Predict: The Next Generation Console Tech

[fellix]

Wow -- hold on, now! This is supposed to be a game console, not an enthusiast PC-rip. I don't think Sony would want to sell another over-engineered bulky brick at loss, again.
PS4 should be a refined and streamlined version of the current platform by design, with minimum backward compatibility issues (if any) and polished set assembly of components.

I say step on, of what the semi-conductor manufacturing process can give for the next Cell (clock-rate wise, mostly), with some needed architectural refreshments and invest in a modest but flexible RSX part, like the one in XB360.

 

[jetrii]

Here are my predictions for the 8th generation. Granted, it's more of a wishlist than prediction. I see both companies easing on the CPU and focusing on a more powerful GPU.

Xbox 720:
Power7 based CPU
6 POWER7 cores
4 threads per core(24 total)
4Ghz
ATI R800 based GPU
4GB GDDR4 unified memory
80MB eDRAM

Playstation 4:
Cell 2 CPU
2 PPE cores
12 SPE
4Ghz
GeForce 400 based GPU
2GB GDDR4
1GB XDR2

 

[fehu]

If they follow the same path, i think that both ms and sony will end up with a similar cpu:
- gp ppe derived from power7
- a number of something similar in concept to actual spe

I want to ask a question to developers
What do you think about multithread core?
Are they really useful?
Is in order architecture to keep also in the future, considering the mature tools and the closed/static nature of the console?

 

[Megadrive1988]

Regarding the GPU and API of the next-generation Xbox, should it be DX11 / SM5 or DX12/SM6 or something inbetween?

DirectX11 / Direct3D11 / Shader Model 5 has already been specc'ed out, probably completed, and there should be supporting hardware for it late this year. It'll be part of Windows 7 which releases late 09 or at the latest, very early 2010.

The next-gen Xbox won't arrive until fall 2011 at the soonest, if not later. Given that, there should be either enough time to develop DX12/ D3D12/SM6 or something that's an improvement and well beyond DX11/D3D11/SM5, much like with Xbox 360, it's GPU and API goes well beyond DX9/D3D9/SM3, and even though it's not all the way to DX10/D3D10/SM4, it's got some features that are not even part of the current DX10 PC spec.

It depends on timing. For example, if in the unlikely event Microsoft launches early, in fall 2010, then I don't expect anything beyond DX11/D3D11/SM5. But if Microsoft launches later, say 2012 or 2013, then obviously there would be a chance of the GPU and API getting next-next DX hardware & API spec.

2006/7 ~ Current = DX10/10.1 / D3D10/10.1 / SM4 / 4.1
2009/2010 ~ Upcoming Next = DX11/D3D11/SM5
2012-2014 ? ~ Future = DX12/D3D12/SM6

On the otherhand, the upcoming DX11/D3D11/SM5 might be more than good enough, feature-wise, and it might be more desirable to focus on rendering those features fast, the speed/performance of the GPU, rather than anything beyond the upcoming standard.

Again, it comes down to timing. I don't expect Microsoft to leave the next Xbox in a trailing position regarding rendering features / shader model generation like the PS3 was. The PS3 launched just at the time DX10 hardware was coming out (G80 / 8800) and just before Vista was released with the DX10 API.

 

[kyetech]

In all fairness even DX10 is damn flexible and has not really been exploited. But from an animation and rendering POV subD surface handling/tesselation on the GPU will be an absolute god send for DX11.

Something that article seemed to miss is that if you are handling a raw poly model of say a 20,000 poly character and then trying to do skinning / deformation, that is going to kill the CPU/GPU processing and also means the chance of having good soft body dynamics is almost nill.

Now being able to deform / skin / bone the basic subD surface before tesselation means that animation effects on things like a characters face can now approach movie realsim becuase animation effects are applied to a 2000 vertex model and not the 20,000 vert output model.

The more flexible programming model will help give rise to some of the more illusive rendering techniques like global illumination / radiosity. As well with RnD mainly tied to the console cycle you can expect DX11 to proliferate like DX9 did as a development platform.

I think everything has come together at the right time for this to be an exciting next chapter in graphics technology development. now we just need to see what Unreal Engine 4 will be capable of, which I think may be previewed in 2010?

 

[babcat]

I want unreal engine 4 to pull off what unreal engine 3 was supposed to pull off.

 

[Mobius1aic]

I think the quad cell would make a better processor choice, considering it's much more than double the raw capability per clock cycle. As for the next GPU to be in the next PS product, it'll most certainly be a unified shader system, problably the equivalent of the GTX400 series, lets say 1024 Unified Shaders, 80+ TMUs, and hopefully more than 8 ROPs this time around Ok, ok, I'd say 40+ or so ROPs at least. Though I do wonder by when we will see GPUs with only one type of "pipe" that can do all functions including texture mapping and rasterization, however thinking about it carefully it probably would be best to leave such actions to a separate part of the GPU considering they are vastly different parts of the rendering pipeline.

 

[grandmaster]

We seem to have gone from accepting that Microsoft and Sony will not leap forward in tech terms that much (due to pricepoint) and have gone back to insane tech specs.

GTX295 is the first card to reach 1080p60 on Crysis (on DX9 any way) at maximum settings and surely that is hugely over-engineered compared to what we can expect from the GPUs in the next gen?

 

[paawl]

We seem to have gone from accepting that Microsoft and Sony will not leap forward in tech terms that much (due to pricepoint) and have gone back to insane tech specs.

GTX295 is the first card to reach 1080p60 on Crysis (on DX9 any way) at maximum settings and surely that is hugely over-engineered compared to what we can expect from the GPUs in the next gen?

I was thinking the exact same thing.

If, for no other reason than to avoid the wrenching software transformation that accompanied the transition from PS2 to PS3, I would expect Sony to learn from Microsoft's example and attempt to maintain software tool continuity from this generation to the next. I expect Microsoft will do the same, since they so clearly benefited from a smooth transition from Xbox to Xbox 360.

This, plus the aforementioned cost benefits, I think, clearly argue for extensions of current architectures rather than radical new ones. The recognition that there are diminishing returns from graphical improvements alone, and the importance of rapidly reaching a mass-market price point, argue against over engineering the next gen consoles.

So here are my two-cent predictions:

Both consoles will feature blue-ray drives as the primary distribution mechanism.

Both consoles will feature optional hard drives, both because the hard drive is a significant impediment to reaching a low-cost (the cost of drives essentially never drop below $50), and because the selling of add-on drives to consumers who bought diskless consoles, or who want a mid-life storage upgrade, can be a significant profit stream.

As for the details, I think that Sony has the most straight-forward upgrade path for their CPU. Developers have already been forced to come to grips with the Cell architecture and writing code for the SPEs. The Cell roadmap already includes a 2 PPE + 32 SPE variant, and this seems a logical choice for PS4.

Microsoft did considerable, interesting research during the design of the 360 and concluded that six to eight CPU threads represent the maximum that can be productively used in games, with more-parallel tasks being better suited to running on the GPU. Assuming that this generality holds throughout this generation (and it has so far), I expect MS to employ, at most, a four-core variant of Xenon, possibly with more L2 cache, for their next CPU.

Microsoft has the most straight-forward upgrade path for their GPU, as I think that something like RV770 with eDRAM would slot nicely into a new console.

Sony's GPU poses the largest problem; I really haven't a clue what they might do here. RSX is already looking a bit long-in-the-tooth, feature-wise, and without eDRAM, framebuffer bandwidth will always be a problem. OTOH, switching to a new GPU, whether G200-like or anything else, will break backward compatibility with PS3 games and game engines, and may not be necessary if more and more rendering tasks are to be performed on PS4's 32 SPEs.

So, my predictions:

Xbox 360 CPU: 3 or 4 cores (with SMT), 1 or 2 MiB L2 cache, 4+ GHz.
Xbox 360 GPU: RV770-ish with 32 MiB eDRAM, 700+ MHz, 2+ GiB GDDR5 @ 5+ GHz on a 128-bit bus.

PS4 CPU: 2 PPEs + 32 SPEs, 4+ GHz.
PS4 GPU: ??

 

[Sound_Card]

I doubt the GPU will be RV770 based with EDram. Heck, I don't even expect a EDram chip on the next xbox. The GPU will be more RV9ish or not an ati chip at all.

 

[3dcgi]

Both consoles will feature optional hard drives, both because the hard drive is a significant impediment to reaching a low-cost (the cost of drives essentially never drop below $50), and because the selling of add-on drives to consumers who bought diskless consoles, or who want a mid-life storage upgrade, can be a significant profit stream.

With the arrival of solid state drives I wonder if the $50 theory changes. Seems like it probably does. You could include a small SSD and still sell upgrades for users that want larger drives.

 

[fehu]

I don't even expect a EDram chip on the next xbox.

why?
i've read somewhere that it's underutilized, or at least utilized for something else, but i've read too that nextbox will use it and maybe as big as 80MB for a full 1080p framebuffer

if they remove it how can emulate the 360?

 

[novcze]

Both consoles will feature optional hard drives, both because the hard drive is a significant impediment to reaching a low-cost (the cost of drives essentially never drop below $50), and because the selling of add-on drives to consumers who bought diskless consoles, or who want a mid-life storage upgrade, can be a significant profit stream.

I can`t imagine next-gen without mendatory hard drive, not after Sony`s approach this gen. But I can imagine diskless consoles sold to those who want buy their own hard drive in desired capacity (on free market, not overpriced accessory from console manufacturer).

 

[sebbbi]

i've read somewhere that it's underutilized, or at least utilized for something else, but i've read too that nextbox will use it and maybe as big as 80MB for a full 1080p framebuffer

if they remove it how can emulate the 360?

1080p framebuffer + zbuffer (at 32 bit each) is 15.82 megabytes. 1080p with 2xmsaa is 31.65 megabytes. I doubt they would make the EDRAM larger than 32 megabytes on the next gen box, as the current 10 megabytes is not even enough for 720p 2xmsaa (you need to split the scene to two tiles for it). With 32 megabytes the developers could choose between 1080p with 2xmsaa (31.65 megabytes) or 720p with 4xmsaa (28.125 megabytes). Or optionally they could split the scene to half and render 1080p with 4xmsaa if they wanted.

I doubt the EDRAM itself is a requirement for the next gen console backwards compatiblity, as it is mainly there for performance reasons. You could easily emulate the EDRAM by allocating a constant 10MB block of graphics memory for Xbox 360 titles backwards compatiblity.

 

[RudeCurve]

1080p framebuffer + zbuffer (at 32 bit each) is 15.82 megabytes. 1080p with 2xmsaa is 31.65 megabytes. I doubt they would make the EDRAM larger than 32 megabytes on the next gen box, as the current 10 megabytes is not even enough for 720p 2xmsaa (you need to split the scene to two tiles for it). With 32 megabytes the developers could choose between 1080p with 2xmsaa (31.65 megabytes) or 720p with 4xmsaa (28.125 megabytes). Or optionally they could split the scene to half and render 1080p with 4xmsaa if they wanted..

Thanks for doing the math...it makes 32MB eDRAM much more likely to be the target memory footprint for next Xbox. At 1080p you don't really need much AA so it's a win-win IMO.

 

[paawl]

For the Xbox 720, a four core CPU (8 hardware threads) would likely be the minimum considered. But I expect to see more cores, or either a much improved clock frequency to differentiate the performance enough from the current gen. If they want to keep the backwards compatiblity it has to be a PowerPC based core (big endian). x86 CPUs are not instruction compatible and store values to memory in little endian format.

I was actually thinking of the current Xbox 360 CPU, with the same pipelines, ISA, etc., possibly bumped to four cores, likely with a larger cache.

I have a gut feeling that MS won't increase the number of cores beyond four, since their own research concluded that, "for the types of workloads present in a game engine, we could justify at most six to eight threads in the system (Jeff Andrews and Nick Baker, "Xbox 360 System Architecture", IEEE Micro, March-April 2006, p. 35)."

Unless a real breakthrough in parallel programming is made in the next couple of years, I think it's more likely that, should MS feel the need for more CPU power, they would choose to enhance three or four small, in-order cores in other ways. Possibilities include a larger shared L2 cache, 16-way vector units, and/or new instructions to support common tasks, like the dot-product and Direct3D (de)compression instructions in the current 360 CPU.

I doubt the EDRAM itself is a requirement for the next gen console backwards compatiblity, as it is mainly there for performance reasons. You could easily emulate the EDRAM by allocating a constant 10MB block of graphics memory for Xbox 360 titles backwards compatiblity. All the copy operations (resolves) from EDRAM could be emulated by shaders. This of course assumes that the next gen GPU is considerably faster than the current GPU (and has a huge bandwidth and shader processing power). Having a ultra fast 32 MB EDRAM chip (for render target) would of course benefit the future titles a lot, especially if they still consider using unified memory architecture in the Xbox 720.

FWIW, every PS3 with PS2 backwards compatibility had to have the PS2's eDRAM. Unless I'm mistaken, no eDRAM, no BC.

 

[kyetech]

Thanks for doing the math...it makes 32MB eDRAM much more likely to be the target memory footprint for next Xbox. At 1080p you don't really need much AA so it's a win-win IMO.

If they had something like 8x aa for edges, and 2x aa for the rest of the scene (or even better a fully adaptive system based on pixel contrast) + a good level of anistropic filtering, then rendering would be very clean and pretty cheap on the hardware.

Next gen consoles will really nail clean rendering. imo.

 

[Sound_Card]

With the arrival of solid state drives I wonder if the $50 theory changes. Seems like it probably does. You could include a small SSD and still sell upgrades for users that want larger drives.

I'm willing to bet, that the next gen consoles (well, PS4 and nextbox) to have built in internal solid state drives on all models with the capability to add an external drive... proprietary or not.

....

1) I don't see life for power PC beyond the 360. I expect a radical change on the CPU.
2) More willing to bet knowing MS, that BC will not be a key feature and emphasis will be placed on "360 originals" via download on xbox live. Besides, I can't remember where I found it, but the amount of people playing last gen games on next gen systems was very small. To me, not worth building next gen systems around old hardware. Time to progress.

 

[pc999]

I was actually thinking of the current Xbox 360 CPU, with the same pipelines, ISA, etc., possibly bumped to four cores, likely with a larger cache.

That would be a real minor update to the CPU, no? specially given how weak each core is, probably many todays dual cores (at least if costum) would beat it...
and would keep all of the problems.

 

[MarkoIt]

I think the quad cell would make a better processor choice, considering it's much more than double the raw capability per clock cycle. As for the next GPU to be in the next PS product, it'll most certainly be a unified shader system, problably the equivalent of the GTX400 series, lets say 1024 Unified Shaders, 80+ TMUs, and hopefully more than 8 ROPs this time around Ok, ok, I'd say 40+ or so ROPs at least. Though I do wonder by when we will see GPUs with only one type of "pipe" that can do all functions including texture mapping and rasterization, however thinking about it carefully it probably would be best to leave such actions to a separate part of the GPU considering they are vastly different parts of the rendering pipeline.

1024 Stream processor?! Even in the "worst" case, with the next generation consoles delayed to the end of 2012 due to the economic recession, there will be no way to have 1024 superscalar ALU on a 200mm^2 die.. it would require a 16nm pp, expected in 2014 or later, not in 2012.
For the next Xbox, if Microsoft will use again an Ati chip, it would probably be son of the next DirectX11 architecture (R800).
I expect the use of the 28nm/32nm, and a ratio between shader and tmu 6:1.
From the spec point of view:
200 mm^2
256bit/192bit@GDDR5 or XDR2 -> bandwidth 200-300 Gb/s
20ish ROPs
400-500 Vec5 ALUs
60-80 TMU
One mempool of 4 to 8 Gb GDDR5/XDR2

If they stick with IBM, the CPU will be son of Power7 architecture.