

It's questionable how much CPU performance actually helps you sell consoles anyway (e.g. MD > SNES, Saturn [supposedly] > PS1, Xbox > PS2, PS3 > 360). Might as well make it easy to make games for and focus on graphics hardware ...

Exactly, that was the point I was trying to make earlier.

In the end the GPU is more important to making the screen look pretty.

 

[TheChefO]

Answer me this...

Depends on how well those gpu's can run physics calcs.

I hope (fingers crossed) physics based gameplay becomes a factor by next gen (2011). Graphics are nice, but accurate and varied physics will enable more interesting interaction with these pretty graphics which in my book is worth more than a prettier picture.

I'd be happy with a slight improvement to this gen graphics (slightly higher poly count + more texture ram) w/16xaa along with extreme physics interaction rather than the predicted linear increase in graphics ability and interactivity.

Now if this can be done just as well on the gpu(s), I say go for it. If it isn't comperable to cell in this regard, I'd say find another solution. Perhaps it is a cell-like cpu, or perhaps a coprocessor (ppu). Perhaps a hybrid: Terascale/Hero + Xenon(3-6core).

However they get there, I hope interaction scales along with graphics as it seems to be lagging currently on the scale in most nextgen games.

 

[Ragnarok]

Answer me this...
Which console would win if this were to happen?

A PS4 with 4 CELL blade CPU's and one RSX

or

A 360 with 4 Xenos GPU's, and one Xenon.



Keep in mind that we already have half teraflop GPU's. No telling what GPU's would climb to when we are in the year 2010.

I think a happy medium might prove to be the most efficient:

3xCell (24 SPU's) + 2xRSX + 8GB RAM =

 

[Rangers]

How is that insinuating what MS was attempting to do? I was pointing out that OOOE is hard, and thus something you can't just tack on, meaning that they couldn't be just surprised at the last moment that OOOE wasn't going to be there. I think you are way too sensitive about this stuff.



Please don't make an appeal to authority here. We've never heard MS or IBM engineers say it was possible, only that they couldn't do it. And going by existing products, logic would dictate that a 3-core OOOE 3.2Ghz CPU on the 90nm is totally impossible within console costs and power consumption limits.

Who says they "needed" (or wanted) 3.2ghz? This hypothetical OOOE CPU planned for 360 could have just as easily been 2ghz. And, we should know that mhz means jack all to CPU performance. Look at how a 2ghz A64 smoked a 3.4ghz Northwood P4 not too long ago. Even Intel has now actually decreased the mhz of it's chips quite a lot with Core2dou. Hell if anything, high mhz seems to be a negative as it really ramps power and heat. Core2dou performs stellar at comfortable 2-3ghz ghz ranges currently.

Maybe, but as debated throughout this thread, this is quite a challenge. If they're stuck with a Quad or Sextuple core OOOE CPU, and Cell2 has something like 32 SPEs, then they would lose by an order of magnitude in performance. Even worse, Cell2 could have 1 or 2 OOOE CPUs as the PPE, thus getting the best of both worlds. That's why I suggest that MS should simply license a variant Cell for the Xbox3. Any other strategy would be very expensive or very underperforming.

You neglect that imo, we're simply not seeing the benefits of Cell in PS3 games. Where are the awesome things they're doing that simply aren't possibly on games running on 360 or PC CPU's? It was always said you'd see more CPU power show up in the physics or AI..well, I haven't seen stunning examples of those that put other games to shame on Cell just yet. In fact by far the best physics I'm seeing is Crysis, running on those dodgy OOOE PC Cpu's.

Personally, I think a more traditional CPU design is better for gaming. I even think, cell2 could be a hindrance in PS3, especially if MS counters with a big beefy symmetrical CPU, like intel's monster nehalem. That's just my personal opinion.

Too me it's about multi-threading the engine, and that seems hard to do on Cell, because it's asymetrical, the 256kb LS limtations, etc. Maybe they'll figure it out.

But there's no way Cell is showing up in Xbox4, that's a pipe dream I promise. MS is not going down that road, that is Sony's baby. Like I said, if nothing else pride will prevent that.

 

[Rangers]

I think a happy medium might prove to be the most efficient:

3xCell (24 SPU's) + 2xRSX + 8GB RAM =

2x RSX will be a joke by next gen

 

[Ragnarok]

2x RSX will be a joke by next gen

2x(whatever the RSX of that time is)

 

[pjbliverpool]

No i'm telling you that, given a multi-threaded architecture, nobody in their right mind would develop software systems based on a single-threaded model in the hopes to gauge any kind of performance metric for the system as a whole.. How can someone say "Cell sucks because it stinks at single-threaded games" when the game itself leaves 7 whole cores sitting idle at run-time..?

There are certain aspects of code, including game code which require single threaded performance and cannot benefit greatly from multiple threads (so I understand anyway). Im not saying "Cell sucks", im saying its not better at everything. When you require single threaded performance then Cell will be poor compared to a modern x86 Core. Its no good saying the devs wouldn't code for single threaded performance if that particular piece of code cannot be spread across multiple threads.

Such comment leads me to believe that your the kind of person of the mindset that Just because of Cell's highly touted FLOPS performance it has no other practical use other than uber fast number crunching.. The truth is Cell IS great at vector math but that's not all it can do either.. & the truth is that the more of your code you can efficiently vectorise and push through the chip, the faster it [your code] will go..

Sure, and I don't dispute that. But your post read to me as though you were saying any code can be re-written to take advantage of Cells superior vector performance and thus run better on Cell than an x86. Regardless of what that code is.

Im saying that some code simply won't benefit all that much and thats were Cells advantages evaporate. It may still be comparable or even superior in other areas but the point is that its not "an order magnatude" in everything "as long as you optimize your code". Some code just won't work all that well on Cell (relatively speaking). The big question is whether games code fits into this catagory - not GPU style graphics code - but the type of code you can't run on a GPU and you need a seperate CPU for.

How about looking at the PC market as a whole instead..? How about comparing the market share of a company like Intel with that of a company like AMD..?

The gaming PCs are nothing but a fraction of the total number of PCs sold each year and I'm pretty darn sure a company the size of Intel couldn't survive off that demographic alone..

Of course it couldn't, but it doesn't change the fact that gaming performance sells a lot of CPU's at all levels, and not just because of gaming, but because of the the mind share that CPU gets from being perceived as faster as measured in the majority of benchmarks sites publish when a new CPU launches. If it were as simple as "add 20% more FP performance and add 20% gaming perofrmance" then im sure AMD or Intel would be jumping at it. Or we could just look at the classic example of a 3Ghz Celeron vs a 2Ghz A64. The Celeron has the higher peak FLOPs performance yet much lower games performance. Clearly there are other more important factors at play.

 

[Bubonic]

Depends on how well those gpu's can run physics calcs.

However they get there, I hope interaction scales along with graphics as it seems to be lagging currently on the scale in most nextgen games.

I'm not too worried about how well 4 xenos GPU's can calculate physics processing.


I would really like to see a multi GPU ecosystem enviroment inside a console. Their is just so many different possiblities that can be used at disposal. Here are some ideas of mine that are possible in a multi GPU console, in this case 4 gpu's.

1) A game app that uses all 4 GPU's for rendering. Their are different outcomes for this option. A devloper can use all 4 GPU's horse power to make a game look as close to realistic as possible, or use all 4 GPU's to make a semi good looking game silky smooth.

2) Use either one, two or three of the GPU's for rendering and another one for nothing but physics.

3) Use either one two or three of the GPU's for rendering, and the other one for applying AA/AF.

4) a combination of both options 2 and 3.

 

[Deusp]

It should be 4xCell + 2xGeforce8800.

 

[Rangers]

2x(whatever the RSX of that time is)

Again you're talking massive power, heat, cost, etc. Why does this never seem to be an issue with building a hypothetical PS4

It seems all computing is going less power, heat, but PS4..

 

[Bubonic]

I think a happy medium might prove to be the most efficient:

3xCell (24 SPU's) + 2xRSX + 8GB RAM =

3x Cell = 21 SPU. At least when talking about PS3's cell.

 

[Shifty Geezer]

Why on earth would you go out and put 4 CELL processors in a console?

Because that'll be the level of CPU power. You know. Each gen you put in a bigger, badder CPU than last gen. I can't imagine anyone saying 'let's use the same CPU we've been using for 5 years and just put in a better GPU this time (not even Nintendo think that. They couple the same CPU with the same GPU )

How exactly would the PS4 benifit from that? How would you put four CELL processors to good use?

Totally morphic worlds, where all matter can be acted upon, combined, burnt, assembled, broken. Think LucasArts Digital Molecular Matter to the nth degree. Massive procedural content. Full realtime behavioural animation and lifelike facial motions. EyeToy-like interfaces with extremely advanced recognition algorithms. Lots of complex processing stuff, even more complex than what we can do with Cell. Cell isn't the limit of what's useful!

Now tell me, would it not be more ideal to put multiple GPU's instead of CPU's? A closed system would benifit much more from GPU power than CPU, after all, we are talking about a game machine. I can't stop thinking about the possibilties of a multiple GPU ecosystem in a console.

You'd be stupidly hampered by the CPU. You have to create vertex data for those GPUs, and that's the CPU's job - to fetch meshes, apply physics, animations, and produces triangles for the GPU to shade. You may well be able to move a lot of that onto the GPU next gen, but using Cell will allow the use of tried-and-tested techniques developed now on PS3, and scalable across Cell's archtiecture. Basically rather than recreating your engine, you can copy it over and double everything up (thinking of things like SPURS job distribution across available SPEs). All the benefits of masses of vector processing power like GPUs, but with the coding niceties of a known system.

Answer me this...
Which console would win if this were to happen?

A PS4 with 4 CELL blade CPU's and one RSX

or

A 360 with 4 Xenos GPU's, and one Xenon.

The actual choice would be more like a 4 Cell (probably more than that) CPU and G120 GPU, or Quad Core Pentium with 4 R900s or whatever. Obviously that's hard to visualize, so if we normalize this to our current tech, you would choose Cell and a 9800 over a 1.5 GHz P4 and G80? I know I would, because I'd prefer the better gaming options Cell gives over the better graphics of the G80. Of course if GPUs have become far more general purpose by then, the issue is muddled...

 

[liolio]

I'm pretty sure that Ms will firstly consider the gpu.
Where will be directX at this time, may be directX 11.

Ms will first consider wich can of job can be done properly on the gpu or the cpu.
feature like memexport supposedly could bring some flexibility in calculation for such a thing as folding.
How many improvement amd/ati and nvidia will be able to bring by 2010/11?
How flexible gpu will be.
i've the feeling that with :
geometry shader
tesselator
possibly collisions test running on the gpu
more and more lightning done on the gpu
some physics effects.
etc.

Ms will chose a cpu that is very efficient at general purpose tasks, and with good acceleration for thing like network code and decompression.
For me Ms will spend a lot more transistors on the gpu than on cpu.

 

[Shifty Geezer]

3x Cell = 21 SPU. At least when talking about PS3's cell.

The current disabled SPE won't be an issue going forwards, though redundancy for yields on massive multicore will likely be a reality, same on GPUs.

 

[Ragnarok]

3x Cell = 21 SPU. At least when talking about PS3's cell.

I'd like to think that by PS4 days Sony would have attained high enough yields to include the actual 8 SPU Cell.

 

[Nite_Hawk]

Hi Guys,

IBM came out to here to give a Cell tutorial today and tomorrow so I'm actually writing code for it now. At least at the most basic level it's not all that different from writing MPI code. You can move stuff around between the PPE and the SPEs via DMA and do fairly standard simd and vector operations. After figuring out mailboxes and what methods you can use for DMA transfers, it's pretty clear how things work.

I'm sure there are going to be much more subtle and difficult things that crop up that I haven't had to deal with yet, but honestly getting into it isn't nearly as bad as I thought it would be. You can have a basic program using the PPE and all your SPEs up and running in under a half an hour.

Nite_Hawk

 

[ban25]

This must be some kind of belated April Fool's thread. Project Hero is Sun's failed DARPA supercomputing bid, which was eliminated in favor of solutions by IBM and Cray. It's also quite old (back in 2003, if I recall correctly).

 

[JardeL]

If Sony will release PS4 I don't think they will take a risk again, full of negative responses around for the Cell now ... When programmers learn how to run Cell @ full performance , probably at 2009 - 2010 , i don't see any reason to move for another CPU... Here is Cell BE Roadmap ;

2 PPE + 32SPE by 2010 ... Probably powerful enough to compete with its rivals [ with a powerful GPU , designed from the beginning , not a last second move like RSX ] ... My thoughts based on current Cell's success , but if it fails it'll be dramatic for Sony , IMO ...

 

[Shifty Geezer]

I'd like to think that by PS4 days Sony would have attained high enough yields to include the actual 8 SPU Cell.

In a big CPU, you get defects, which stop bits of the processor working. If PS4's Cell is 300 mm^2 again, they'll be taking a hit on chips. I've no idea how many transistors we'll be at by then, but let's just say for the argument that 4 PPEs and 32 SPEs fit on 300 mm^2. Yields on 100% working chips will be as low then as 1:8 Cells are now. If Sony are willing to lose up to 4 PPEs, yields will be much improved. Alternatively they go with a smaller chip with better yields, such as a 2:16 Cell at 150 mm^2, or even 4 1:8 Cells networked together.