Feasibility Of a Portable Wii U?

[Grall]

I remember GameCube well enough. When has Nintendo ever publicized things like a shrink?

Ah, I see the reasoning you're using; because there wasn't any announcements, it happened. Well, that makes sense...

Buy a Wii from the store now and open it up. You'll see.

Right. Why don't you just post a weblink backing up your claim? Considering your level of confidence that ought to be easy enough for you.

...Ah. Forgot. Because Nintendo doesn't announce these things, you can't. Too bad!

Circular logic at its finest.

 

[BobbleHead]

I'm not buying and disassembling a Wii to try to prove your point. Fortunately there are available shots: http://www.eurogamer.net/articles/df-hardware-nintendo-wii-mini-review You can't really conclude anything from pictures of packaged chips except a maximum size, and I don't feel like trying to cross reference that with some other component whose size is known (that and I can't find dimensions for the mini). But at least you can see that they're still using two separate CPU and GPU chips. This site says that there are unconfirmed reports of shrinks. They do show that power consumption went down, and while that could be in part due to a reduction in other components I suspect at least one of the two major chips got at least one shrink. On the other hand, you'd expect better improvements if both were shrunk all the way down from 90 to 40nm.

Okay then... Compare the picture in your link with the 3rd one in http://wii.mmgn.com/News/Inside-the-Wii. The big capacitor below the main packages in yours is fairly standard and can be used as a reference to the two below the GPU in my link.

In the mini, the cap is 41 pixels wide, the GPU is 168 wide (4.1x) and the CPU is 86 wide (2.1x).
At launch, the cap is 52 wide, the GPU is 240 wide (4.6x) and the CPU is 162 (3.1x).
Both the CPU and GPU are relatively smaller packages than the original. Somehow they shrunk.
Even if you don't believe the capacitor is the same size in both cases, the ratio between the CPU size and GPU size is different. 168/86=1.95 and 240/162=1.48. They obviously are not the originals.

Of course this only shows that the packages are different. Do you really think they shrunk the package without shrinking any of the die inside? Package size is dictated by either the size of the die on it or the number of BGA balls necessary to cover power and IOs. The only way they could make it smaller is if the die got smaller or if they needed fewer balls. They couldn't significantly change the functional pinout because they didn't eliminate any of the major interfaces, therefore the only way the ball count goes down is if the power requirements go down. That means a shrink.

 

[Grall]

No, it doesn't mean a shrink without actually verifying the dies themselves. You're assuming it shrunk, but you have no proof of that. Using a cap as indicator of relative scale also seems risky, these things do come in different sizes after all. Go for the USB port instead, that one's sure to be reliable.

Anyways, pixel-counting will only give the appx. size of the package, without saying anything about what's inside. The hollywood package size is largely dictated by what's in it (IE, the dies of the GPU and its 1T SRAM memory, not the number of balls underneath. Wii originally had two RAM dies on the package. These could just as well have changed form factor instead of the GPU, thus allowing a smaller package to be used.

 

[BobbleHead]

Would they ever change the form factor to make the GPU package bigger? Pretty unlikely, isn't it?

Let's assume you're right and the GPU package got smaller because they only changed the form factor, and that made it smaller. If the GPU package were made smaller without changing the CPU package, the ratio between the two should have gotten smaller. Those two pictures show the opposite. Originally it's 240/162 = 1.48/1 and now it is 168/86=1.95/1. The GPU is now quite a bit bigger relative to the CPU.

Now we have 4 possibilities:
1. The GPU is exactly the same but the CPU package got much smaller without changing the die.
2. The GPU is a smaller package without die changes and the CPU got an even larger package reduction without die changes.
3. The CPU is exactly the same and the GPU package got much bigger.
4. Both the CPU and GPU had shrinks to reduce cost and both wound up on smaller packages.

Which do you think is more likely?

 

[Grall]

You're just engaging in polemics. Come back when you have solid fact to back up your position (IE: die shots.)

 

[Deleted member 13524]

FWIW, I do remember something on the webz about the Gamecube changing to a smaller process by the time it sold for 120€.
I think this appeared in one of those investor meetings or something.

 

[BobbleHead]

You're just engaging in polemics. Come back when you have solid fact to back up your position (IE: die shots.)

It doesn't matter, we've gone off topic long enough. You can believe what you want to believe.


Back on topic, it probably wouldn't be until 16/14nm that they could hope to make it into a portable, and they would have to integrate CPU and GPU into a single die. That is pretty far away.

 

[Exophase]

Of course this only shows that the packages are different. Do you really think they shrunk the package without shrinking any of the die inside?

Of course. If the CPU is only 19mm^2 then neither the original nor final package is anywhere close to that small.

Shrinking the die/migrating it to a new manufacturing processor has got to be a lot more expensive than just shrinking the package as better packaging and PCB assembly technology is available (cheaply, with low failure rates). But with the latter alone they capture most of the benefits.

Package size is dictated by either the size of the die on it or the number of BGA balls necessary to cover power and IOs. The only way they could make it smaller is if the die got smaller or if they needed fewer balls. They couldn't significantly change the functional pinout because they didn't eliminate any of the major interfaces, therefore the only way the ball count goes down is if the power requirements go down. That means a shrink.

You're missing one: the size and pitch of the BGA balls shrunk while the die stayed the same. There isn't a one to one relationship between the die size or the die's pads and the size and pitch of the balls or leads in the package.

I'll attempt an absolute size comparison. There are some better PCB screenshots of older Wiis here:

http://hackmii.com/2009/08/wii-hardware-a-history/

This demonstrates the first smaller package for the CPU, and also that the capacitor sizes are a poor reference. So I'll take Grall's suggestion and use USB instead. USB-A sockets a little over 12mm wide. I used the middle picture for the original dimensions since it's less distorted. In this shot the USB is just on the edge of the picture so I estimated it, it could be a little off. But this is what we get:

Original Wii USB: 100 pixels wide
Original Wii CPU: 172x158 pixels (21x19mm^2)
Original Wii GPU: 246x226 pixels (30x27mm^2)

2008 Wii USB: 92 pixels wide
2008 Wii CPU: 118x122 pixels (15x16mm^2)
2008 Wii GPU: 222x246 pixels (29x32mm^2)

Wii Mini USB: 27 pixels wide
Wii Mini CPU: 32x32 pixels, 14x14mm^2
Wii Mini GPU: 58x58 pixels, 26x26mm^2

According to reports, Wii's original CPU die (at 90nm) is 19mm^2 and the original GPU die is 70mm^2. That works out to about 4x5mm^2 and 8x9mm^2 respectively.

The die size for Wii could be wrong, but it fits well with the die size report for Gamecube, and more importantly fits with the die size for Wii U that can actually be measured. So assuming that these sizes are in fact correct we see that even the current packages are much larger and can still easily accommodate an original die without shrinks.

 

[BobbleHead]

From http://www.chipworks.com/blog/technologyblog/2013/02/04/looking-at-the-wii-u-graphics-processor/

The (WiiU) die size is ~12.0 x 12.5 mm. (~150 sq. mm.), compared with ~117 sq. mm. for the earlier Wii Black.

The original Wii GPU had 2 big die, 72 sq mm of GPU and 95 sq mm of memory. (http://forum.beyond3d.com/showthread.php?t=62651)

The Wii Black came out in late 2009 or early 2010, combining the two into a single 117 sq mm die made on a smaller process.

Nintendo doesn't shove that type of information in your face. They talk about tech specifics as little as possible because they want to make it all about the games and the overall experience. They never officially said anything about the Wii GPU or CPU. All of the tech details were uncovered elsewhere. Same thing for the WiiU. They've said the minimum (3 core CPU from IBM, GPU from AMD) so far and that is not likely to change anytime soon.

If they made a portable WiiU 6-8 years from now you might hear about how they used 14 nm and TSV memory to fit the power of WiiU into a tablet, but that's about all they would say. Chances are they still would not tell you the clock speed or shader count.

 

[Exophase]

Frankly, the reason why I always thought Nintendo never shrunk isn't because they didn't announce it but because I couldn't ever find someone online who said so - including on the die size thread you just linked to, which must have been updated with those figures relatively recently. And given the lack of information they're probably just based on the comment in the Wii Mini review where the reviewer says his sources claimed two shrinks, and the Wii Black GPU size which was only publicly released by Chipworks today.

It's not like anyone ever needed a reveal by Nintendo for this information. Someone simply needed to keep popping head spreaders and measuring the thing themselves. I guess Wii was just never interesting enough to encourage people to do that. But I don't need to be told that Nintendo doesn't reveal anything about their hardware anymore.. I haven't been living under a rock.

I'm still a little skeptical that they really shrunk to 40nm for the Wii Mini, because it isn't reflected in the power savings (although without further information I still have no reason to believe they ever shrunk the CPU die). Furthermore, for an alleged transition from 90nm to 55nm plus a die integration the savings of only 43% area reduction is underwhelming.

I also still stand by what I said earlier - package size doesn't really say anything about die size. For further illustration, here's the original Wii's GPU package delidded:

http://www.bunniestudios.com/blog/?p=227

If the package size was dictated by the balls underneath I don't see how shrinking the dies would have helped if they couldn't also decrease the ball size/pitch. And if they could do the latter I don't see why they'd have had to shrink the dies to do so either.

 

[BobbleHead]

Someone simply needed to keep popping head spreaders and measuring the thing themselves. I guess Wii was just never interesting enough to encourage people to do that.

That's why it has been difficult to argue this point and I should've just kept my mouth shut originally. I know it happened, but there isn't much on the internet to point to as "proof." It's physically out there because they sell millions of them, but someone has to pop it open and see. The ChipWorks guys did it because it's their job. They don't publicize everything they do, but you can be sure their clients know about it when it happened. MS/Sony also must have known.

I don't think it's because the Wii wasn't interesting. It was because Nintendo didn't use it for publicity or change the Wii form factor at the time. When MS or Sony switched to a smaller or quieter box it made people ask what changed and it gets investigated. Without some sort of trigger to make you want to open it up why would you bother? As far as you know there is no reason to think anything has changed.

If the package size was dictated by the balls underneath I don't see how shrinking the dies would have helped if they couldn't also decrease the ball size/pitch. And if they could do the latter I don't see why they'd have had to shrink the dies to do so either.

Keeping the same ball size and pitch, a die shrink can reduce the number of total balls required. You still need the same number of functional IO balls, but a smaller die requires less power. That allows you to reduce the number of power/ground balls. Combining 2 die into 1 creates additional savings. Converting a die-to-die interface into an on-die interface eliminates all of the IO power for that interface. Consolidating onto a single die can also allow you to eliminate separate voltage rails for the two dies. All those cuts let you move to a smaller package with the same ball size and pitch.

 

[DuckThor Evil]

I'm still a little skeptical that they really shrunk to 40nm for the Wii Mini, because it isn't reflected in the power savings

It isn't? Eurogamer specified a drop from 18W to 13W. In a such low power device other components such as the DVD-drive etc. consumes quite a bit percentage wise from the total draw. If the rest of the components are consuming close to what they used to, the GPU and CPU power draw drop is very substantial. Perhaps from 10W to 5W or something like that.

 

[Exophase]

It isn't? Eurogamer specified a drop from 18W to 13W. In a such low power device other components such as the DVD-drive etc. consumes quite a bit percentage wise from the total draw. If the rest of the components are consuming close to what they used to, the GPU and CPU power draw drop is very substantial. Perhaps from 10W to 5W or something like that.

They weren't specific on the testing but I didn't think it was done while the drive was spinning, although it could have contributed to a long term average. What else would use a lot of power? The RAM would contribute, but there's a very good chance the RAM used is more efficient than what was in the launch hardware.

Unless this chipset is in other hardware or Nintendo has more ambitious future plans for it it seems strange that they'd do a shrink just for the Mini, then only sell it in Canada..