IBM power7 in PS4 ?

[Love_In_Rio]

I have read a rumour about Sony going to include a variation of the future power 7 processor from IBM in Playstation 4. I have also read rumours about the vector unit of the power7 cores based on SPUs, which would make PS4 with this CPU easily backwards compatible with PS3. What do you thing about this ?. Do you know something about the Power 7 CPU and the posibility of being in a console ? If the vector unit is based on SPUs would that indicate that Sony could have exclusive rights to use it ?.

 

[V3]

PS4 doesn't need Power 7 or it's variant. Sony best bet for PS4 is to used current Cell with all the SPUs and all FLEX I/O lanes enable @ 4+ GHz and go with twin 250-300 mm2 GPUs from either PowerVR, AMD or NV, where the GPUs can be made into single die as process improved.

CPU made minimal impact this gen consoles, it'll be even less of impact next gen.

 

[Henriok]

The vector units in POWER7 is not based on the SPUs in Cell, they are regular Altivec units that has been present in G4, G5, Power6 and Cell (in the PPE), and Sony has not exclusive rights to the SPEs (Toshiba uses them in their SpursEngine). I guess that IBM could make a Cell like processor based on POWER7 cores and attach SPEs to it. Personally I think would be a pretty nice idea.

 

[Grall]

CPU made minimal impact this gen consoles, it'll be even less of impact next gen.

Not really sure what basis you use for that statement, I can't imagine it to be particulary accurate by a long shot.

CPUs this gen (not counting Wii, for obvious reasons) made a HUGE impact by their multicore nature, and by being very weak performance-wise unless you really exploited the multiple threads offered by these chips.

It was a BIG paradigm shift for sure.

 

[taucias]

Not really sure what basis you use for that statement, I can't imagine it to be particulary accurate by a long shot.

CPUs this gen (not counting Wii, for obvious reasons) made a HUGE impact by their multicore nature, and by being very weak performance-wise unless you really exploited the multiple threads offered by these chips.

It was a BIG paradigm shift for sure.

Also the CELL has assisted the RSX in a lot of PS3 games.

 

[V3]

Not really sure what basis you use for that statement, I can't imagine it to be particulary accurate by a long shot.

CPUs this gen (not counting Wii, for obvious reasons) made a HUGE impact by their multicore nature, and by being very weak performance-wise unless you really exploited the multiple threads offered by these chips.

It was a BIG paradigm shift for sure.

Exactly my point, next gen if Sony put a 32 cores Cell or Power 7 even, there aren't going to be many developers that are going to take advantage of it. I mean just look the situation currently. Where as if you stuck a beefy GPU, it's easier for developers to take advantage of it and far easier for core gamers to notice the difference too.

Also the CELL has assisted the RSX in a lot of PS3 games.

Sony were better off putting better GPU in PS3 while going for smaller CPU. PS3 had some large silicon real estate ~ 500 mm^2. Even if the trade off between CPU and GPU are quite small this gen, next gen, 32 nm Cell will be around or less than 70 mm^2. if Sony stuck with Cell, they can put 400 - 500 mm^2 into GPUs real-estate.

Sticking with current Cell for PS4, they can have access to the spare SPU that was disable in PS3, they can also clocked Cell higher plus the GPU doesn't need to be assisted. So there will be improvement. Going with 32 cores Cell or Power 7, is like PS2 or PS3 all over again, we will be waiting for developers to take advantage of the extra cores, or it will baby sit a weaker GPU, etc. It's not a good trade off to have 50:50 split in silicon real-estate for next gen.

 

[sunscar]

Exactly my point, next gen if Sony put a 32 cores Cell or Power 7 even, there aren't going to be many developers that are going to take advantage of it. I mean just look the situation currently. Where as if you stuck a beefy GPU, it's easier for developers to take advantage of it and far easier for core gamers to notice the difference too.



Sony were better off putting better GPU in PS3 while going for smaller CPU. PS3 had some large silicon real estate ~ 500 mm^2. Even if the trade off between CPU and GPU are quite small this gen, next gen, 32 nm Cell will be around or less than 70 mm^2. if Sony stuck with Cell, they can put 400 - 500 mm^2 into GPUs real-estate.

Sticking with current Cell for PS4, they can have access to the spare SPU that was disable in PS3, they can also clocked Cell higher plus the GPU doesn't need to be assisted. So there will be improvement. Going with 32 cores Cell or Power 7, is like PS2 or PS3 all over again, we will be waiting for developers to take advantage of the extra cores, or it will baby sit a weaker GPU, etc. It's not a good trade off to have 50:50 split in silicon real-estate for next gen.

I don't think it's your argument that's flawed so much as they way it came off.

It's undeniable that this generation of consoles did as well as they did because of the immense processing resources they provided. But it also cannot be denied that there has been a shift away from the CPU toward to GPU in order to facilitate process intensive applications. I.E. we no longer need a massive CPU as much as we need a disproportionately large GPU that can do those otherwise CPU-related tasks faster/better.

Hence my stance that next-gen consoles atleast will likely be represented by a small, relatively boring CPU, tied to a fairly general and considerably more massive GPU. (Your current CELL+future GPU is a good example of this)

 

[flynn]

Exactly my point, next gen if Sony put a 32 cores Cell or Power 7 even, there aren't going to be many developers that are going to take advantage of it. I mean just look the situation currently. Where as if you stuck a beefy GPU, it's easier for developers to take advantage of it and far easier for core gamers to notice the difference too.

I agree that 32 cores is overkill but you still need some CPU power so you don't become CPU bottlenecked. If you go with a very powerful GPU that means moving subsystems like physics and maybe AI to the GPU.

By now most of the top PS3 developers have enough experience with the Cell, it's no longer the new architecture they need to learn. If Sony gave them a more powerful version, specially if they don't have to adapt their code for limited local storage, and they pair that with a good GPU you could have great looking games since day one.

My hope for this generation is to have games running at 1080/60p with no compromises and higher (way higher) res textures.

If Sony pushes their motion sensor technology that's going to require some CPU power as well, right? So I wouldn't settle for a non-powerful CPU just because we cannot think of how to put 100% of it to use. Either Sony or the devs themselves will come up with ideas that will use that processing power so it better be there even more so if they expect a 10 year life cycle.

 

[Color me Dan]

I'm under the impression that the amount of SPU cores doesn't have to scale linearly. Doing so would be utterly useless (for the PS4 at least) unless developers need/want them. In a entirely hypothetical situation I think Sony would look at the collective worth of their SPUs (i.e "do we want to move forward with this architecture?"), ask how many more of them developers want/need, look over what they need for the OS or whatnot themselves and then decide on the final number of SPUs that will be integrated into the design.
I'm guessing that any update to the PPE would be based on recent IBM PowerPC developments by default unless they've decided to branch it off into a new direction with it.

 

[V3]

By now most of the top PS3 developers have enough experience with the Cell, it's no longer the new architecture they need to learn. If Sony gave them a more powerful version, specially if they don't have to adapt their code for limited local storage, and they pair that with a good GPU you could have great looking games since day one.

More powerful as in faster clock, if Sony increases the number of SPUs or even the size of the local store or timing of local store, devs need to relearn and write another code to take advantage of those. That's the disadvantage of using local store instead of cache, which you trade to get better performance at lower size and power.

If Sony pushes their motion sensor technology that's going to require some CPU power as well, right?

Wii seems to do fine with their CPU. Look a single SPU is plenty to do that kind of calculation. PS Eye is like SD resolution, even if it goes HD next gen, a single SPU is still plenty to process the image. If they need processing power, they can put Cell into whatever peripherals. Cell will be cheap and low power by then.

So I wouldn't settle for a non-powerful CPU just because we cannot think of how to put 100% of it to use. Either Sony or the devs themselves will come up with ideas that will use that processing power so it better be there even more so if they expect a 10 year life cycle.

The SPUs are now used to assist RSX, when they don't have to do that next gen a lot more resources will become available to them and they have to relearn how to put it to use again. Increasing the clock is still the best way from dev perspective to get extra performance from little work.

In addition, I don't understand about Playstation is needing to progress for 10 years. But doing so by sabotaging your first few years so you can see the improvement in later years. Sony is legendary for doing this, PS2 jaggies and texture problems, down clock in PSP and OS memory footprint in PS3. It's like they're trying to do it on purpose.

Besides processing visible improvements are easier to gain with better GPU. Even today sane PC gamers will spend more on the GPU instead of CPU to get their value for money. Moreover, if devs can leverage those SPUs in Cell instead of falling back to PPU, Cell is quite more potent and offers better value than any of today's x86 CPUs.

 

[archangelmorph]

More powerful as in faster clock, if Sony increases the number of SPUs or even the size of the local store or timing of local store, devs need to relearn and write another code to take advantage of those. That's the disadvantage of using local store instead of cache, which you trade to get better performance at lower size and power.

Err.. No we don't.. We just create more work to fill the new SPUs & increase the job sizes of our old systems to maximize load (provided the bus-width increases with the local store size, otherwise there'd be little point of a size increase IMO)..
& shared cache will never give you better performance especially as no. of threads increase..

The SPUs are now used to assist RSX, when they don't have to do that next gen a lot more resources will become available to them and they have to relearn how to put it to use again. Increasing the clock is still the best way from dev perspective to get extra performance from little work.

I'm not sure what your on about with all this "they'll have to re-learn everything" stuff..? We don't have memories like fish you know..

& I'm not sure it's even remotely feasible to expect clock speed increases upwards of 3.2 GHz without compromising the convenience of cheap cooling solutions..

 

[V3]

Err.. No we don't.. We just create more work to fill the new SPUs & increase the job sizes of our old systems to maximize load (provided the bus-width increases with the local store size, otherwise there'd be little point of a size increase IMO)..
& shared cache will never give you better performance especially as no. of threads increase..

I'm not sure what your on about with all this "they'll have to re-learn everything" stuff..? We don't have memories like fish you know..

That's what I meant, when I said dev need relearn ( not that dev forgot ) how to put those SPUs to use with new code. I mean you can give those SPUs any tasks, but considering the cost of larger number of SPUs is transistors from GPU, can you give those SPUs jobs that'll give better visual improvement compare to just using better GPUs.

Can't the 8 SPUs in the current form of Cells be enough to handle those jobs instead of going, say to 32 SPUs Cell ? With better expected GPU do you have jobs for those 32 SPUs that gamers can recognize and enjoy ? Do you see any clear benefits that you can do with 32 SPUs Cell that no GPUs in the world can't do for games ? Or are you still wait and see like how dev approached PS3. The current Cell was like a solution looking for problems, instead of a solution to problems, which GPUs are.

& I'm not sure it's even remotely feasible to expect clock speed increases upwards of 3.2 GHz without compromising the convenience of cheap cooling solutions..

Well from the clock speed / power curve, on 45 nm the power don't ramp up as steep as it was on 90nm, so on 32nm, higher clock speed should be possible.

 

[Nesh]

Wii seems to do fine with their CPU. Look a single SPU is plenty to do that kind of calculation. PS Eye is like SD resolution, even if it goes HD next gen, a single SPU is still plenty to process the image. If they need processing power, they can put Cell into whatever peripherals. Cell will be cheap and low power by then.

Thats because the Wii isnt a good example. It does not aim for the high standards of the competition. It does fine in having games with motion controls and acceptable graphics. The market isnt nitpicky with Wii's visuals. They are with the PS3 and 360. And they will be a lot with PS4 too

 

[V3]

Thats because the Wii isnt a good example. It does not aim for the high standards of the competition. It does fine in having games with motion controls and acceptable graphics. The market isnt nitpicky with Wii's visuals. They are with the PS3 and 360. And they will be a lot with PS4 too

Wasn't my point, I was saying that Wii CPU function fine with it's motion control solution. Even MS thinks that Natal will do fine using Xbox 360 CPU. You don't need that much CPU for motion control solution.

 

[Shifty Geezer]

Wasn't my point, I was saying that Wii CPU function fine with it's motion control solution. Even MS thinks that Natal will do fine using Xbox 360 CPU. You don't need that much CPU for motion control solution.

That depends entirely on your choice of system. Image analysis can be extremely demanding.

 

[Nesh]

Wasn't my point, I was saying that Wii CPU function fine with it's motion control solution. Even MS thinks that Natal will do fine using Xbox 360 CPU. You don't need that much CPU for motion control solution.

It depends what you are trying to do.

 

[V3]

What exactly are devs going to do with image analysis for motion control that are going to require 8-16 SPUs ? I was into image analysis research some ten years ago, as far as I know face recognition and other sort of recognition was running alright on some old Pentium 3 or 4. Sure if you're running finger prints or other image indentification through large database, you could use more processors, but that's for traversing the database rather than the image analysis itself.

Anyway, it would be cheaper for Sony to license Natal, if their solution was going to require that kind of processing resources.

 

[Shifty Geezer]

What exactly are devs going to do with image analysis for motion control that are going to require 8-16 SPUs ?

Very little for console purposes. But there's a world of difference between 8 SPUs and Wii's Broadway! You can't point to Wii and say (or infer) its CPU is all that's needed. Optical background removal is an area that's very challenging, and I'm sure aiming for a better experience, a developer could gobble up most of Cell's clock-cycles. We have to make the distinction that there's different kinds of motion interface requiring different amounts of processing to pull off.

In the broader scope of the subject though, I agree with you that 8 SPUs should be enough for future workloads if the GPU is a good'un. If RSX was Xenos in PS3, what would devs be doing with Cell nowadays without the need to prop up the visuals to such an extent? I dare say they wouldn't be putting in as much effort to extract performance from Cell and letting cycles go idle.

 

[V3]

Very little for console purposes. But there's a world of difference between 8 SPUs and Wii's Broadway! You can't point to Wii and say (or infer) its CPU is all that's needed. Optical background removal is an area that's very challenging, and I'm sure aiming for a better experience, a developer could gobble up most of Cell's clock-cycles. We have to make the distinction that there's different kinds of motion interface requiring different amounts of processing to pull off.

My point was why choose motion control methods that will goble up your CPU ? Nintendo managed to do their motion control solution using Broadway. If your motion control is going to gobble up CPU, it's time to go to dedicated hardware built into the device.

If RSX was Xenos in PS3, what would devs be doing with Cell nowadays without the need to prop up the visuals to such an extent? I dare say they wouldn't be putting in as much effort to extract performance from Cell and letting cycles go idle.

Exactly my point. It is better to just give PS4 better GPU rather than a 32 SPUs Cell. Their choices for PS3 seems to create their own problems so Cell could solve them. It's very silly. My guess is they did not have the current PS3 design in mind when they started Cell. I think PS3 was supposed to only have 256 MB XDR with Cell and a successor to GS with eDRAM. Obviously that would have been more underpower than what PS3 is now. 32+ SPUs Cell will only make sense with that sort of configuration, not pair with today's or future GPUs, not for a console anyway.

 

[Shifty Geezer]

Their choices for PS3 seems to create their own problems so Cell could solve them.

I'm not so sure about that. PS3 seems to follow some ideals of PS2, with a very powerful, flexible CPU performing some fo the graphics tasks. the graphics problems exist and need processing to solve them, and Sony chose to deal with lots of that processing on the CPU rather than on the GPU. Taken as a whole system, the programmability of PS3's graphics is unequaled this gen. The problem is it needs developers to make use of it. In contrast, XB360's system hit a good flexibility with far less hassle, while PC's with mightier GPUs have far less flexibility, which devs don't care about as they don't want that much flexibility at the moment! They just want to render pretty graphics, and using the conventional methods makes that possible and easy on their hardware of choice.

PS3 isn't a bad design per se; just, relative to the approach of other systems, it's not a good fit for the current market.