The Future of the Playstation

[nAo]

You are right, PS4 needs to be the first console pushing Renderman for Games .

Well..do we really need Renderman in games? What would it buy to us? Better DOF, AA and motion blur for sure, but at huge costs (to composite and resolve every element in the scene per frame would take a tremendous amount of bandwidth). Maybe the main advantage would be to completely doucple fragments shading frequency and fragments sampling frequency, but I can see that happening with rasterizer as well (deferred renderers can do that in theory, and look at this evil evil work: http://developer.download.nvidia.com/presentations/2007/siggraph/stencil_routed_a-Buffer_sigg07.ppt )
Do we really want an hardware implementation of Renderman to render images for interactive/real time applications? I'm not sure about that..
Marco

 

[Panajev2001a]

Well..do we really need Renderman in games? What would it buy to us? Better DOF, AA and motion blur for sure, but at huge costs (to composite and resolve every element in the scene per frame would take a tremendous amount of bandwidth). Maybe the main advantage would be to completely doucple fragments shading frequency and fragments sampling frequency, but I can see that happening with rasterizer as well (deferred renderers can do that in theory, and look at this evil evil work: http://developer.download.nvidia.com/presentations/2007/siggraph/stencil_routed_a-Buffer_sigg07.ppt )
Do we really want an hardware implementation of Renderman to render images for interactive/real time applications? I'm not sure about that..
Marco

Marco, after all these years you still take that Micropolygons comment from me seriously ?

.

Still, I am glad you take the time to write a thought out response (thank you) . I love hearing pro's opinions on the subject.

I like Renderman for how it jsut seem to "make sense" and how clean it appears from a 3000 ft perspective: everything before a certain phase is micro-polygons and everything after a certain stage is sub-pixel samples to be resolved. Each stage is decoupled from the other and in each stage you can approach things in an uniform way.

I think that people working right now on PS4 and Xbox 720 will be toying with the whole micro-polygons+stochastic AA (along with some RT for shadows, reflections, refractions, and light transfer computations) idea while working things along: you have to admit that it is lovely to read the REYES SIGGRAPH paper .

Still, they will know that it is not the best and most R&D budget efficient solution for what they want to achieve and they will go for more complex and less "clean" solution. On that end it is almost sad to hear Carmack's explaining (Quakecon 2007's Q&A) why he decided to abandon unified dynamic lighting and shadowing he worked so hard to achieve with Tech 4 for a non unified solution full of pre-computed light and shadow maps mixed with real-time lighting and shadowing for Tech 5 games.





P.S.: edit: thanks for the paper, it is very interesting and it seems to work with current HW too (not on the PC side where they use D3D10 calls): it seems something that can be don on RSX and Xenos (I am referring to Slide 10 and Slide 16: both console GPU's seem able to turn off multisampling and render in a MSAA buffer as well as using a variety of functions not exposed under stock DX9... isn't it what games like Lost Planet and some PS3 titles do for some of their particle rendering systems ?)

what do you think ?

 

[Panajev2001a]

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?

I think CELL can already handle graphics workloads quite well right now (it is just not specifically designed to excel at it like RSX is) so I doubt that a future CELLv2 will have problems working together with a discreet GPU.

CELLv2 will have its own evolution path decided by the IBM-Sony-Toshiba group according to the need of all three companies. I do not think highly parallel and performance bound tasks such as 3D graphics will lack from the simulations they will run while developing the evolution of the CELL architecture.

For sure, developers will be encouraged more and more to think of SPU's as a virtualized resource and not to think of their engines as system made of a fixed number of pieces that need to be mapped on physical SPE's.

IMHO, if you start thinking of SPE's as a sea of resources you will do yourself a favour in preparation to PS4.

Of course right now you also have to worry about using optimally the four rings in the EIB inside CELL (and thus worry about which physical SPU does what) to allow maximum bandwidth utilization and even though it is not an easy thing to track things down to this level for large projects it will be something people will try to do as much as they can (Faf, crazy recursive DMA chains guy, I am looking also at you )

(http://www.ibm.com/developerworks/power/library/pa-expert9/#main)

but it is not like people programming for massively multi-CPU's systems do not worry about what processing node does what and near which node data is being stored when developing their code.

 

[weaksauce]

It will have a "Cell"-like CPU system, but with much more cores.

Guess it depends on who's in charge of the project. If it would be Kutaragi it would have to be 1000 times faster.

But c'mon, cell is gonna be old as hell at the time. Lets hope they'll use optic processors or something. Is that the only new processor tech underway, accept maybe quantum computers?

 

[Npl]

But c'mon, cell is gonna be old as hell at the time. Lets hope they'll use optic processors or something. Is that the only new processor tech underway, accept maybe quantum computers?

we may see optic interconnects between Chips, but optic computing is still ways off. And quantum computing is way to specialised to find its way into Home-Computers/Consoles (alone), its magnitudes faster for a set of problems but unusable for all others.

I dont see why a Cell2 wouldnt be very competive in 5-7 years time. I read somewhere that the sweetspot for the LS was around 4MB per SPU, based on IBMs simulation, so im pretty sure we`ll see more LS and likely more SPUs.

 

[joker454]

I also think we will not see an HDD in PS4, but a large, and user-upgradeable Solid State Storage solution

I'd be shocked if their next box didn't include a hard drive standard. Playstation is as much about gaming as it as about Sony controlling the users tv, basically being a media hub. That needs tons of storage for the obvious reasons, but also so that Sony can sell the user stuff. The more storage space the user has, the more stuff they can buy. Plus, you'd think they would need at least 80gb of storage on their next machine to make migrating everything over from a PS3 possible. I can't imagine 80gb of solid state memory being cost effective, when a 300gb 2.5" drive will probably be dirt cheap by 2012.

 

[Panajev2001a]

I'd be shocked if their next box didn't include a hard drive standard. Playstation is as much about gaming as it as about Sony controlling the users tv, basically being a media hub. That needs tons of storage for the obvious reasons, but also so that Sony can sell the user stuff. The more storage space the user has, the more stuff they can buy. Plus, you'd think they would need at least 80gb of storage on their next machine to make migrating everything over from a PS3 possible. I can't imagine 80gb of solid state memory being cost effective, when a 300gb 2.5" drive will probably be dirt cheap by 2012.

You are right, but if Solid State storage keeps dropping in price per GB as it has been doing so far, I think you will see quite big Solid State HDD's come 2012.

On the other hand, my preferred road would be for them to have once again a removable drive like they had with PS3 yet giving a large HDD in the default model.

You are right about the HDD prices too: http://www.newegg.com/Store/SubCategory.aspx?SubCategory=14&name=Internal-Hard-Drives

750 GB for $200, today.

 

[AlphaWolf]

You are right, but if Solid State storage keeps dropping in price per GB as it has been doing so far, I think you will see quite big Solid State HDD's come 2012.

I don't think solid state will have come close to being more cost efficient than magnetic disks by 2012, if they are going to include storage, its probably going to be in the form of a magnetic disk HDD. There's also always some interesting optical devices lurking but I don't see any of them becoming reality in 5 years.

 

[Shifty Geezer]

In terms of mass economy, I don't think flash will have caught up with HDDs. But at a minimum cost, they'll be cheaper. If wanted, the entry level console could come with $5 worth of flash storage, 32GB or whatever that is, with the more media-centric version having a minimum price $20 HDD of 180GB or whatever the bottom is then. This assumes multiple SKUs will remain in vogue, with a broad range of features in the machines but a target at the bottom end price of primarily a games machine with download content.

 

[Crossbar]

I read somewhere that the sweetspot for the LS was around 4MB per SPU, based on IBMs simulation, so im pretty sure we`ll see more LS and likely more SPUs.

Do you have a link? I thought I read somewhere the sweetspot was estimated ton 1 MB, but my memory may have failed me and I could not find any link to back it up.

All I found was this old nice article, which describes how the LS ended up to be 256 kB in the current versione of Cell.

We increased the size of the local store twice: first, in the initial concept phases from 64 KB to 128 KB, and later, when it was doubled again to 256 KB. In both of these cases, programmability was the driving factor.

 

[wco81]

Is Kutaragi out of the picture for the next PS?

If so, how will that affect the design?

 

[Cheezdoodles]

Is Kutaragi out of the picture for the next PS?

Kutaragi basically got fired back in april. Well, they forced him to retire anyway..

If so, how will that affect the design?

It means the engineers that design this thing will not have to ask themselves questions like: "WTF did he mean by 4d capability???"

 

[Blackraven]

You are right, but if Solid State storage keeps dropping in price per GB as it has been doing so far, I think you will see quite big Solid State HDD's come 2012.

On the other hand, my preferred road would be for them to have once again a removable drive like they had with PS3 yet giving a large HDD in the default model.

You are right about the HDD prices too: http://www.newegg.com/Store/SubCategory.aspx?SubCategory=14&name=Internal-Hard-Drives

750 GB for $200, today.

I am with you on the SSD thing.

I too think that PS4 SHOULD use Solid State Drives instead of magnetic HDD storage.

Projections (from Samsung especially) seem to indicate that if this year, 1x SDD = 5x HDD, then by year 2010, 1x SDD = 3x HDD. That rate should continue to decrease as well by 2012 (potential launch date for PS4) and that's when Sony can make their move.

The start-up time and boot times will improve dramatically and loading times will be greatly REDUCED. Once more, it doesn't have to go through an drive bay parking proceedure that is used on magnetic HDDs so when you shut off a PS4, it would only take less than a second (from the time when standby/off button is pressed).

This HDD parking-proceedure costs 5-10 extra seconds on the PS3 (which is slower than the shut-off time on a PS2, and OLDER-GEN CONSOLE) which gives the notion that a newer gen console powers-down longer than an older-gen. If an SSD were to be used as standard feature on the PS4, then you can expect fast shut-off (as fast as a PS2 or faster).

In a power outage, those extra seconds could mean longer lifespan for your console and lesser repairs (due to improper shut-down ; HDD parking failure, etc.)

So that's one technical advantage that a standard SSD would offer for the PS4 when it launches in 2012-2013 (over magnetic HDD)

 

[kimg]

Well, the consoles need to keep up with the PC devlopment, right?

And as it look now, the PC's will start use some of the trademarks of old console tech: integrated multifunctionality.

As in, general purpose gpu's who is programable, maybe GPU and CPU in same chips for PC, with a lot of hybrid processors serving as both CPU and GPU's, etc.

If this comes for PC, and the memory acces for it is not stopped by puny local storages, would not a next gen playstation need to keep up? would LS in SPU's really keep up with a, say, next gen GPGPU shader array in a integrated gpu inside the multicore CPU in a pc?

Of couse this would probably be a lowend GPU solution for the PC marked, but the tech could, if something as this come, and GPU's turn to, lets imagine, a co processor plugged into a motherboard socket, it would use the console's speed advantage in integrated construction(for the price of less flexibility in choosing gpu/cpu combos), but with upgradable RAM.

Im not a programmer, but i keep imagining LS in SPU's will be a problem when games goes beyond 10+ cores someplace in future, with future's game engines.

 

[Crossbar]

If this comes for PC, and the memory acces for it is not stopped by puny local storages, would not a next gen playstation need to keep up? would LS in SPU's really keep up with a, say, next gen GPGPU shader array in a integrated gpu inside the multicore CPU in a pc?
....
Im not a programmer, but i keep imagining LS in SPU's will be a problem when games goes beyond 10+ cores someplace in future, with future's game engines.

I think you got things backward. The local storage (LS) attached to each SPU is just like a software controlled cache memory.

Your comments about the LS do not make any sense. Actually I could not make head or tail of much of what you wrote.

 

[Npl]

Do you have a link? I thought I read somewhere the sweetspot was estimated ton 1 MB, but my memory may have failed me and I could not find any link to back it up.

Im certain the value was 4MB, but I cant vouch for the exact wording. It was not the "optimum" in terms of performance per mm², but rather IBM did run simulations for Cells target applications and AFAIR 4MB was the spot where more memory wouldnt improve performance much - purely based on 1 SPU.
If you think of Cell as whole, aparrently more SPUs with less LS are the better choice and you have to put "sweetspots" in context.

Would like to find this article myself, cant find it anymore.

 

[Color me Dan]

Is Kutaragi out of the picture for the next PS?

If so, how will that affect the design?

I'd think the next Playstation will be more "cost effective", Kutaragi had vision that often surpassed what was acceptable economically. But his investment in new tech this gen will probably make the next much more profitable. Looking at the modular design of the PS3 it wouldn't be hard to imagine it scaled up and still being acceptably good. I think somewhere Kutaragi knew this, but thats just me.

How would a combined CBE-Cell work? Would Sony and nVidia pool designers to create a design or would there be some form of split, making two parts and then joining them with some form of commonly agreed interface in silicon?

 

[pc999]

One should think in a few things, before think in pure raw tech.

I mean till today techprice alone would make the design of the console, but that is/will not be true again because of several points:

1- Production cost of the games, what is the point of having 20x more power if the devs cant tap it or just produce enought content to it.
2- There will be less chances of reducing costs, i.e. if they release it in the 32nm there will not be many things they can do fast to reduce costs.
3- Market evolution, I trully belive that after Wii no console will be released being just a XB3/PS4 (I did not said that they will follow Nintendo path), but I think that wii (and a few other games, EndWar seems like one off them) and services like made/will make a point strong enought to see a general evolution here raw tech will mean much less.
4- Reduced visual/gameplay beneficts from having so much power (at least used in a traditional way (meybe in some new ways, like new interfaces, AI...).

Meybe one could find more reasons, but I really think that they will first try to find a good way to deal with all those problems, only then they will think in tech. So i personally think that there will be a (big, but not nearly as big) evolution in HW, quite probably resulting in advances in Cell tech and a mid-end modern GPU for the time, but always thinking in the above points.

 

[babcat]

Hello Everyone,

Quite frankly I don't want the PS4 to launch until the PS3 has been around ten full years. I want the PS4 to be a revolutionary jump above the PS3, and I don't think that will be possible if it launches too soon.

Additionally, Sony has lost a ton of money on the PS3. I think it's going to take them time to recoup all the losses and start turning the PS3 into a money maker. In my opinion, from a financial stand point there is NO reason for them to rush the launch of the PS4. Look at how well the PS2 is doing right now AFTER the PS3 has launched! I think PS3 games and systems will be selling well for a long time. Launching the PS4 too early would be a mistake.

Also, there are MANY things Sony can do with the PS3 without launching the PS4. They could produce additional models of the PS3. For example, they could launch a multimedia model with a huge HDD and a faster Blu-ray drive. These upgrades would NOT benefit the performance of PS3 games, but could enhance the ability of the PS3 to be something like a TIVO or multimedia center.

Additionally, at some point they could launch a LINUX enhanced model. For example, LINUX could be completely and totally already installed. Also, the system could allow access to the RSX and also have more SYSTEM RAM for LINUX applications. Again, this RAM could NOT be used for PS3 games. It would only be used for LINUX applications.

I'm sure there are other possibilities as well. Oh, and I almost forgot. We will of course see size reductions in the PS3 as well. I'm sure there will be a slim line model at some point.

Now, what do I want the PS4 to be like?

First, I want to see it using a highly evolved version of the CELL processor. Now, remember I'm talking about something that could exist in ten years. I would like to see at least two CELL processors linked together. Each CELL chip would consist of one or two PPEs and 32 SPUs. I would like for the PPEs and SPUs to be redesigned to be more powerful and better suited to working on graphical tasks.

I would like to see the CELL more closely integrated with an even more custom designed GPU from NVIDIA. I would like to see the CELL being able to "easily" cooperate with the GPU and be able to help with EVERYTHING (depending on of course what the developers want to do with each game). I don't think this is far fetched. SPU's can already help with vertex work, geometry shading, pixel shading, and other tasks.

I would like the GPU to be highly powerful and dwarf the RSX. This could really happen if the PS4 is not launched until the PS3 has been around ten years.

Additionally, I want to see the system having at least 5 to 10 GB of RAM (the fastest available) with tons of bandwidth.

I know this is ambitious, but I think it is possible when the PS3 has been around ten full years.

 

[wco81]

They won't be able to wait 10 years.

You know MS and Nintendo won't.

Frankly, I would have no problem paying another $600 if they at least doubled the RAM, got a faster Blu-Ray drive and a newer GPU in there. That's after owning the machine less than a year.

Now considering that most of the PS2 installed base hasn't moved over to the next gen...

But they should start thinking about a 2010 or 2011 launch.