The Future of the Playstation

Allthough relativly early in the production cyckle of this generations consoles and Sonys estimated 10 year actively developed lifespan, i was thinking of the investments of technology in the Cell Broadband Engine and related technology from the PS3 and how that would be used going forward within the brand.

Very early ideas for the PS3 were said to have been multiple ES+GS chips, Later it would become two Cells and from there to one Cell and RSX. Going forward Sony has said that they will use the experiance gained off course.

As a thought experiment based on near future and current technologies, would it be safe to assume that a future PS system would contain a developed Cell processor? I am assuming that a future PS product would be evolutionary instead of revolutionary to cut costs. Alot of developers also felt the PS2 coding knowledge learnt as a bit of a waste because it was not applicable to the PS3 or indeed any other system, so keeping and updating hardware to reduce learningcurve would be a popular move? A philosophical question would also be "how good is good enough" for the next generation when you consider what we can do with this one. If the PS3 was 100 times more powerful then the PS2, how manny more times powerful would the PS4 have to be to give devs what they need for the next gen?

Obviously i'm no hardware nor software engineer, but like most on these boards i'm interested in technology. So how do you see the next PS system working?
 
As a thought experiment based on near future and current technologies, would it be safe to assume that a future PS system would contain a developed Cell processor?
Yes, exactly for the points you've mentioned. The Cell architecture is scalable on multicore and multichip levels. PS4 will almost certainly take the Cell technology of then, likely the same structure 1 PPE : n SPEs though with perhaps greater double precision performance and a stronger PPE with OOO, and stick one or more in there. Existing Cell code should run directly, with suitably written code scaling across available cores. I think ease of development will be a key point next time around. The new paradigm had to come in eventually. Now it's started, Sony will want to keep everything straightforward. 'Write the same code you have been writing for a few years now, only you'll have more cores to play with.'
 
Cell is a platform. Cell was always as much about the PS4 and beyond than it was the PS3. What happens with the PS4 will really be determined by how the PS3 pans out, how silicon technology advances over the next couple years as well as the risk analysis of future silicon issues, and how well developers adapt to Cell in general. How Cell does in other markets may also have an impact on the Cell design in the PS4, as well as how NV GPUs evolve over the coming years. Fundamentally just as important is the continued development of memory technologies.

I think our crystal balls will be foggy for a while, depending on how well the PS3 does. Sony may cut it short and try to match, or even beat, MS to market with the PS4 which would significantly alter what the PS4 is. I also think Sony will be more price conscious next time.
 
I think our crystal balls will be foggy for a while, depending on how well the PS3 does. Sony may cut it short and try to match, or even beat, MS to market with the PS4 which would significantly alter what the PS4 is. I also think Sony will be more price conscious next time.
Whatever Sony's plans, Cell is adaptable enough that it makes sense we'll see it in PS4. If they want to push performance, they can stick two Cell2 1:32's in there. If they want to be cheaper and release earlier, they can take a small single Cell2 1:16. These choices should be implementable at the drop of a hat, so Sony could react to MS if they choose. Changing their mind between different Cell's in the same mobo architecture shouldn't be any problem at all, assuming the Cell's their using are commodity items. If they choose a custom Cell solution, that'll tie them into a design. For that reason I think it makes sense for them to stick with the standard architecture they've started. Accept Cell in whatever form it is being produced in a few years' time.
 
You say "how NV gpus evolve", Is there a contract between NV and Sony for the next Playstation i'm unaware of? I wonder if Sony will try and make its own GPU or if they will keep using modified mainstream GPUs, I'd imagine the costs might be prohibitative for such an endevour but you never know. Bandwidth being a prime concern in consoles, do you think we would see less typical memory and more flash based memory in the future. Not being very technical i remember reading (as an example) about Vista readyboost, which uses a USB device because it is faster then ordinary ram. Could we see a similer development in the next generation consoles? Seeing as costs are coming down rapidly.
 
nVidia and Sony are supposed to be in cahoots. nVidia spoke up the relationship, and are IIRC said to be contracted into another Sony device which we're mostly assuming is PSP2. nVidia also said they 'like where Sony is going' and it matches their idea of the 'the future'. Could that be a Cell-like processor that incorporates GPU type functions? Could be. Thus there may be a Cell that includes PPE, SPE and GPE, with the Graphical Processing Element having the sort of memory subsystem needed to handle texture data in the way Cell can't. Though GPE would be a misnomer as, ideally, all cores could be set to task with whatever they're dealt with, such as SPE being vertex + geometry shaders. Fundamentally you could just distinguish between cores by data formats. Erratic data accessing wants an OOO core, SIMD stuff works with SPEs and LS, and broad random-access data structures want a GPU like system.
 
It would certainly be interesting to see such a system, i like the fact that most consoles keep getting better looking games as they progress and the hardware is better utilized. A Cell-GPGPU (if you can call it such) sounds like it would give devs quiet a bit more legroom to do what they please, as seen on Cell taking on certain graphics processing elements on SPEs to help RSX.
 
nVidia and Sony are supposed to be in cahoots. nVidia spoke up the relationship, and are IIRC said to be contracted into another Sony device which we're mostly assuming is PSP2. nVidia also said they 'like where Sony is going' and it matches their idea of the 'the future'. Could that be a Cell-like processor that incorporates GPU type functions? Could be. Thus there may be a Cell that includes PPE, SPE and GPE, with the Graphical Processing Element having the sort of memory subsystem needed to handle texture data in the way Cell can't. Though GPE would be a misnomer as, ideally, all cores could be set to task with whatever they're dealt with, such as SPE being vertex + geometry shaders. Fundamentally you could just distinguish between cores by data formats. Erratic data accessing wants an OOO core, SIMD stuff works with SPEs and LS, and broad random-access data structures want a GPU like system.

If that were the case, what would traditional GPU's play in console's architecture?
 
If that were the case, what role would traditional GPU's play in console's architecture instead of primary being used for graphics ?
 
If that were the case, it wouldn't! The idea would be to take the silicon budget of discrete CPU and GPU and combine them into a shared pool of processing that can handle any task. Given the talk from nVidia, it's my gut feeling that that's what they're looking to do. And it'd be a good move against ATi too. nVidia could do with being inside a one-stop solution, and a Cell with graphics built in is probably as ideal solution as nVidia can hope for.
 
I wonder if their work on CUDA might be preperation for this in the longterm, adding a layer for programers to use in order to get more out of idle parts of the "gpu" in the Cell. Would be an interesting solution not only for that but also to free up SPEs, or alliviate them, when doing physics for example.

Again though i'm being very interested in the memory architecture they'd use, if flash based memory is so fast then perhaps they'd use their own MemoryStick architecure for PS4; just kidding, but an interesting idea for them to lessen their dependance on other companies IPs and licensing costs.

It's my understanding that Sony is generally a little uneasy about using other makers parts, unless they are close.
 
Yes, exactly for the points you've mentioned. The Cell architecture is scalable on multicore and multichip levels. PS4 will almost certainly take the Cell technology of then, likely the same structure 1 PPE : n SPEs though with perhaps greater double precision performance and a stronger PPE with OOO, and stick one or more in there. Existing Cell code should run directly, with suitably written code scaling across available cores. I think ease of development will be a key point next time around. The new paradigm had to come in eventually. Now it's started, Sony will want to keep everything straightforward. 'Write the same code you have been writing for a few years now, only you'll have more cores to play with.'


There is already a 32SPE / 2PPE Cell on IBM's proposed roadmap for 2009/10. This sounds like a candidate for the PS4. I've no idea what they'll do for the PPE but if they want to beef it up a cut down POWER6 core should be just the job, I don't see OOO going back in. 45nm with highK should cut it's power consumption by more than half and simplification will reduce this further (I don't think a console really needs mainframe grade reliability features).


I actually wrote a paper on what I thought a future PS4 should look like
http://www.blachford.info/stuff/PS4.pdf here. I suggested the SPEs be modified with sub cores dedicated to raytracing (these were based on existing raytracing hardware). I have evolved the idea since then until with the sub cores looking a lot closer to GPUs with extra bits tagged on.

With Intel pushing raytracing so much these days I can see future consoles also being based on raytracing. Quite whether it'll be the PS4 generation I don't know.
 
What I find slightly funny is that virtual the same hypothetical CPU with, say, 4 PPUs and 16 or 24 SPUs can be used in both a hypothetical PS4 and Xbox3, emulating with little effort their respective predecessor's CPUs. Will IBM go that route?
 
Not being very technical i remember reading (as an example) about Vista readyboost, which uses a USB device because it is faster then ordinary ram. Could we see a similer development in the next generation consoles? Seeing as costs are coming down rapidly.

Again though i'm being very interested in the memory architecture they'd use, if flash based memory is so fast then perhaps they'd use their own MemoryStick architecure for PS4; just kidding, but an interesting idea for them to lessen their dependance on other companies IPs and licensing costs.

I think you're a bit confused about the speed of flash RAM... It's "faster" (i.e. lower latency access) than magnetic storage (i.e. hard drives) however the throughput is much lower, and it's nowhere near as fast as the DRAM used for system memory in either case.
 
If that were the case, it wouldn't! The idea would be to take the silicon budget of discrete CPU and GPU and combine them into a shared pool of processing that can handle any task. Given the talk from nVidia, it's my gut feeling that that's what they're looking to do. And it'd be a good move against ATi too. nVidia could do with being inside a one-stop solution, and a Cell with graphics built in is probably as ideal solution as nVidia can hope for.

So, multiple cells that can handle graphics. Its sounds way out there but at the same time it makes a lot of sense. Sony could make PS4 cheaper than its cost for PS3.
 
With Intel pushing raytracing so much these days I can see future consoles also being based on raytracing. Quite whether it'll be the PS4 generation I don't know.
Intel can push RT as much as they want, this doesn't mean we are necessarily going to use it.
I can certainly see a future where RT and rasterization work together, though I still don't get why we should completely remove rasterization and replace with it RT.
 
So, multiple cells that can handle graphics. Its sounds way out there but at the same time it makes a lot of sense. Sony could make PS4 cheaper than its cost for PS3.

Sony and Toshiba did consider it while CELL was still in the R&D labs, but they quickly understood it would not have worked. I doubt that nVIDIA and ATI will stand still till PS4 is ready and waste chances to evolve their offerings.

To make a long story short, I do not think that packing as many PPE's and SPE's as needed to match a more conservative CELL chip paired with a discreet GPU would end up being the cheaper alternative for Sony, even in 2011-2012. Maybe I lack vision, but I really do not see full software rendering based on CELLv2 or using a CELL-optimized-for-graphics chip is the cheapest solution Sony can take R&D wise.

I think Sony is out of the "GPU race", I agree with nVIDIA's boss that PSP's GPU is probably the last custom Graphics Processor SCE is going to design fully in-house. Especially given how tighter and tighter the R&D budget is going to be on future PlayStation projects. While I do think that they will still design their projects with "hardware loss at launch and progressive and aggressive cost reduction as time goes by" (especially PSP2 where the semiconductor scaling will be less of an issue: I seriously doubt they will go with more advanced manufacturing technologies than 65 nm with PSP2: tried and perfected manufacturing process == cheaper manufacturing process), I do not think that they will want to take as big of a loss per unit sold at launch as they have done in the past.

Especially if they manage, like they do want, to establish Blu-Ray in the market well enough to make it a worthwhile inclusion in PS4: a good Blu-Ray drive should not have nearly the same impact to launch date and per unit losses as what we have witnessed with PS3. It will be like including a DVD drive this generation. I also think we will not see an HDD in PS4, but a large, and user-upgradeable Solid State Storage solution (you would be able to add more memory through USB Flash drives, or other pluggable Flash/Solid State memory solutions... I say flash, but it could be other solid state technologies as well). We have seen 32+ GB Flash HDD's last year and with the pace Solid State memories have been progressing at I do not see 100-120 GB being impossible to wish for by 2011-2012 and not incredibly expensive either. The downside is that with HD video, being able to use your console as PVR, downloading games and large high quality movie content from some online stores would fill up those 100-120 GB raher quickly.

I personally like the idea of having a 2.5'' HDD slot and the possibility of changing the HDD yourself thus enabling you to quickly upgrade your PS3's HDD capacity. Maybe Sony will decide to offer two SKU's again. One with Solid State memory and another one with a 0.5-1 TB HDD (you can buy 1 TB external/USB 2.0 based HDD's now for less than $300).
 
Last edited by a moderator:
Sony and Toshiba did consider it while CELL was still in the R&D labs, but they quickly understood it would not have worked. I doubt that nVIDIA and ATI will stand still till PS4 is ready and waste chances to evolve their offerings.

To make a long story short, I do not think that packing as many PPE's and SPE's as needed to match a more conservative CELL chip paired with a discreet GPU would end up being the cheaper alternative for Sony, even in 2011-2012. Maybe I lack vision, but I really do not see full software rendering based on CELLv2 or using a CELL-optimized-for-graphics chip s the best route Sony can take.

Ok then, but you do believe that Sony will optimized Cell processors to handle more of the graphic workload even if their is a true GPU in PS4?
 
Back
Top