Wii U hardware discussion and investigation *rename

[Dogface]

They've said the same CPU as in Watson, which it clearly isn't. And as IBM haven't shown any small-scale, low power POWER7 CPU yet AFAIK, it wouldn't be wrong for people to understand talk of a POWER7 CPU being in WiiU as being a monster server CPU.

No. You can't expect 100+ participants to read every technical documentation in the world relating to a subject. Where someone's understanding is limited, someone more knowledgeable can point to the right information and quote+link. This way we share specialist knowledge from 100+ individuals instead of needing everyone to go learn everything personally.

With that out the way, reference to key components of the POWER7 architecture and their usefulness in a low power console would be very beneficial. I went looking for power consumption and thermals and couldn't find anything other than "twice as much performance per watt as POWER6". Any info showing POWER7 can be operated with 3 or 4 cores at...I dunno, ~50 watts, would be an important bit of info in favour of a POWER7 processor fitting Wii U.

Right, which is one of the key concerns about P7's relevance in a console. That L3 is for working with large datasets. What use is 16+ GBs of eDRAM L3 in a game console? Maybe it is an ideal solution, but it needs someone to make a case for it.

its 512 GB/s l3 shared by 8 cores. On the 8 core model.

Yes, Ibm most certainly HAS announced a small scale power 7 model that can go into games consoles, one that draws less than half the power, and generates less than half the heat, than a 2 core p6, IN their power 7 documentation. Ibm's ENTIRE document talks about the low power low heat low cost benefits of the chips design.

I'm not asking you to read every document on every component in the wii u. Im asking you to read ONE document on the ONE component you have admitted to blindly, ignorantly arguing about.

Read it.

 

[kagemaru]

I'm not asking you to read every document on every component in the wii u. Im asking you to read ONE document on the ONE component you have admitted to blindly, ignorantly arguing about.

Read it.

That's rather harsh. Shifty is definitely not one of the blind or ignorant members here, so a bit of respect should be in order.

Unless you've worked on the CPU in the Wii-U, there's no reason to be taking comments about the CPU so personally.

 

[Kaotik]

They've said the same CPU as in Watson, which it clearly isn't. And as IBM haven't shown any small-scale, low power POWER7 CPU yet AFAIK, it wouldn't be wrong for people to understand talk of a POWER7 CPU being in WiiU as being a monster server CPU.

They've said
1) It uses "same Power7 chips" ( https://twitter.com/IBMWatson/status/78473693843562498 )
2) Custom chip using same processor tech as Watson ( https://twitter.com/IBMWatson/status/240241146213842944 )

Not that it's Watson, unless you see "same Power7 chips" as that.

 

[Dogface]

That's rather harsh. Shifty is definitely not one of the blind or ignorant members here, so a bit of respect should be in order.

Unless you've worked on the CPU in the Wii-U, there's no reason to be taking comments about the CPU so personally.

Its not harsh, its blunt.

And it doesn't matter WHO you are, if you are arguing about something you have no knowledge on, you are blind and ignorant on that subject.

THAT SUBJECT. Not as a person, but that SUBJECT.

This is why they make technical documents. You can't possibly know everything about someone elses design, you have to read about it. Particlularly something as different as p7.

Now, he didn't say he didn't have access to the documents, he flat out said he refused to read them. A very immature notion for someone who appeared to be rather passionate about the subject.

In that case, a little bluntness can go a long way to getting minds back into gear, opening closed minds to make progress.

Everything hes brought up is covered extensively in ibms documentation.

He just needs to read it. And him flat out saying NO is just not condusive to progress.

 

[AlphaWolf]

I expect your stay will be short.

 

[kagemaru]

Its not harsh, its blunt.

And it doesn't matter WHO you are, if you are arguing about something you have no knowledge on, you are blind and ignorant on that subject.

THAT SUBJECT. Not as a person, but that SUBJECT.

This is why they make technical documents. You can't possibly know everything about someone elses design, you have to read about it. Particlularly something as different as p7.

Now, he didn't say he didn't have access to the documents, he flat out said he refused to read them. A very immature notion for someone who appeared to be rather passionate about the subject.

In that case, a little bluntness can go a long way to getting minds back into gear, opening closed minds to make progress.

Everything hes brought up is covered extensively in ibms documentation.

He just needs to read it. And him flat out saying NO is just not condusive to progress.

To be fair, I understand the blunt direction. I also understand being bothered by discussions over matters that people are clueless on. We see it on the net all the time, including here.

However, Shifty has a very good point that it would be more beneficial to the discussion if the more knowledgeable few just did a direct quote and link to the specific info in question than to expect people to read every document covering every topic at hand. I understand you only expect us to read "ONE" document, but your mentality would also apply to every subject and that's just not a realistic expectation.

I understand blunt, I understand being annoyed by ignorance, I'm exactly the same way, but neither really applies here.

 

[Dogface]

I expect your stay will be short.

Since not a single person here has bothered to educate themselves on the subject matter they are debating about during the course of this and when given the oppurtunity for hard information relevent to every question they asked, flat out refused the information....

I suspect my stay here will be short as well.

I mistook this forum for what it used to be. Now its yet another internet arena for slinging idle unsupported opinions, and the people within aren't even interested in the actual information they are arguing about, just the argument.

Have fun arguing. When the systems torn down, think of me.

 

[willardjuice]

Since not a single person here has bothered to educate themselves on the subject matter they are debating about during the course of this and when given the oppurtunity for hard information relevent to every question they asked, flat out refused the information....

I suspect my stay here will be short as well.

I mistook this forum for what it used to be. Now its yet another internet arena for slinging idle unsupported opinions, and the people within aren't even interested in the actual information they are arguing about, just the argument.

Have fun arguing. When the systems torn down, think of me.

I'll unban you if the WiiU's cpu is actually Power 7 (and not something loosely based off Power 7 that Nintendo claims is Power 7). And if by some miracle the great and knowledgeable Dogface is wrong, you can PM Shifty an apology and he can decide if he wants to unban you (doubtful).

 

[AlphaWolf]

Why would anyone think of the guy who offered nothing? It's a discussion forum.

 

[Shifty Geezer]

its 512 GB/s l3 shared by 8 cores. On the 8 core model.

I don't see that being of particular value to a console CPU. For GPU that BW obviously has massive benefits. But for CPU workloads, how much benefit would it be? I can see it's use in AI. Most of the console workloads we've talked about this gen have been stream-processing, but that's no doubt in considerable part because the CPUs are designed that way. Still, random access pattern 4 MB datasets strike me as a niche function. It'll just take one of resident devs to set me straight on that though.

Yes, Ibm most certainly HAS announced a small scale power 7 model that can go into games consoles, one that draws less than half the power, and generates less than half the heat, than a 2 core p6, IN their power 7 documentation. Ibm's ENTIRE document talks about the low power low heat low cost benefits of the chips design.

What chip's that? The smallest P7 I've found is the 6 core 750. A quote would be useful. Regards thermals, IBM have only described P7 compared to POWER6 AFAIK, which was a hot chip. They haven't given actual thermal figures, so we don't know how it compares to other well-documented processors.

I'm not asking you to read every document on every component in the wii u. Im asking you to read ONE document on the ONE component you have admitted to blindly, ignorantly arguing about.

You miss the point. The principle of a discussion forum means people share their understanding and research instead of everyone going and doing their own thing.

You want me to read that tech document to understand the POWER7 option. Okay. Then on principle I also have to read the Intel documents on Atom, i3, i5, and i7 to understand them as options for consoles. I have to then go read up everything about ARM. I also have to go find and read the Global Foundries and TSMC documents and what manufacturing options there are. Then read up all the GPU stuff. That's an awful lot of time being dedicated to learning what's just a hobby interest.

Rather than read the whole document (162 pages of this red paper for one) for every tech, I typically look to technical websites to read them and digest them into more manageable summaries. I then look to specialist knowledge from those with more research or personal experience to build on that. In a discussion, I have my understanding which I'm well aware can be wrong (and everyone posting on a discussion forum should be aware that any post could be wrong), which is where someone comes along and corrects me.

If you've read through the whole POWER7 documents, then you are in a position to quote a few important key points that help educate us all to our misconceptions about POWER7 as a console CPU.

They've said
1) It uses "same Power7 chips" ( https://twitter.com/IBMWatson/status/78473693843562498 )

Not that it's Watson, unless you see "same Power7 chips" as that.

Yes. The chip is specifically the implementation of the architecture in an arrangement of cores on a silicon die (or dies) in a chip package. It can't be the same POWER7 chip without being the same number of cores with the same amount of cache and everything else. Thus that tweet is saying that it's the 8 core 750 in Wii U. This is highly implausible. Hence the reference to POWER7 can't be that exact. As Dogface says, POWER7 doesn't have to just be these monster chips (although the eDRAM seems an essential part of any P7 CPU) and there could be a small form P7 that would fit a lot more of the tweets and statements we've heard. Such a part would probably be custom, unless someone can find a particular reference to this small console-friendly version that Dogface reckons has been described by IBM.

 

[Graham]

I'll unban you if the WiiU's cpu is actually Power 7 (and not something loosely based off Power 7 that Nintendo claims is Power 7). And if by some miracle the great and knowledgeable Dogface is wrong, you can PM Shifty an apology and he can decide if he wants to unban you (doubtful).

Dogface, if you are reading this - I support your ban. I imagine you could be a very valuable member of this forum as you clearly are a knowledgeable person - however the attitude you have shown (in just 4 posts!) has demonstrated you lack the maturity to engage in a constructive conversation.

Unfortunate. Best of luck, I trust you learn from the experience.

 

[Shifty Geezer]

On the subject of POWER7 as a console CPU, Acert93 started one of his well-documented threads on the viability of POWER7 as an architecture for a console CPU here. and with some important datapoints : 8 cores at 4 GHz is consuming 250W in a 567 mm^2 die @45nm. A true 'Watson' POWER7 in WiiU is out of the question, and we also see that there'll have to be significant changes to the chip to produce a small, low wattage console part. I can see a 4 core, low clocked part being an option, but I'm still unsure of the value of the eDRAM.

 

[bgassassin]

I haven't posted in this thread in a minute. Personally I think the best chance of a "direct" relation to POWER7 would be the core with 2MB of L2 cache being some kind of modified P7 core. Other than that, the Watson talk means nothing to me anymore.

 

[liolio]

On the subject of POWER7 as a console CPU, Acert93 started one of his well-documented threads on the viability of POWER7 as an architecture for a console CPU here. and with some important datapoints : 8 cores at 4 GHz is consuming 250W in a 567 mm^2 die @45nm. A true 'Watson' POWER7 in WiiU is out of the question, and we also see that there'll have to be significant changes to the chip to produce a small, low wattage console part. I can see a 4 core, low clocked part being an option, but I'm still unsure of the value of the eDRAM.

A while ago Aaron Spink spoke about a crossing point in performance between SRAM and EDRAM.
Sadly he has given more details. I was at first thinking about frequencies, edram if the clock speed is low enough doesn't act that differently from Sram.
IBM gives some figure in that presentation (page 13):
https://www.power.org/events/2010_ISSCC/Wire_Speed_Presentation_5.5_-_Final4.pdf

Spoiler

I'm not sure what IBM mean by application on that slide.
That means that the value for the edram cells (cycle and latency) are for the cell by self and the better result (application) once implemented in the cache where there are other overhead than the characteristic of the cells them selves?

There are obvious big win in density and power consumption.
They give other detail in the white paper (power a2 again)

Deep trench (DT) embedded DRAM (eDRAM) is used instead of SRAM in the
four 2MB L2 caches. The 2MB L2 caches are implemented as 16 1Mb eDRAM
macro instances, each is composed of four 292Kb sub-arrays (264 WL × 1200
BL). The eDRAM cell measures 152×221nm2 (0.0672μm2). This eDRAM implementation
allows substantially larger cache sizes (3×>SRAM) with only ~20% of
the AC and DC power [4]. Use of eDRAM provides a significant noise reduction
benefit in the form of DT decoupling capacitance. DT provides at least a 25×
improvement over thick-oxide dual-gate decoupling devices in capacitance per
unit area. Coupled with a robust power distribution, DT decoupling reduced AC
noise 30mV and reduced power more than 5W.

But still I'm not sure that was the only reason why Aaron stated what he stated.
At the beginning of the first presentation (page 4) they say "smaller memory line".
My understanding is what it is, I'm not sure to get that right but I read it as "tinier cache lines than if they were to use SRAM" in that context.
So I wonder if not because of their density and power characteristic you can use tinier cache lines than if you were to use SRAM. Not that it would be impossible using SRAM cell but pretty costly in silicon, power, etc.
So I've been wondering since Aaron stated that what he really meant, I've a gross understanding about how cache are acting but I wonder if this is what he meant; EDRAM may allow some cache structures that would while desirable would can with undesirable characteristics.

If any body is clever enough to enlighten us he is welcome, I'm struggling to make my -self clear on something I don't really understand, using a naughty blend of guess and "physics" instinct...

 

[TheLump]

The addressable market for Nintendo at $300 is too small and at $350 it is non-existent. Don't forget this is an 8GB of flash machine with a recycled CPU, smaller psu and overall package.. the 6 inch low resolution resistive touchscreen can be bought in bulk for a few dollars. Take a look at rehashed junk like NSMB U and low tech junk like Nintendo Land and P-100, Nintendo is going very low investment in both hardware and software.

Not sure whats low-tech about P-100. We also have no idea if the CPU is "recycled" and in fact the info we do have is that its a custom chip - so technically its new

Also intrigued as to what "8GB of flash machine" actually means. Thats a nice amount of flash storage to have considering you can add any sized external HDD/SDD you want on top of that.

Not arguing with you - just failing to see how those statements relate to your point - or what your point was in raising them?

 

[TheLump]

Sorry for double post - but:

As for Power 7, my knowledge extends as far as Wikipedia

According to that, the cpu is available in 4, 6 or 8 cores. So its certainly possible there is a lower powered 4 core variant out there.

We've only heard that the WiiU CPU is 3 cored though. I certainly think bgassasins theory of a single POWER7 core backed up by two smaller less powerful cores is plausible.

As of October 2011, the lowest number of cores in the available POWER 7 range is the 6 core single chipe 710 express, at 3.7GHz. They have also made a BladeCentre which uses 4 cores at 3.0GHz.

 

[upnorthsox]

They've said
1) It uses "same Power7 chips" ( https://twitter.com/IBMWatson/status/78473693843562498 )
2) Custom chip using same processor tech as Watson ( https://twitter.com/IBMWatson/status/240241146213842944 )

Not that it's Watson, unless you see "same Power7 chips" as that.

You do realize that "they" is actually a computer tweeting this stuff, right?

 

[Kaotik]

You do realize that "they" is actually a computer tweeting this stuff, right?

Err, did you even read the profile? My underlining.

The official twitter feed of IBM Watson. The account is administered by Chris Nay and Michael Holmes, and follows the IBM Social Computing Guidelines.

 

[Shifty Geezer]

According to that, the cpu is available in 4, 6 or 8 cores. So its certainly possible there is a lower powered 4 core variant out there.

There is a 4 core P7 in IBM's BladeCentre PS700 as mentioned in Wiki. A 4 core POWER7 is still a massive, hot chip though.

 

[Kaotik]

Remember it's "custom chip", not off-the-shelf, so why would 1-core Power7 + 2 "enhanced broadways" be out of the question?
Or even 3-core Power7, scaled down a bit from the "off-the-shelf-parts"?