PS4 to be based on Cell?

[Gubbi]

One route for Sony would be to reduce the emphasis on SPEs. Produce a CELL chip with 8 SPEs, but stick 4 good solid OOOe cores in there instead of the PPU. The SPEs are small dense cores, so wouldn't add too much deadweight, - and they might even be useful for a whole bunch of tasks that would be hard for a CUDA (or similar) enabled GPU.

Backwards compatibility and instant developer friendly hardware.

Cheers

 

[Paul]

Few things

It would be absolutely suicide for Sony to release a PS3 3.5 in 2010 or 2011. Expect to see a PS4 no earlier than 2012 with a most robust Cell processor(1TFLOPS+), an expanded GPU which I think will be a more custom SCE and nVIDIA built with cell elements, not just an off the shelf rip off this time around.

Blu-ray will still be there, though they may use higher capacity disks?

Expect the thing to have some Wii elements as well, target price should be 349 dollars

 

[wco81]

If they launch early, it would be more expensive to implement BC if they don't use nVidia for GPU?

Yet, if they launch early, BC might be more important to prod users to upgrade ...

 

[Tahir2]

I am hoping that PS4 is based around Cell and will have a more robust GPU that is designed around the needs of a console rather than sticking an off-the-shelf GPU that has limited the PS3 to a degree.

My speculation would be an increase in SPE's and a slight bump in clock frequency and a GPU that from either NVIDIA or AMD - don't see Sony designing a custom unit at all.

 

[nAo]

FirewalkR: CUDA is not the only reason. Having two massively parallel architectures on the same system means that you have to learn how to use both of them and you need two advanced compilers/debuggers and possibly APIs. Also inter-communication between these two ICs might be problematic, to not mention costs and power consumption.

I'm sure Sony and MS want to drive costs down, so I won't expect bitchin'-fast next gen consoles. They need something reasonably faster, simpler and easier to use.

To me CELL merits are not so clear on a platform that will (probably) have a many-cores architecture designed to be first and foremost efficient at graphics and physics. My base assumption is that next (next) gen GPUs will be all LRB-like, perhaps not as flexible as LRB, but good enough for a lot of non-graphics related work.

The second option you mention sounds better to me; a relatively small OOOE CPU with 2 or 4 cores to primarily drive gameplay code and/or all non graphics related code on applications that don't need a massive amount of computations.

 

[INKster]

Well, i'm throwing my spoon into the pot too, just for the fun of it .

- Launch sometime in 2010~2011.
- 32nm CELL at release, relatively moderate clockspeed increases from today's 3.2GHz to, say, 4~5 GHz.
- 2~4 GB high-speed GDDR5 (across a 256bit bus shared between CELL and the GPU).
- Mechanical hard drive (SSD sounds nice, but its gigabyte-per-dollar value will likely be too low for a games console, even then).
- Blu-ray drive, of course (Sony has too much at stake here to just drop optical drives altogether like Phil Harrison had predicted).
- Much smaller footprint (the current PS3 is just too bulky and heavy to go against a simple design like the Wii -and that one can shrink too-, even with the rumored future model "on a diet").

What do you think ?

 

[ArturNOW]

I had no doubts from the beginning i.e. where SONY was talking about CELL in PS3. I'm rather wondering how many CELL processors will PS4 contain. SONY used RSX just to ease programmers live, PS4 might be solely CELL base, but there can be a problem with backward compatibility so they might live RSX which won't cost much by 2011-2012 (rather 2012). I think PS3 is great platform to let developer learn CELL architecture. If there are many studios that are familiar with CELL (remember that CELL is supported by universities) PS4 won't have problems known from first years of PS3.

 

[eastmen]

Would there be a benfit of using 3 chips in the system. 1 cpu which is just an enhanced cell chip. This would be used for normal cpu things , a.i and what not. A ppu that would be just for physics and then a graphical processer ?

Even just taking an off the shelf 2010 gpu (assuming a 2011 release) will give you many times the power of the ps3. The gpu in the ps3 is 7800gtx levels or lower. Even todays gpus like the radeon 4870 and the gtx 260 run circles around it. Adding in 2-4 gigs of ram would also allow them to create levels and models far beyond whats in game today. The only thing left to really enhance would be the pyhsics. I'd love to play a map where I can actualyl destroy everything in teh game and that assigning objects real propertys isn't just a few items here and there. For instance cover systems need a boost. using a ppu you should be able to assign the correct propertys to everything. That way you don't get a fence that you can shoot through but a wood door you can't. If the bullet has the proper physics to pierce it then it would do so.

I think thats far more important than just graphics that will naturely get improved.

 

[scificube]

I haven't seen any indication that next (next) GPUs will be available in 2011. I also don't know how much next (next) GPUs are going to be like LRB given what we've heard about DX11. They will be more flexible but "how" that is accomplished doesn't sound as if it would be in the same manner at all as of yet.

So long as there is a discrete CPU and a discrete GPU the interconnect issue persists regardless.

As long as a discrete CPU and GPU persists different compilers, debuggers, etc. will also. This is now becoming a major problem? It to me seems a single arch design seems more of an opportunity than a cure to some major affliction that exists now. Perhaps some will see that as an argument without a distinction but I do.

The likelyhood of an OOOe CPU from either MS or Sony is very low especially one with only a few cores. It's leaves them both quite vulnerable to being 1-uped with ease on the spec and by extension to suffer a great deal of marketing malign from the other vendor.

I still see a discrete CPU and a discrete AMD/Nvidia GPU in both machines. I see the GPU being used to pump out graphics as graphics sell instead of being hitched with other extraneous tasks. If one is not using the GPU for graphics somebody else *is*.

 

[rekator]

nAo, I'm thinking you're right with the statement of "the drive costs down", a other >400$/€ price would probably not desire by publishers and naturally by us, the customers.
From a dev POV, you're thinking that a few multi-core OOOE CPU is more suitable than a massive in-order multi-core (with my old little coding experience I'm can understood that), but that did you think of a hypothetical LRB-like CPU tale for console?
And for LRB, are we go with the same dual API's problems you mentioned for the CELL+ and next gen GPU, or the x86 heritage is more friendly?
I'm also interest to known that is the "ideal" design CPU+GPU from a IA and Physic POV. Personally, I'm hoping a big increase in these two sectors for the next gen, despite the fact that a lot of increase would probably more come from the software than the hardware.
Thanks for your time.

 

[FirewalkR]

To me CELL merits are not so clear on a platform that will (probably) have a many-cores architecture designed to be first and foremost efficient at graphics and physics. My base assumption is that next (next) gen GPUs will be all LRB-like, perhaps not as flexible as LRB, but good enough for a lot of non-graphics related work.

The second option you mention sounds better to me; a relatively small OOOE CPU with 2 or 4 cores to primarily drive gameplay code and/or all non graphics related code on applications that don't need a massive amount of computations.

The problem is... where does all the research on the Cell go to? The supercomputer market exclusively? The trash can? It seems such a waste.. Perhaps Sony could "go nuts" and add both SPEs and functionality to each SPE so it could go up against regular GPUs or LRB, but it would be risky and could go terribly wrong.

What about integrated CPU and GPU? A bit like the Cell but instead of a PPE have several OOOE cores and a GPU instead of the SPEs, for physics+graphics. I think I read about AMD going in this direction, probably here at B3D...

Or they could get the general-purpose cores, plus some SPEs, plus the GPU all in one chip. And then they'd watch programmers all over the world commit suicide.

 

[Shifty Geezer]

[modhat]I don't really understand why this thread exists when it's covering everything being talked about in the next-gen console thread. Unless someone can explain the difference, the threads should be merged IMO.[/modhat]

 

[scificube]

Would there be a benfit of using 3 chips in the system. 1 cpu which is just an enhanced cell chip. This would be used for normal cpu things , a.i and what not. A ppu that would be just for physics and then a graphical processer ?

Even just taking an off the shelf 2010 gpu (assuming a 2011 release) will give you many times the power of the ps3. The gpu in the ps3 is 7800gtx levels or lower. Even todays gpus like the radeon 4870 and the gtx 260 run circles around it. Adding in 2-4 gigs of ram would also allow them to create levels and models far beyond whats in game today. The only thing left to really enhance would be the pyhsics. I'd love to play a map where I can actualyl destroy everything in teh game and that assigning objects real propertys isn't just a few items here and there. For instance cover systems need a boost. using a ppu you should be able to assign the correct propertys to everything. That way you don't get a fence that you can shoot through but a wood door you can't. If the bullet has the proper physics to pierce it then it would do so.

I think thats far more important than just graphics that will naturely get improved.

For all intents and purposes the PS3 is already a 3 chip design...if you look at Cell from arch perspective it's a heterogeneous core with 2 distinct architectures.

Sony needs to make SPUs more friendly without nerfing them and have a PPU that is "matched" to the capability of the SPUs. The PPU is...well...very easily improved upon.

I guess I am saying the system needs to be "balanced" better AND be easier to wrangle with.

Grabbing an off the shelf GPU is dangerous because it will run hot, and it's memory interface will complicate the motherboard. Something will have to be done to negate the affect of an 384bit to 512bit bus to memory. I wouldn't be surprised to see eDram appear again to handle framebuffer access. With the exception that it would be integrated with the GPU from the start and the GPU would have a separate pool of memory over a <= 256bit bus for textures etc that is not shared with the CPU. That would place two easier to implement and discrete buses on the mobo that have a better chance of costing less to produce over time.

edit:

Actually another fast bus would be needed so that the GPU and CPU can communicate effectively. That may be difficult to swallow so a "unified" mem pool for the GPU and CPU becomes attractive again for better or worse. That would solely be based off just how fast the bus is.

end edit:

Then again recently I've heard some concerted hate for eDram. I wonder what other solution there is to providing these GPUs the memory peformance they will demand.

 

[Mobius1aic]

Why the rage against Larrabee? Anyway it would definitely *not* be that chip in this scenario, for being a more acute reason of why nAo laments the combination to a 2011 GPU; they are treading on each others toes if extrapolated along their present trajectories, which means API hassles and/or wasted transistors.

To do it right, Cell should be slightly evolved and have the computational power scaled via clock and core, and the GPU should be designed to complement it; in 2011 that would be a very custom job, and is NVidia willing to take the time/effort to do that? It's not been their pattern, and unless it were to change, well... the result would be inelegant.



I think if you ask many Wii owners whether it is underpowered or not, they would say 'no.' That view is more held by those that resent the system than those that actually own/enjoy it... and that later group is the majority, for better or worse.

Consumer expectations still rise. Sure those who are won over by the Wii are still likely to shell out the money for a new system from Nintendo rather than for MS or Sony due to consumer confidence in the Wii product line. However, I think the next Nintendo system can't just be a barely upgraded architecture with clock speed increases (although we know the real upgrade from GC to Wii is the RAM system ). The next Nintendo system will be expected to be HD, 720p minimum. Sure it doesn't have to be PS4/XBox 720 level quality but it'll definitely be a new system throughout, not minimum hardware increase, which means a new array of components, CPU, GPU, RAM, etc. Whether the next Nintendo system releases upon the usual console cycle or after an extended life for the Wii, consumers will expect something better than what they have now. Because of the high penetration of the Wii, Sony and MS are paying attention - close attention. On the Larrabee comment, I just said goddamn because I like too

Back to the topic. Yes, Cell will have to scale to similar power usage into a new version with more cores/SPEs. It's the current nature of processor design, x86 or otherwise.

 

[dobwal]

Would there be a benfit of using 3 chips in the system. 1 cpu which is just an enhanced cell chip. This would be used for normal cpu things , a.i and what not. A ppu that would be just for physics and then a graphical processer ?

Even just taking an off the shelf 2010 gpu (assuming a 2011 release) will give you many times the power of the ps3. The gpu in the ps3 is 7800gtx levels or lower. Even todays gpus like the radeon 4870 and the gtx 260 run circles around it. Adding in 2-4 gigs of ram would also allow them to create levels and models far beyond whats in game today. The only thing left to really enhance would be the pyhsics. I'd love to play a map where I can actualyl destroy everything in teh game and that assigning objects real propertys isn't just a few items here and there. For instance cover systems need a boost. using a ppu you should be able to assign the correct propertys to everything. That way you don't get a fence that you can shoot through but a wood door you can't. If the bullet has the proper physics to pierce it then it would do so.

I think thats far more important than just graphics that will naturely get improved.

If you have an enhanced Cell with a more robust "off the shelf" GPU why would you need a PPU. You basically have two processor that will be well suited to handle physic, so a third processor dedicated to physics would seem redundant and rather costly to the overall BOM of the console.

 

[V3]

[modhat]I don't really understand why this thread exists when it's covering everything being talked about in the next-gen console thread. Unless someone can explain the difference, the threads should be merged IMO.[/modhat]

Its about the merit and rumour of Cell in PS4. The other thread is speculation thread.

 

[mrcorbo]

To me CELL merits are not so clear on a platform that will (probably) have a many-cores architecture designed to be first and foremost efficient at graphics and physics. My base assumption is that next (next) gen GPUs will be all LRB-like, perhaps not as flexible as LRB, but good enough for a lot of non-graphics related work.

The second option you mention sounds better to me; a relatively small OOOE CPU with 2 or 4 cores to primarily drive gameplay code and/or all non graphics related code on applications that don't need a massive amount of computations.

So, in short, you're saying the need to have a GPU on board, which if it is based on where PC GPU designs are today and appear to be heading is going to be based around massively parralell stream processing basically makes Cell redundant in a future console.

Is there any scenario where going their own way on the GPU design and further developing Cell makes sense for Sony, or is it just throwing more money and resources at a dead-end?

 

[Crossbar]

I'd like to meet these developers Perhaps you wanted to write "restrictions of IOE"..

Yep that was my intention but I somehow failed mid-sentence......

Anyway going back to the original rumour. If Sony is asking game developers for input now I find it pretty hard to believe that they will have something ready for the market by holiday 2011 considering how long time it takes to go from start of development to first silicon and from first silicon to mass production of the IC and then move on to mass production of the complete console. 2012 maybe.

Here are some neat illustrations from the Goto article that can be helpful when predicting the next Cell:

 

[Carl B]

Anyway going back to the original rumour. If Sony is asking game developers for input now I find it pretty hard to believe that they will have something ready for the market by holiday 2011 considering how long time it takes to go from start of development to first silicon and from first silicon to mass production of the IC and then move on to mass production of the complete console. 2012 maybe.

If Sony really wanted to, 2011 is not a problem. Cell at 32nm (and thus a true revision) should be doable by then, if they went Intel or AMD, well... it's really just Sony and either of them coordinating, and if they went a new design out of IBM, consider how quickly the XeCPU came together. So 2011 is very doable, it's just a matter of what their actual internal intentions, constraints, and goals are.

 

[Shifty Geezer]

Then again recently I've heard some concerted hate for eDram. I wonder what other solution there is to providing these GPUs the memory peformance they will demand.

TB/s RAMBUS RAM. TBDR. With a fixed 'low resolution' of 1080p, BW shouldn't be an issue next-gen.

Its about the merit and rumour of Cell in PS4. The other thread is speculation thread.

But the actual content here ignoring title is the whole system including GPU talk, alternative CPU architectures, etc. A fair bit of which has already been had in the other thread.