Predict: The Next Generation Console Tech

[pc999]

One way to extend their current model would be to use a variation of the PPC970FX or MP. The PPC970FX took 62mm2 at 90nm lithography, so at 45nm it would probably be some 20mm2 or very roughly the size and price of the current Wii CPU at 90nm. The PPC970MP was a dual core implementation, with not quite twice the die size since the cores shared some processor elements - this could also be an option, particularly at 32nm. The performance of the PPC970 is, clock for clock, similar to Intels C2D, with a collection of perks. If I were Nintendo and wanted to apply this core, I would target some 2GHz for very low power consumption. That number is grabbed out of thin air, the goal being sub 5W CPU power draw at full blast, the 90nm PPC970FX was shipped at up to 2.7GHz. There would likely be some additional work required on the memory bus/controller side to create a suitable console environment.

Going beyond dual core probably wouldn't make any sense, and I'm not certain that two cores would be sufficiently utilized to justify the additional cost and heat. It may well be a better idea to simply increase the clock of the single core as long as we're in a fairly sympathetic part of the clock vs. power function as we should be at these low power levels.

Another option could be to use an adapted PPE, but I'm not sure that this would necessarily be a better move.

Of course, this assumes that they want to extend their current CPU architecture upwards. They may choose to go ARM Cortex9 or some such, break backwards binary compatibility, but gain in other areas.

Probably for a dual core they would be better served if they just used some acelerator, would keep dev easier, so unless it is very revamped (not in order, for multithreading) a PPE would be out of question.

Anyway what about modern cores like a single core phenom/i7/power7, or even other future cores like buldozer and such?

 

[Quaz51]

No, 12x is the current maximum, but Pioneer have a lab version of a 16x drive.

BD 16x speed rotation is equivalent to DVD SL 22.54x. it's over physical limit?? (and read 150 nanometer pit at 283km/h - 456 mph is amazing )

 

[NathansFortune]

BD 16x speed rotation is equivalent to DVD SL 22.54x. it's over physical limit?? (and read 150 nanometer pit at 283km/h - 456 mph is amazing )

http://www.pocket-lint.com/news/news.phtml/22578/23602/sony-optiarc-AD-7240S-dvd-burner.phtml

It is possible. I don't think it is probable though...

 

[Quaz51]

http://www.pocket-lint.com/news/news.phtml/22578/23602/sony-optiarc-AD-7240S-dvd-burner.phtml

It is possible. I don't think it is probable though...

thanks

 

[pjbliverpool]

The performance of the PPC970 is, clock for clock, similar to Intels C2D, with a collection of perks.

I remember it being more in line with the A64/X2 which would put it about 20% slower than Conroe clock for clock.

 

[Jugix]

BD 16x speed rotation is equivalent to DVD SL 22.54x. it's over physical limit?? (and read 150 nanometer pit at 283km/h - 456 mph is amazing )

283 km/h == 175,9 mph
456 mph == 733,9 km/h

 

[Quaz51]

283 km/h == 175,9 mph
456 mph == 733,9 km/h

yes it's 176mph of course

 

[Blazkowicz]

an AMD Regor would make a fine console CPU I think : dual core 45nm K10.5 "Hydra" (newer revision of the process), 2x1MB L2, nice performance and efficiency to expect of it. It also exist.. PPC970 was the last high end general public PowerPC.

 

[grandmaster]

On the topic of noisy DVD drives, my Zephyr debug is astonishingly quiet. The fans are much louder than the drive, and yet it's a basic Hitachi 78 or 79, the same that I have in my Elite which is frightfully loud. Do the debugs run the drive at a slower speed?

 

[Entropy]

Anyway what about modern cores like a single core phenom/i7/power7, or even other future cores like buldozer and such?

x86 - incompatible, too big, too power hungry(/unit performance if you want to quibble) and thus too expensive.
Power7 - too big, too expensive.
Other exotica - incompatible, too big, too power hungry, too expensive, lacking in tools.

I'm boring, I know.

ARM - incompatible, licenseable IP, small, cheap, low power, suitable for SoC building, lends itself to process shrinks/adjustments when it suits the licensee, unlikely to offer quite the same performance level, may offer stationary-mobile convergence.

I remember it being more in line with the A64/X2 which would put it about 20% slower than Conroe clock for clock.

Depends on what you do with it. 20% is way below the effect of compilers, problem domain, specifics of memory organization and access and so on. In my area of interest, the dual 64-bit MAC-capable single cycle FPUs per core, coupled with a remarkably decent memory subsystem capable of supporting quite a bit of physical memory without jumping through hoops made the PPC970 machines pretty spiffy performers. If you compile gcc or some such branchy integer code using Intels reference compiler, the C2D will shine. Horses for courses.

Besides, the bus organization and memory subsystem is very important and typically quite unusual in consoles compared to general purpose computers. Look at the current Wii CPU where the available benchmark data is all using vanilla SDRAM (!), whereas in the Wii, it primarily uses the 1T-SRAM. A world of difference, and I'd love to see some quantitative data. The PS3 uses its XDR with way higher effective bandwidth than a typical PC, particularly at the time when it was designed. If Nintendo would put the 970 core to use, it would be no surprise if the core itself would be somewhat adapted, but it would be really surprising if bus/memory controller didn't get a major overhaul or rather a complete replacement. Which would render older PPC970 benchmark data irrelevant.

It's available, it's compatible, it would be small so it would be cheap. A PPC970 derivative is an option, for sure. But Nintendo surprised me the last time around, so why shouldn't they do it again?

 

[Dargakis]

On the topic of noisy DVD drives, my Zephyr debug is astonishingly quiet. The fans are much louder than the drive, and yet it's a basic Hitachi 78 or 79, the same that I have in my Elite which is frightfully loud. Do the debugs run the drive at a slower speed?

yes, 4 speed

 

[pc999]

x86 - incompatible, too big, too power hungry(/unit performance if you want to quibble) and thus too expensive.
Power7 - too big, too expensive.
Other exotica - incompatible, too big, too power hungry, too expensive, lacking in tools.

Cant say for the others (we dont have enought info on them) but if you take 3 cores and the L3 cache (you "dont need" it in a single core configuration) from a Phenom II/i7 even if you double the L2 you will end up with a CPU with at the most 60-70mm^ @ 45nm (being pessimistic), that is very cheap and I guess that power consuption/heat would be low enought that a 32nm version would be possible in a console like Wii, with performance enought to run crysis(+?).

Others PPC architetures would be a good guess?

 

[Blazkowicz]

current single core on 65nm is 77.2mm² and pretty low power especially if you lower the voltage.

The dual core AMD Regor I was talking about is said to be 80mm² here : http://www.techpowerup.com/79451/AMD_Propus_and_Regor_Die_Sizes_Surface?.html


that cpu would make a great one for a console, mostly comparable to xenon or spanking it depending on the workload. but then you need a GPU with its own ram pool, or using the system ram with a sucky bandwith.
on the Wii 2 timeframe you would probably have an AMD Fusion with two bobcat cores instead..

 

[Squilliam]

Does anyone know what the likely cost of single-threaded vs dual/triple threaded designs, I.E the pros/cons of going single vs multi thread per core?

Also 80mm^2 on 45nm isn't bad at all.

 

[pc999]

Current C2D are also around 80-110 mm^, depending on L2 size, just giving a lot of room to costum work and extra cache in the "guess" above.

Does anyone know what the likely cost of single-threaded vs dual/triple threaded designs, I.E the pros/cons of going single vs multi thread per core?

Also 80mm^2 on 45nm isn't bad at all.

From what I recal:

Cons: draws more power, more heat, and a bit more on the cost (but not that much), it needs to be programed in a multithreaded environment (and some designs dont give you much)

Pros: It help geting the most of the cores, the more complex the core the better.

 

[Dargakis]

will xbox720 employ motion controls btw?
We have had some rumours about it getting a new controller voor "banjo kazuki three", maybe that was for the next xbox? who knows..
I think the controllers will stay the same, seeing how the ps3 controller has been around for every generation now, give or take rumble, dual analog or sixaxis

 

[Entropy]

current single core on 65nm is 77.2mm² and pretty low power especially if you lower the voltage.

The dual core AMD Regor I was talking about is said to be 80mm² here : http://www.techpowerup.com/79451/AMD_Propus_and_Regor_Die_Sizes_Surface?.html


that cpu would make a great one for a console, mostly comparable to xenon or spanking it depending on the workload. but then you need a GPU with its own ram pool, or using the system ram with a sucky bandwith.
on the Wii 2 timeframe you would probably have an AMD Fusion with two bobcat cores instead..

Just to make the differences perfectly clear, that Regor die size at 45nm is five times the die size of the current Wii CPU at 90nm. As with all x86 CPUs offered,
- it is binary incompatible with all codes for all Nintendo devices.
- it is unlikely that Nintendo will be able to order a variation with for instance a bus interface that is suited to simultanously low cost and high performance, or on-chip graphics at a Nintendo specified performance point, or... It's a PC chip, pure and simple. Nor is Nintendo likely to be able to determine lithographic process choices or whether or not to scale their particular processor to finer geometry when available. In this respect they would be in the same situation Microsoft was in with the XBox, but for no good reason.
- it is unlikely to offer particularly good performance/dollar or performance/watt. Hell, even now you can't find an x86 processor that performs significantly better than the Wii processor while drawing less than 5W, and at the same (estimated) pricepoint as that 15mm2 PPC chip. Even with two generations of lithographic advantage.

Even with the trade-offs Sony and Microsoft were willing to make, x86 processors were deemed unsuitable for console use. (And Microsoft had way more reason than anyone else to go x86). For Nintendo who have focussed on efficiency both in terms of cost and energy, I can't see that x86 is relevant even for wild speculation. Sure, Nintendo is difficult to predict, but a PC processor? Might as well propose an Itanium.

 

[pc999]

Well 1 core of Regor would be 2,5-3x as big in 45nm, at the worst 2x as big in 32nm, that would cost what, (worst case) more 5 dollars?

Anyway I was not predicting a X86 CPU for Wii2 (it is Nintendo so how knows), just pointing that a "high end" single core comparable to those on the best PC CPUs could be a great option.

Being both a great performer (in relative terms, as it wouldnt have any problems with a engine/workload like crysis) and cheap/low power.

Unfurtanately X86 CPUs are the best know so easier to compare.

Anyway it is quite possible they just update/grade a PPC970, like they did to the 750, but at 32nm it would be smaller than Broadway (and IIRC that CPU had some high heat/power requirements for higer speeds). Do anyone think they would have a CPU even smaller (unless a lot of L2 and costum features make is way into it).

 

[Entropy]

Anyway it is quite possible they just update/grade a PPC970, like they did to the 750, but at 32nm it would be smaller than Broadway (and IIRC that CPU had some high heat/power requirements for higer speeds). Do anyone think they would have a CPU even smaller (unless a lot of L2 and costum features make is way into it).

You're right that it can become both impractical and probably economically suboptimal to package a very small CPU on its own. At some point it makes sense to do a SoC, but then again you can't just throw together different functionality from different vendors for both IP and design reasons. This is one reason that Nintendo may change the architecture of the Wii quite a bit with their next generation, since just shrinking the current doesn't seem like a very viable path forward. They could go with updated CPU and GPU architectures (very possible) or go even further along the low cost and low power draw trajectory and do a SoC, at which point they may or may not opt to change instruction set, and if they do ARM is the only real option to sticking with PPC. But then backwards compatibility by emulation would be a challenge. If backwards compatibility is a concern for Nintendo, a PPC processor is the safest bet, and if they go that route my guess would be that they would also opt for a discrete GPU.

 

[Squilliam]

Current C2D are also around 80-110 mm^, depending on L2 size, just giving a lot of room to costum work and extra cache in the "guess" above.



From what I recal:

Cons: draws more power, more heat, and a bit more on the cost (but not that much), it needs to be programed in a multithreaded environment (and some designs dont give you much)

Pros: It help geting the most of the cores, the more complex the core the better.

Thanks.

Hmm... CPU architecture is a tough question to answer really. Possibly the toughest of them all as there are so many correct solutions to a single problem.