Predict: The Next Generation Console Tech

[Silenti]

Couple of questions.

Has MS ever continued its talk of "forward compatibility" or whatever term they used. Where older games would look better on the newer hardware? How would that concept effect what your choices are next gen? Pretty much mean there would have to be EDRAM no?

With the, seeming, dominance of the importance of 3rd party developers, isn't anything you include which is exotic just going to mean its something that is not really properly used? From my understanding, a bunch of the development done to make use of Cell helped on the 360 as well, but the EDRAM and some of what the GPU in the 360 can do wasn't really pushed. (I'm tired and sketchy on details here. Something about Rare and Viva Pinata being one of the few games that used even some of these features.) Maybe 1st party comes back big time, but MS seems to have gone the 2nd and 3rd party route heavily. Maybe they won't benefit from something that is too much of a step away from what their competitors are doing. Why pay for it and include it if it doesn't get used?

 

[hoho]

Pretty much mean there would have to be EDRAM no?

I'm fairly certain that with high enough bandwidth it's relatively trivial to emulate eDRAM. I mean 128bit memory interface GPUs have around 64GB/s bandwidth vs 32GB/s to eDRAM and 22 to VRAM for XB360. I would think going with 192bit interface VRAM would be cheaper than 128bit VRAM+ whatever is needed for eDRAM and would provide more than enough raw throughput to emulate eDRAM

 

[AlNom]

Has MS ever continued its talk of "forward compatibility" or whatever term they used. Where older games would look better on the newer hardware?

Well, super-sampling ought to help a lot of shader-heavy games significantly.

From my understanding, a bunch of the development done to make use of Cell helped on the 360 as well,

That probably had a lot more to do with keeping code tight within limited cache space. Xenon does only have 1MB for 6 threads.

Why pay for it and include it if it doesn't get used?

Well, aside from the edram and tiling, the 360 wasn't really exotic.

32GB/s to eDRAM

This is the wrong figure to look at. The heavy bandwidth ops are writing pixels within the eDRAM/ROP chip (256GB/s).

 

[hoho]

This is the wrong figure to look at. The heavy bandwidth ops are writing pixels within the eDRAM/ROP chip (256GB/s).

That is assuming games would actually use 4xMSAA when emulated instead of using some smarter and far more efficient AA algorithm

 

[AlNom]

Yeah, but that's still the wrong comparison (the interlink between the eDRAM and the shader dice).

instead of using some smarter and far more efficient AA algorithm

With respect to emulation, some edge blur isn't going to do crap for shader aliasing or any thin edges. Emulation is bound to have wonky performance anyway.
What they will need is considerable bandwidth for say z-only sampling, which can be done at much higher rates in newer hardware than just the pixels (sans MSAA).

 

[Ruskie]

"For the home equipment the PS3 still has a product life," he said, "but this is a platform business, so for the future platform - when we'll be introducing what product I cannot discuss that - but our development work is already under way, so the costs are incurred there."

http://www.eurogamer.net/articles/2011-05-26-sony-is-working-on-playstation-4

More speculation!:smile:

 

[hoho]

Yeah, but that's still the wrong comparison (the interlink between the eDRAM and the shader dice).

Well, they might need some rather tricky emulation (on-the-fly recompiling some stuff?) but considering most games were able to squeeze roughly similar image quality on PS3 that has far lower GPU computing power and excluding XDR just around 23GB vram bandwidth vs that 22+32/256GB/s I don't think it would be that much of a problem.

So in short, it should be doable even if it's tricky

 

[RudeCurve]

I could see PS4 using a CELL processor with more cores maybe 12 SPEs and clocked at 4GHz? The GPU would be an upgraded version of whatever is in the PS3 + more RAM. That should cut down R&D expenses significantly and manufacturing would be fairly straight forward. This means 100% backwards compatibility and developers could use existing software base without having to start from scratch.

 

[DuckThor Evil]

PS3 had one SPU disable in its Cell processor. If they did it right, they would have save some money just making a 7 SPU Cell.

1 SPU only increased the size of the chip about 6-7% and the added redundancy probably helped quite a bit with the yields. I'm thinking that was probably a good decision.

edit:
The yields would have to have been really good for a 7/7 spu chip to have made sense over 7/8 chip. 6/7 chip is another thing.

 

[BoardBonobo]

I guess sticking with the CELL is the smart thing to do. Just add dual precision to the SPU's and OOE to the PPU. Up the clock speed slightly. After the current CELL has proved itself to be quite a beast who's true potential is held back by the supporting hardware. In the case the RSX.

Unify the memory and up it to at 1Gb and tack on a 90Mb eDram pool to the RSX2 (even a broader faster G80 would be very nice in there!)

 

[Lucid_Dreamer]

I guess sticking with the CELL is the smart thing to do. Just add dual precision to the SPU's and OOE to the PPU. Up the clock speed slightly. After the current CELL has proved itself to be quite a beast who's true potential is held back by the supporting hardware. In the case the RSX.

Unify the memory and up it to at 1Gb and tack on a 90Mb eDram pool to the RSX2 (even a broader faster G80 would be very nice in there!)

I thought we've learned that eDRAM is not worth the trouble. Sony dropped it from the PS3. It's very likely that MS will drop it for their next console as well (never to return). The PS4 will most likely fore go it, too.

Well, it's more like the amount of RAM held back the true potential of the Cell. Because there was no GPU out at the time that could allow Cell to only focus on non-graphics tasks, that looks more like a moot point. Of course, there are exceptions like Motorstorm: Apocalypse, etc (Cell not used for big graphical tasks in those instances). If anything, that has allowed the world to see all the versatility the Cell has. Otherwise, how would you know that the Cell was good at certain graphical tasks? We would only know it was good at A.I., physics, etc. I'm just saying...

 

[Brad Grenz]

I think the plan was for the PS4 to have reduced R&D expenditure from the start of the PS3 planning. I've always thought of the Cell as a multi-generational processor. I've thought of Blu-ray being in PS3 and PS4. Plus, the mention of a GPU roadmap starting with the RSX points to the same. Of course, I always thought the PS4 would include the evolution of those original investments.

The Cell was a processor design based on the best hardware performance one could get when unencumbered with legacy support (for die space and money). One of the goals was to be ahead of Intel's processor performance for, I believe, four or five years from launch. They have accomplished that. STI know they can, easily, push the clock speed up to 5 GHz. They, also, know they can drop up to a 36 core Cell on a 45nm process. That research is done. That screams thinking beyond the PS3 to me.

They made a lot of money from the PS2. I believe the thinking was, at some levels, about an investment into a long-term future. It seems to be an investment that spans, at least, two console generations. The software development of libraries for the Cell screams that it would not be complete in only one console generation.

This frees up options for the future. Either you can scrap everything and start over next generation, including very high R&D costs. Or, you can have much lower R&D costs next generation with high performance and ease of development (all while your competition, typically, has to invest heavily and start over).

Think of it as the filming of Matrix 2 & 3 or Lord of the Rings. It's cheaper to film the movie as one, instead of separately. The difference being that it looks worse on paper ($225 million instead of $75 million per film). In the case of the PS3 & PS4, the development appears to go beyond the short-sightedness of a single console generation.

Yeah, my hope is that they'll carry forward the Cell to the next system. Improve the PPU, maybe add a second one, and bump up to like 48 SPUs and they can leverage all the research Sony developers have done into new programming models, and they can couple it with whatever unified shader nVidia GPU fits their price/performance needs. Whether or not they planned it, it should allow Sony to keep costs down while still delivering a lot of performance in the next generation.

 

[Shifty Geezer]

I think the plan was for the PS4 to have reduced R&D expenditure from the start of the PS3 planning. I've always thought of the Cell as a multi-generational processor...That screams thinking beyond the PS3 to me.

Though I agree Cell was developed to provide a long-term hardware platform and its uniform scalability provides a natural progression for future hardware, its lack of penetration means Sony may be looking at other routes. The notion of Cell being ubiquitous in different scales in different devices would have made Cell in PS4 a given, and meant Sony's investment was extremely shrewd. But investments don't always work out. It looks like Cell is something of a dead-end, sadly. And rolling Cell components (SPUs) into another processor is going to add RnD cost that PS4 shouldn't have to burden.

 

[function]

I thought we've learned that eDRAM is not worth the trouble. Sony dropped it from the PS3. It's very likely that MS will drop it for their next console as well (never to return). The PS4 will most likely fore go it, too.

In the last two generations only the least successful console has not used embedded memory. Sony didn't have the option of using it this time after scrambling to use a version of the GPU Microsoft rejected.

Embedded memory needs to be more flexible (in the case of the next Xbox) to match the increasing flexibility of GPUs, but it doesn't need to go away.

Well, it's more like the amount of RAM held back the true potential of the Cell. Because there was no GPU out at the time that could allow Cell to only focus on non-graphics tasks, that looks more like a moot point.

RSX allows Cell to focus on none graphics tasks. The thing is that even with all that Edge type stuff running on Cell, even in first party games that don't have to run on the 360, the CPU is still sitting so idle that games can implement CPU AA (and sometimes "in a day").

Cell has had plenty of opportunity to focus on "none graphics tasks" but always ends up twiddling its thumbs, waiting to do graphics. Might want to think about using a better graphics chip next time! Having a good graphics chip always trumps having a faster CPU.

Next gen Sony need to decide what software they want the PS3 to run and spec the cheapest CPU that can do it. In other words, they need to handle multiplatform games as well as the next Xbox and do it as cheaply as possible. That's it. If Cell can do that then they'll use Cell, if it can't then its dust.

 

[Rangers]

I guess sticking with the CELL is the smart thing to do. Just add dual precision to the SPU's and OOE to the PPU. Up the clock speed slightly. After the current CELL has proved itself to be quite a beast who's true potential is held back by the supporting hardware. In the case the RSX.

Unify the memory and up it to at 1Gb and tack on a 90Mb eDram pool to the RSX2 (even a broader faster G80 would be very nice in there!)

...wow, that's just terrible.

EDRAM is awful, guys. 90MB? Probably cant even be manufactured, let alone the massive black hole of cost sucking that would be.

It would/should/better have at LEAST 4GB of RAM.

And a GPU that eats G80 for lunch...

Cell should be ditched...too difficult to program by far. Maybe it's got some corner case advantages, that's all. And I've always though, there's no point spending CPU $ to help the GPU, which is mostly what Cell does. That's a big misappropriation of $.

 

[Laa-Yosh]

I still have doubts about using 4GB of RAM...

The most memory intensive kinds of data are:
- textures: should be handled with virtual texturing
- sound/voice: should be streamed
- various graphics buffers: don't expect anyone to go above 1080p, memory bandwidth is also going to limit how much more these can require
- animation: that's the only one that can't be helped
In short, I don't really see what content they'd need 4GB for. That might change, of course, as new graphics and game technology is invented, maybe they'll find a way to do hair that requires lots of memory, or start using blendshapes for animation, and so on. Anyway, a 4x increase and an 8x increase aren't that much different, and scaling up content is happening slower every day because of asset production costs.

On the other hand, adding that much memory will cause problems elsewhere:
- cost
- component complexity (bus width on the system mainboard)
- heat and power consumption
- lack of a fast enough background storage to keep loading times manageable (this generation is already pushing at the limits of my patience)

 

[hoho]

In the last two generations only the least successful console has not used embedded memory

Gamecube had some eDram as well you know. Or in other words, having an edram had pretty much nothing to do with the success of the machine.

One pretty bad thing about edram is if you can't stack it then you'll be needing a whole lot of extra pins on your chip.

 

[Lucid_Dreamer]

That's an incredible amount of effort you're putting in to spin this.

Sony never intended for this generation to be something of a disaster, and I doubt that RSX was actually an expensive, ahead-of-its-time investment in PS4.

MS will have "ease of development" next generation without without sacrificing this generation, and I expect the same thing from Nintendo.

There is no spin. It just is what it is. Why are you talking about disaster and I'm talking about forward thinking R&D concepts? BTW, how do you know MS will have "ease of development" next generation? That sounds a bit like wishful thinking to me. It may end up being the truth, but you have nothing to base that on, now. Multi-threading will be the order of the day, next-gen. Cell's practices completely carries over to the future.

What?

You read that just fine. Take a look at one of DICE's recent slides with one stat on culling, I believe. There are others. That wasn't an "out of the blue" statement. There are tasks that Cell is about twice as fast at than modern 4-core Intel offerings.

Though I agree Cell was developed to provide a long-term hardware platform and its uniform scalability provides a natural progression for future hardware, its lack of penetration means Sony may be looking at other routes. The notion of Cell being ubiquitous in different scales in different devices would have made Cell in PS4 a given, and meant Sony's investment was extremely shrewd. But investments don't always work out. It looks like Cell is something of a dead-end, sadly. And rolling Cell components (SPUs) into another processor is going to add RnD cost that PS4 shouldn't have to burden.

If Cell is something of a dead end, what would make it a success (or just not a dead-end), in your eyes? It's it TVs, laptops, servers, and home consoles. Are they in every electronic device? No, but does it need to be? How many millions need to be made to not be a dead-end? I'm just curious.

What R&D costs does it add to the PS4? I thought a new product always incurs additional R&D costs. I, also, thought a future Cell was mostly researched already. Additional R&D costs, on the matter, should be held to a minimum; in my eyes.

 

[Lucid_Dreamer]

In the last two generations only the least successful console has not used embedded memory. Sony didn't have the option of using it this time after scrambling to use a version of the GPU Microsoft rejected.

Embedded memory needs to be more flexible (in the case of the next Xbox) to match the increasing flexibility of GPUs, but it doesn't need to go away.

Now, THAT'S "spin".

RSX allows Cell to focus on none graphics tasks. The thing is that even with all that Edge type stuff running on Cell, even in first party games that don't have to run on the 360, the CPU is still sitting so idle that games can implement CPU AA (and sometimes "in a day").

Cell has had plenty of opportunity to focus on "none graphics tasks" but always ends up twiddling its thumbs, waiting to do graphics. Might want to think about using a better graphics chip next time! Having a good graphics chip always trumps having a faster CPU.

Next gen Sony need to decide what software they want the PS3 to run and spec the cheapest CPU that can do it. In other words, they need to handle multiplatform games as well as the next Xbox and do it as cheaply as possible. That's it. If Cell can do that then they'll use Cell, if it can't then its dust.

New first party games (big titles) don't, usually, have idle CPU time. They keep finding different ways to use Cell to make the GPU more efficient. That concept works in ANY home console gaming setup. Bigger GPU's won't change the fact that some burden can be taken off of the GPU to make games look better! That within itself defeats the argument of having a better graphics chip. The fact is that, at any given time, there is only one top GPU you can have from a given manufacturer. In order to push graphics further than what's available at that time, you need to have a stronger CPU to ease that GPU's burden with would be idle time.

 

[Shifty Geezer]

If Cell is something of a dead end, what would make it a success (or just not a dead-end), in your eyes? It's it TVs, laptops, servers, and home consoles. Are they in every electronic device? No, but does it need to be? How many millions need to be made to not be a dead-end? I'm just curious.

To not be a dead end, it needs an architecture roadmap that's being actively developed. Cell would need IBM/STI to be developing new Cells, and if there's no market for these processors, there's no point developing the chips. Cell was developed for a larger market than just PS3. Now that's apparent that it was only PS3 that provided a market, retrospectively we can say the cost wasn't worth it. If Cell2 is only going to feature in PS4, the cost of developing it likely won't be worth it. Now it may be Sony can just double up a current 1:8 Cell and that'd suffice, without needing much development, but it wouldn't provide the ease of development of a more rounded processor (think Cell with better OOO Power core or whatever).

What R&D costs does it add to the PS4? I thought a new product always incurs additional R&D costs. I, also, thought a future Cell was mostly researched already. Additional R&D costs, on the matter, should be held to a minimum; in my eyes.

It depends on what format the new CPU takes, but if it's a full fledged new processor, it'll cost as much to produce as any advancement of a CPU. Think 286 to 386, or SGX 4 series to SGX 5 series. The cost is basically however much money you want to through at it! The more you spend, the more you can do. What Cell for PS4 is lacking is a wider market to ameliorate the RnD. A billion dollars spent on developing Cell 2 would have to be paid back by PS4, earing the higher price tag, whereas if there was a market in CE devices, the RnD cost could be spread around more and have less cost on PS4, being less of a ball-and-chain around its price/profitability.

The fact that Toshiba tried Cell in high-end TVs but has dropped it in favour of custom ASICs pretty much seals Cell's demise IMO.