Predict: The Next Generation Console Tech

[eastmen]

XP 32bit can use more than that, though. But there is no need, yet. As I said, there are games, that can use more than 1.5GB of RAM, but none see the need. Resolutions won't rise much anymore, as humans simply won't be able to see any difference... textures will reach the nyquist frequency before long... there'll be no need to raise texture resolution anymore, as there's no visual difference in doing so. And several games already have "perfect" textures, too.

I guess, the need rather goes to faster RAM, than bigger. 2 to 4GB will be more than enough, next generation, to server all needs. But beyond that, there'll be a need to address the other things... more and higher quality models on screen without heavy LOD, world physics, nothing baked, realtime GI... all those things won't need much RAM, at least compared to textures, unless you ramp up your GTA4 with several hundreds of car types creating traffic jams and crashing into buildings, destroying them and their car along with it, simultaneously.

The way PS2 and to a degree the 360 handle it, are good. A big pool of cheap general purpose RAM plus expensive additional fast RAM for other stuff. I am still quite in love how PS2 handled framebuffer operations like it did... made for some cool surreal games.

I don't see how windows xp can use more than that , even with the switch your still limited to 3.5gigs of ram.

As for needing more ram , people will allways want more textures and for everything to look better. The more unique textures you can use in a frame the better the game will look.
Then there is the added benfit of dumping the whole game into ram once it becomes cheap enough.

You can fit all of crysis into 16 gigs of ram and have 6 gigs or so left for other tasks. You'd get instant loading of the game which would make many gamers very happy.


As for ram speed and capacity , it seems like every two micron drops new dimm sizes become the norm. Right now 2gig dimms are the sweet spot and have been for at least 2 micron drops. With 42nm it seems like 4gig dimms will be the sweet spot in pricing , 1 gig dimms will most likely go the way of the doodoo . When we hit 22nm we may see 8gig dimms become the norm. As ram becomes cheaper more and more people will buy larger quanitys of ram. As the average skews towards 8 gigs instead of 4 you will see games and other applications grow to fill that extra ram. It wasn't long ago (with the launch of 7) that 8 gigs of ram was expensive. Now you can get it for less than a $100 even without sales

 

[AlNom]

So no one PMed you to look at this link?

Wasn't me. He just "knows".

It also requires SOI rather than bulk processes which therefore would add to the expense. It would also mean there would be no convenient half nodes to shrink to either.

hm... AMD is still designing for SOI for 32nm. I haven't checked the roadmaps recently, but are they planning on switching to bulk

 

[MfA]

It also requires SOI rather than bulk processes which therefore would add to the expense. It would also mean there would be no convenient half nodes to shrink to either.

There's a decent chance bulk might die altogether after 22 nm at the moment.

 

[rpg.314]

There's a decent chance bulk might die altogether after 22 nm at the moment.

I think Intel will push on with bulk after 22 nm as well.

 

[AlNom]

Well, until Intel gets into a home console, it'll be somewhat irrelevant what those magical Intel fabs can do. The dioxide thickness is certainly a concern, but I suppose they have all their fancy tri-gates etc. AMD/Global Foundries/TSMC on the other hand... :s I'd hope for something coming out around 2014 considering all the blundering/delays with processes. It'll at least give a fighting chance for any mid-high end Fusion-type chip.

 

[Squilliam]

hm... AMD is still designing for SOI for 32nm. I haven't checked the roadmaps recently, but are they planning on switching to bulk

Well, their mainstream fusion parts are fabbed on bulk and they did mention how the extra cost of SOI was hurting them in their competition against Intel. I would say that they'd at least be considering it for their entire range of processors.

There's a decent chance bulk might die altogether after 22 nm at the moment.

Apparantly theres also a decent chance the normal shrink pattern might die after 11nm. But with the semi-industry I think I'll just wait to see before I believe.

 

[bomlat]

The console industry hit the wall five years ago.
If they want to lift by one magnitude the processing power of the consoles then the electrical consumption easly can go to the 300-500 watts range (liquid cooling,or big fans),and the price tag on the 1000$ range,which mean 500 $ loss on each console,if they want to make a market.(5 billion loss in the first year.....)

The last version of the PS/XB break the back of the console division of the biggest consumer electronicy/software company.The next iteration will break the back of the complete company.

So,there is no chance for a new console in the next 3-5 years. (probably a WII HD,but that will have probability if the sales will drop dramaticaly)

 

[MfA]

I think Intel will push on with bulk after 22 nm as well.

Maybe, but Intel doesn't care about lack of tools or developers who want to stick to what they know ... their margins and volume allow them to recoup a lot more development costs. Non-planar logic with bulk is a bigger jump than simply going SOI ...

 

[Squilliam]

The console industry hit the wall five years ago.
If they want to lift by one magnitude the processing power of the consoles then the electrical consumption easly can go to the 300-500 watts range (liquid cooling,or big fans),and the price tag on the 1000$ range,which mean 500 $ loss on each console,if they want to make a market.(5 billion loss in the first year.....)

The last version of the PS/XB break the back of the console division of the biggest consumer electronicy/software company.The next iteration will break the back of the complete company.

So,there is no chance for a new console in the next 3-5 years. (probably a WII HD,but that will have probability if the sales will drop dramaticaly)

It's like you're working off the assumption that all consoles will be released using the 90nm process node again

 

[AlNom]

It's like you're working off the assumption that all consoles will be released using the 90nm process node again

It'll be interesting to see 28nm Cayman's die size with similar specs and clocks to Cypress (to account for design changes like VLIW4 etc), which would put it at more than 10x Xenos' theoretical max (in terms of flops if that's the metric we want to really use). That's a full node from 40nm as well, so at best we'd be looking at ~120mm^2 die size (vs. 334mm^2 Cypress) assuming everything scaled perfectly and without considering any changes to ROP/texture unit counts to take into account their usefulness when paired with a 128-bit bus to GDDR5. Of course, there's the notion that they'd be designing specifically for 28nm so the scaling/die size comparison will be off as well.

 

[bomlat]

It's like you're working off the assumption that all consoles will be released using the 90nm process node again

No,I just considering the fact:the PS3 using 60watts of power currently (latest),and if you increase the die size tenfold then you will have min. 300-400 watts.
If you use 90nm then the power consumption could be in the 1300 watts,so that is the PS5 specification if you improve the node size

 

[AlphaWolf]

No,I just considering the fact:the PS3 using 60watts of power currently (latest),and if you increase the die size tenfold then you will have min. 300-400 watts.
If you use 90nm then the power consumption could be in the 1300 watts,so that is the PS5 specification if you improve the node size

Wasn't RSX over 200mm at 90nm? Who is expecting anything 10x that size?

And I don't think a 300watt console is out of the question, the ps3 and xbox360 were both around 200 at release.

 

[bomlat]

Wasn't RSX over 200mm at 90nm? Who is expecting anything 10x that size?

And I don't think a 300watt console is out of the question, the ps3 and xbox360 were both around 200 at release.

They are out of question.
No one can put onto the table the enormous upfront money for a ps3/xb3.


And the 300 watts is more likely an optimistic calculation-the realistic is at least 400 watts.

 

[AlphaWolf]

They are out of question.
No one can put onto the table the enormous upfront money for a ps3/xb3.

Microsoft certainly can afford to, the only question is if they want to do it.

And the 300 watts is more likely an optimistic calculation-the realistic is at least 400 watts.

Why? It seems like you're just pulling numbers out of your nether regions.

 

[DuckThor Evil]

No,I just considering the fact:the PS3 using 60watts of power currently (latest),and if you increase the die size tenfold then you will have min. 300-400 watts.
If you use 90nm then the power consumption could be in the 1300 watts,so that is the PS5 specification if you improve the node size

PC GPUs already have about 10x the transistors of RSX at 40nm and power consumption in normal operation is around 200w or even less. They should also have room to tweak for consoles, not to mention that 28nm is coming soon.

GTX 580 uses a bit more power, but it still has very elegant and relatively small cooling system. IMO you're way too pessimistic on just about every point you've made so far.

 

[bomlat]

There is three independent factor:
1.,The power consumption ratio between the console generations,and the power consumption of the ICs inside the consoles.
2.,The ps3 25xx have 60 watts power requirement-if you expect 0.5 linearity,then roughly you have 300 watts.
3., the power consumption of the PC increased by the same amount during this time period.

 

[Squilliam]

It'll be interesting to see 28nm Cayman's die size with similar specs and clocks to Cypress (to account for design changes like VLIW4 etc), which would put it at more than 10x Xenos' theoretical max (in terms of flops if that's the metric we want to really use). That's a full node from 40nm as well, so at best we'd be looking at ~120mm^2 die size (vs. 334mm^2 Cypress) assuming everything scaled perfectly and without considering any changes to ROP/texture unit counts to take into account their usefulness when paired with a 128-bit bus to GDDR5. Of course, there's the notion that they'd be designing specifically for 28nm so the scaling/die size comparison will be off as well.

The comparison with Cayman/Cypress is complicated further by the fact that they both sport a 256 bit and in the former case, a high speed GDDR5 bus. That is completely out of the question with regards to a console design. IMO it is better to start with a design which is closer to the console specifications as they stand currently and will likely stand in the next generation consoles. I would say that Juniper is the best candidate especially with much of the non GPU performance related improvements removed, it ought to come close to the Xenos die size on 40nm whilst yielding 3* the memory bandwidth and 6* the GFlops of compute performance. From there it is relatively easy to extrapolate that with similar area devoted to the GPU on the 28nm process it ought to yield about 4* the memory bandwidth and 10* the GFlops when you consider the overall die size without extranious features without going above 60W for the GPU on its own.

 

[bomlat]

PC GPUs already have about 10x the transistors of RSX at 40nm and power consumption in normal operation is around 200w or even less. They should also have room to tweak for consoles, not to mention that 28nm is coming soon.

GTX 580 uses a bit more power, but it still has very elegant and relatively small cooling system. IMO you're way too pessimistic on just about every point you've made so far.

200W GPU + 100 W CPU + 100W everything else- and we are in the water cooling/big box air cooling range.

 

[bomlat]

Microsoft certainly can afford to, the only question is if they want to do it.
.

They can't afford it.
In that case they supposed to use all free cash /OI from all business unit just to support the XB3 project.Suicide tactic,and there is no board that will approve it.

 

[Squilliam]

200W GPU + 100 W CPU + 100W everything else- and we are in the water cooling/big box air cooling range.

You seriously believe consoles will consume over 300W of power? Thats ridiculous.