Commission IBM to design, same difference.

What I am suggesting is that what is available to us commercially (Power7 at 45nm) is not necessarily what is available to their partners for future products. We both agree that Nintendo's CPU is going to be based on some IBM Power architecture (probably a lower end power7 from what IBM hinted at with the eDRAM and same technology as watson references) and we know that IBM's 32nm process is available for chip design (whether you design your own chip or contract somebody else to do it) so there are no technical barrier to have a custom Power7 be designed for IBM's 32nm process.

That IBM doesn't yet produce power7 (or a successor) processors at 32nm today has no bearing as we are talking about what they have available today to design tomorrow's CPU's. IBM is not gonna tell their clients to wait until they have finished to design their power7 processor at 32nm before they begin designing theirs.

Anyway, this was just an example of Nintendo not using cutting edge technology. A good reason for not doing so is to avoid the yield ramp up problems typical of new processes. For example the Radeon HD 5000 series was seriously constrained at launch because TSMC's 40nm process was not mature enough (and it was even worth for Nvidia's 400 series at launch). Nintendo will need to sell a large number of console near the launch to make it an attractive platform to third parties, being out of stock because you are making loads of consoles and selling them all (like the Wii) is one thing, being out of stock because your partner cannot manufacture enough of a critical component is quite a different thing. 

Personally I would love to be proven wrong and to have Nintendo have their GPU manufactured at an aggressive process (32 or 28 nm) but I do not expect it to be so and the higher process will mean that a smaller processing power will fit in the same TDP.

@ethomaz: I know, I was not comparing specific process but generation, with 40 and 45  nm processes being close enough to be in the same generation, like PS3 and 360 are in the same generation even though they are different.

I was using Nintendo's choice of a 45nm process for its CPU as a good indication that it is more likely to choose a 40nm process for its GPU, which in turn has some implications wrt power dissipation especially when combined with the size of the console and the power usage of a stock 4870.

And yes, I know the 4870 is not at 40nm, I just have serious doubt about the feasibility of using a 4870 shrunk down to 40nm in such a small console. If I am wrong and they choose to use a 28nm process then that probably would be enough of a shrink to make it probable (imo).

Viper1 said:

My only rebuttal to that is the simple fact that IBM themselves are not using 32 nm for their Power7 CPU's.  Why not?    Would it not bein their best interst to produce their most pwoerful and energy hungry chips on their smallest process available?  The fact that they do not tells me that IBM knows something we don't regarding the chips and what process to use it with.

I've never seen a chip maker not use their most cutting edge process on their most cutting edge chips unless the yeilds are bad.

My only rebuttal to that is the simple fact that IBM themselves are not using 32 nm for their Power7 CPU's.  Why not?    Would it not bein their best interst to produce their most pwoerful and energy hungry chips on their smallest process available?  The fact that they do not tells me that IBM knows something we don't regarding the chips and what process to use it with.

I've never seen a chip maker not use their most cutting edge process on their most cutting edge chips unless the yeilds are bad.

The SOI is a better choice over the bulk... so the 45nm SOI will be used to make the Wii U's CPU instead 32nm bulk.

Sri Lumpa said:

Isn't IBM's 32nm SOI process released yet? From this paper we know that IBM were developing their 32nm SOI with Global Foundries (and Freescale).

AMD's new Llano processor is produced on a 32nm SOI by Global Foundries.

I find it hard to believe that GloFo and its customers has access to a 32nm SOI process and IBM doesn't when they are part of the same alliance and IBM generally the one driving the research.

I have no difficulty whatsoever to believe that said process is not mature yet and that Nintendo doesn't think it is likely to be mature by the time production of Wii U CPUs is scheduled to start, but GloFo having a 32nm SOI process before IBM? Seems unlikely (unfortunately it is harder to find information about IBM processes on the web as they do not offer retail level processors).

2.Llano is already released? What I know just reviewers received it... the launch was scheduled to June 15th... and yes, it is the fisrt CPU to use 32nm SOI from IBM. Llano was delayed some times because AMD/GloFo had problems to implement the 32nm SOI.

So just Llano uses 32nm SOI... all others Fusion uses 40nm Bulk.

How is it a rebuttal though? I'm saying Nintendo has both IBM's Power7 architecture and IBM's 32nm process available but chose to go 45nm instead, likely to avoid bad yield on a non-mature process. You are saying IBM not using 32nm for power7 today hints at bad yields. I don't see a contradiction to my position there.

IBM may know something we don't but I doubt they know something Nintendo doesn't, and that knowledge would be taken into account in choosing 45nm instead of 32nm.

So, where does it contradict my theory that Nintendo preferred the more mature 45nm process over the less mature 32nm one?

From what I gather from what Sri Lumpa is saying, the design decisions Nintendo makes on its custom chips are based on different factors than those made by IBM on their own projects. Since the WiiU's chip is a custom design for a different timeline and strategy, their choices may not follow the same logic as those of IBM.

If using 32 nm is nothing more than a business decision, then why isn't IBM using 32 nm for their Power7 chips?