Nintendo Conference

[function]

How much heat does it generate?

More or less than the 65W of Power 7 (OoE 8 core, 32 threads, 2,4Ghz lots of edram a monster for any console standards)

or

20W (OoE 4 core, 16 thread 1,4Ghz lots of edram, still more gigaflops than Xenon, if you want to use that metric, still quite good the purpose)

You're comparing server chips with HUGE margins that can sustain shitty yields AND be binned for different profiles to a single-use chip that needs high yields and absolutely, positively has to be as cheap as possible.

It is not a reasonable comparison.

Also, 20W for the CPU alone would be far more than the cooling solution in the Wii has to deal with, and would probably be closer to the total heat output from the Wii U's CPU and GPU combined.

 

[AlNom]

Also, 20W for the CPU alone would be far more than the cooling solution in the Wii has to deal with, and would probably be closer to the total heat output from the Wii U's CPU and GPU combined.

Isn't the 20W for the entire Wii? The RAM and DVD drive use power too.

 

[ERP]

Either they have something to hide or at this point they prefer to wait till the next big gaming show (possible news at tgs?)

When have Nintendo ever "volunteered" specs.
I've only ever seen them publicly through developer leaks.

 

[patsu]

Nintendo president puzzled by investor reaction to Wii U:
http://www.reuters.com/article/2011/06/10/us-nintendo-iwata-idUSTRE7590A220110610

Nintendo President Satoru Iwata said he was surprised at the tumble in the company's share price following the unveiling of a successor to its smash hit Wii games console, adding that the new gadget had to be played to be understood.

...

"At the moment, there is a barrier between the Wii, which is seen as for casual users and the other companies' consoles, which are seen as for core gamers. We are questioning whether that barrier needs to be there," said Iwata.

...

 

[function]

Isn't the 20W for the entire Wii? The RAM and DVD drive use power too.

Yeah, the in-game at-the-wall power readings I've seen around the net vary between 18W and 22W. After PSU efficiency (I doubt they're 85%+ like good PC PSUs), Wifi, DVD, RAM, fan ... you can see how that little aluminium heatsink and little fan can handle things!

Hoping for a 20W CPU and 24 W GPU is probably a little optimistic.

 

[RancidLunchmeat]

"At the moment, there is a barrier between the Wii, which is seen as for casual users and the other companies' consoles, which are seen as for core gamers. We are questioning whether that barrier needs to be there," said Iwata.

Of course the barrier doesn't need to be there. As long as Nintendo doesn't need to make a profit. Let's face it, if the Wii were as powerful as the PS360 with an according price point, it never would have sold in the numbers that it did, and who knows when or if it would have become profitable.

Price was a huge factor in the ability of parents and grandparents to buy them as gifts for their children and grandchildren. Now, they were still extremely expensive toys, but not quite to the price point of consumer electronics like the PS360.

With lower power, you get a resulting lower price. But you also get lower performance which is what creates that 'barrier' for 'core gamers'.

Seems like a relatively simple equation to me.

 

[pc999]

You're comparing server chips with HUGE margins that can sustain shitty yields AND be binned for different profiles to a single-use chip that needs high yields and absolutely, positively has to be as cheap as possible.

It is not a reasonable comparison.

Also, 20W for the CPU alone would be far more than the cooling solution in the Wii has to deal with, and would probably be closer to the total heat output from the Wii U's CPU and GPU combined.

BTW I made a mistake and mixed up Power 7 and Power a2 info.

Second I dont expect a power 7, I just tried to show that there is very powerful designs at 45nm that have very low power consumption.

I also dont expect a 5830, but again itis much more power and 24 w.

Using a custom design they certainly ca get a better middle figure, dont you think

Anyway the Wii U case is a little bigger and not final, would be more interesting to know how much XB360 consumes too?

 

[dobwal]

BTW I made a mistake and mixed up Power 7 and Power a2 info.

Second I dont expect a power 7, I just tried to show that there is very powerful designs at 45nm that have very low power consumption.

I also dont expect a 5830, but again itis much more power and 24 w.

Using a custom design they certainly ca get a better middle figure, dont you think

Anyway the Wii U case is a little bigger and not final, would be more interesting to know how much XB360 consumes too?

I think the Wii-U can achieve PS360 parity very easily within the limited space. I just think people are pointing to tech and hawking on the performance characteristic of that tech creating a belief the Wii will be at a level of power thats a little too much given Nintendo's stated target.

You don't need at the minimum a HD 4850 to achieve PS360 visuals. A console with a 4850 and a comparable cpu would kill the the PS360.

When you really look at it, does Nintendo need an AMD R700 gpu to beat out an AMD R500 gpu? Does Nintendo really need the latest and greatest from IBM server tech to beat out IBM's own tech from 2005.

No not really. IBM based Power7 and AMD R700 tech may end up in the Wii-U but it won't have the awe aspiring power those techs are known for because those derivatives are targetting tech introduced a half of a decade back.

 

[function]

BTW I made a mistake and mixed up Power 7 and Power a2 info.

Second I dont expect a power 7, I just tried to show that there is very powerful designs at 45nm that have very low power consumption.

As I understand it, larger chips that are run slower can often obtain more performance per watt than smaller chips run faster, but at an increased cost for the processor.

The wikipedia page for the PowerPC A2 describes a 428 mm2 45nm chip with 1.43 billion transistors. The Xbox 360 Slim used a combined CPU/GPU consisting of 372 million transistors with a chip size of 169 mm2 (iirc) + edram.

Even if yields were similar the A2 will be a lot more expensive than the 360 processor.

Will Nintendo really want to use processors that are roughly comparable with the 360 for performance but cost several hundred percent more?

I also dont expect a 5830, but again itis much more power and 24 w.

The 5830 mobile uses a Juniper chip that was introduced 3 months earlier as the (desktop) 5750 and 5770 cards drawing a whopping 86 and 108 W respectively (and I bet a "power virus" of the kind the 360 S was tested with would have exceeded these figures).

That's three months of yield improvements and die harvesting before releasing the 5830 mobile.

Nintendo cannot do this because there's no other market where they can sell their custom processors for several months beforehand, and no other market where they can sell their perfectly operating but power hungry chips for 5 - 10 years while selling the Wii U.

Using a custom design they certainly ca get a better middle figure, dont you think

They'll probably get a worse result!

If they bin for only the most power efficient parts there will be nowhere else they can use their custom processors so they'll end up vastly increasing the price of their processors.

If you reject half of your viable processors you effectively double the price of your processors.

MS use a big heatsink and fan on the 360 Slim, they are probably able to use a large percentage of their viable chips. If Nintendo try and match them but using huge chips, most of which are rejected because of power consumption, they'll end up being unable to compete with the 360 on cost terms (don't forget that the new controller will be relatively expensive too).

In short: Nintendo are bound by the same need to have high yield, cost effective processors as MS and Sony. And comparisons to cherry picked mobile parts, or HUGE server processors are not meaningful.

 

[dobwal]

Nintendo cannot do this because there's no other market where they can sell their custom processors for several months beforehand, and no other market where they can sell their perfectly operating but power hungry chips for 5 - 10 years while selling the Wii U.

I agree with everything you said except maybe this part. I think Nintendo will use the latest but low-mid range mobile part from ATI. Simply for this reason, any chips that are defective may be able to be re-purposed as laptop or discrete htpc-centric cards.

 

[mczak]

The 5830 mobile uses a Juniper chip that was introduced 3 months earlier as the (desktop) 5750 and 5770 cards drawing a whopping 86 and 108 W respectively (and I bet a "power virus" of the kind the 360 S was tested with would have exceeded these figures).

That's three months of yield improvements and die harvesting before releasing the 5830 mobile.

I doubt it's got much to do with better yields and die harvesting.
Biggest improvement came from lower clocks and hence lower voltage. Second is the ddr3 ram instead of gddr5. And third die harvesting.
You can see on that amd page that the mobility HD5870 is 50W - this is somewhat comparable to desktop HD 5750 in performance, it's got same core clock, one simd more but lower gddr5 clock (and I bet the memory is not overvolted like on the desktop - probably uses low-voltage gddr5 too).
I see no reason why similar efficient performing parts couldn't be done for a console.

 

[AlNom]

What does the cooling setup look like?

 

[pc999]

As I understand it, larger chips that are run slower can often obtain more performance per watt than smaller chips run faster, but at an increased cost for the processor.

The wikipedia page for the PowerPC A2 describes a 428 mm2 45nm chip with 1.43 billion transistors. The Xbox 360 Slim used a combined CPU/GPU consisting of 372 million transistors with a chip size of 169 mm2 (iirc) + edram.

Is that for a 16core a2 or a 4core a2 I am guessing the first.

Anyway I am not really expecting to see a total (cpu+gpu) 1.43 billion transistors in Wii. But a total 800 millions wouldn't surprise me, as we all saw there is more complex chips with much more transistors (~25% more) using relatively low power like a 6830 (24w too 1080 transistors*).

I am really confident that they can get a good middle ground if they aim/build it to that, and that with 800 millions transistors it will perform better at the same clocks than PS360 with 400 millions.


Or do you think that if they build a single chip of 800 million transistors on low powoer silicone with some power requirements in mind it will use more power than a 1080 million transitors one and perform worst than a 400 millions one

http://www.notebookcheck.net/AMD-Radeon-HD-6830M.43739.0.html

 

[function]

I agree with everything you said except maybe this part. I think Nintendo will use the latest but low-mid range mobile part from ATI. Simply for this reason, any chips that are defective may be able to be re-purposed as laptop or discrete htpc-centric cards.

I suppose they could, but I expect Nintendo to have licensed the tech and be making processors at their fab of choice. I still think there's a good chance it's IBM and a single chip CGPU ...

I doubt it's got much to do with better yields and die harvesting.

They were good to go but waited three months to trickle the part out, missing the Christmas period, for what reason then?

Biggest improvement came from lower clocks and hence lower voltage. Second is the ddr3 ram instead of gddr5. And third die harvesting.

They could have done the lower clocks and lower voltage when the desktop part launched, like they did a year later when they recycled Juniper for the 6700/6800M.

You can see on that amd page that the mobility HD5870 is 50W - this is somewhat comparable to desktop HD 5750 in performance, it's got same core clock, one simd more but lower gddr5 clock (and I bet the memory is not overvolted like on the desktop - probably uses low-voltage gddr5 too).
I see no reason why similar efficient performing parts couldn't be done for a console.

Because Nintendo can't put chips into desktop parts or special OEM only cards that they sell by the hundreds of thousands or millions.

Is that for a 16core a2 or a 4core a2 I am guessing the first.

The IBM presentation and white paper only talk about a single 16 core chip. I can only assume that the 8 and 4 core parts are harvested parts, and it would make these staggeringly bad parts to compare to the Xbox CGPU.

If I've got this wrong and there are multiple chips, please correct me. These are the presentation and white paper linked to from Wikipedia:

http://www.power.org/events/2010_ISSCC/Wire_Speed_Presentation_5.5_-_Final4.pdf
http://www.power.org/events/2010_ISSCC/ISSCC_WSP_Paper_5_05r2.pdf

Anyway I am not really expecting to see a total (cpu+gpu) 1.43 billion transistors in Wii. But a total 800 millions wouldn't surprise me, as we all saw there is more complex chips with much more transistors (~25% more) using relatively low power like a 6830 (24w too 1080 transistors*).

Assuming similar transistor density and yield, Nintendo would be paying around twice as much for their processors in order to meet the stated goal of handling multiplatform games and ports. In reality yields would probably be worse.

Also, the 6800M series is yet another Juniper chip, and is still also able to be put in desktop 6700 parts.

If Nintendo go with a custom chip then they cannot do this.

I am really confident that they can get a good middle ground if they aim/build it to that, and that with 800 millions transistors it will perform better at the same clocks than PS360 with 400 millions.

Or do you think that if they build a single chip of 800 million transistors on low powoer silicone with some power requirements in mind it will use more power than a 1080 million transitors one and perform worst than a 400 millions one

Processors have to be paid for. They also have to cooled.

[Edit] Nintendo must keep the cost of the processors down, which means they can't be too big and they can't throw too many away, especially not fully working ones! The cooling and the cost of the cooling solution is also critically important. Server parts can get away with costing far more per chip, and can be used in different configurations (disabled units, power envelopes). Mobile parts can also be used in different configurations (disabled units, power envelopes, even as desktop parts).

 

[pc999]

The IBM presentation and white paper only talk about a single 16 core chip. I can only assume that the 8 and 4 core parts are harvested parts, and it would make these staggeringly bad parts to compare to the Xbox CGPU.

If I've got this wrong and there are multiple chips, please correct me. These are the presentation and white paper linked to from Wikipedia:

http://www.power.org/events/2010_ISSCC/Wire_Speed_Presentation_5.5_-_Final4.pdf
http://www.power.org/events/2010_ISSCC/ISSCC_WSP_Paper_5_05r2.pdf



Assuming similar transistor density and yield, Nintendo would be paying around twice as much for their processors in order to meet the stated goal of handling multiplatform games and ports. In reality yields would probably be worse.

Also, the 6800M series is yet another Juniper chip, and is still also able to be put in desktop 6700 parts.

If Nintendo go with a custom chip then they cannot do this.



Processors have to be paid for. They also have to cooled.

[Edit] Nintendo must keep the cost of the processors down, which means they can't be too big and they can't throw too many away, especially not fully working ones! The cooling and the cost of the cooling solution is also critically important. Server parts can get away with costing far more per chip, and can be used in different configurations (disabled units, power envelopes). Mobile parts can also be used in different configurations (disabled units, power envelopes, even as desktop parts).

Well if it is custom they could have a custom native 4 core a2, as one possibility, not trying to say they will just to give a example. The same could be said about the 6830, they could do a smaller more efficient chip.

Naturally those native versions, less complex and smaller would have much better yields and a much more mature and low power process leading to better hea/consuption management, making them a lot cheaper and with no need to harvest chips.



BTW Usually at the begining of a console life they cant use all the chips but that improves, and with 32-28nm around the corner that can change fast.

The console is also expected to be pricier to produce, and a bit bigger.

 

[pc999]

BTW HW "details"

http://www.develop-online.net/news/37999/Programming-guru-gives-insight-into-Wii-U-tech

“The multicore architecture of the console is a natural fit for our in-house HD engines, such as the Anvil engine,” Parenteau said during a developer roundtable at E3.
Parenteau, one of the few people in the industry that knows of the Wii U’s deepest hardware specifications, said the console’s “large memory capacity” would allow for enhancements to Anvil tech.
He said the extra processing power would allow for precalculating data and increasing cache sizes.

“All the graphical shaders used in Assassin’s Creed are fully functional on Wii U”, Parenteau added.
He claimed that developers with Wii experience will find a familiar set of APIs in Wii U.
“New features, such as multicore processing, will extend the APIs in a natural way with low-level and straightforward calls,” he added.

 

[function]

Well if it is custom they could have a custom native 4 core a2, as one possibility, not trying to say they will just to give a example. The same could be said about the 6830, they could do a smaller more efficient chip.

Naturally those native versions, less complex and smaller would have much better yields and a much more mature and low power process leading to better hea/consuption management, making them a lot cheaper and with no need to harvest chips.

A chip being custom doesn't automatically make all concerns about heat and power go away. In order to be able to use a good proportion of their chips they may need to settle for characteristics that are less favourable than high end mobile or server parts even though the parts are custom.

If Nintendo focus exclusively on performance per Watt (unlike Sony and MS) I think it'll bite them. Likewise, if they prioritise form factor way above cooling (like laptops have to and consoles don't) they could end up paying for a costly cooling solution for the lifetime of the device.

I'm not saying they can't make something broadly as capable as the PS360 while using less power, but matching a 5830M (for instance) seems really unlikely.

BTW Usually at the begining of a console life they cant use all the chips but that improves, and with 32-28nm around the corner that can change fast.

The console is also expected to be pricier to produce, and a bit bigger.

I don't think the Wii ever had shrinks; Nintnedo's philosophy seems to be that the console should be a viable from day one. If this is applied to the Wii U then we won't be seeing a race to shrink processors and rapidly diminishing cooling requirements, we'll have a modest and profitable console from day one.

BTW HW "details"

http://www.develop-online.net/news/37999/Programming-guru-gives-insight-into-Wii-U-tech

An RV730 or Redwood (104 mm2 @40nm) level part should be able to manage that, even if it was clocked down under 500 mHz to save power (RSX manages with less fillrate and tri-setup than the 360).

Edit: referring to all the shaders running on Wii U bit.

 

[mczak]

They were good to go but waited three months to trickle the part out, missing the Christmas period, for what reason then?

It's a pretty low volume part (on the mobile side) anyway so I'm not sure missing some season matters much. It's also possible the parts got better simply because of TSMCs trouble with 40nm, which isn't going to happen with IBM 45nm...
There could be different reasons why it was launched later on mobile, the mobile release dates usually don't have anything to do with when notebooks appear neither and often are different to desktop launch dates (sometimes earlier, sometimes later). Simply deducing it was launched later due to die harvesting is a bit of a stretch.
Maybe they were simply too lazy to just test the chips at a lower voltage / clocks earlier...

Because Nintendo can't put chips into desktop parts or special OEM only cards that they sell by the hundreds of thousands or millions.

Sure don't disagree. But the differences between good and bad dies likely aren't all that much, you just set the bar low enough that all match it in the end. So if you think only half the chips would qualify for some given TDP and clock, probably all of them would if you lower clock by 5% and hence can lower voltage slightly too.

 

[glw]

A chip being custom doesn't automatically make all concerns about heat and power go away. In order to be able to use a good proportion of their chips they may need to settle for characteristics that are less favourable than high end mobile or server parts even though the parts are custom.

Assuming they are using the POWER A2 then Nintendo would almost certainly want a custom part as Nintendo has no use for the chip-to-chip IO for SMP systems, the accelerators for server and network tasks, nor the numerous network interfaces that are supported. Uncore parts take up more than half the die on the Wire-Speed Processor.

 

[pc999]

I don't think the Wii ever had shrinks; Nintnedo's philosophy seems to be that the console should be a viable from day one. If this is applied to the Wii U then we won't be seeing a race to shrink processors and rapidly diminishing cooling requirements, we'll have a modest and profitable console from day one.

Cant they do something like change strategy if it does make sense

Or are they so stupid that must do only the same they did in the past?

Besides how much would a die shrink save on the Wii?

An RV730 or Redwood (104 mm2 @40nm) level part should be able to manage that, even if it was clocked down under 500 mHz to save power (RSX manages with less fillrate and tri-setup than the 360).

Anything in the point to Wii U being less or even just equal in terms of performance? Or does it sugest that it might be somewhat (how much is unknow), a better performer.

Assuming they are using the POWER A2 then Nintendo would almost certainly want a custom part as Nintendo has no use for the chip-to-chip IO for SMP systems, the accelerators for server and network tasks, nor the numerous network interfaces that are supported. Uncore parts take up more than half the die on the Wire-Speed Processor.

Thanks for the info, it just shows that a custom work can go very high on savings of power,price and still good performance.