Wii U hardware discussion and investigation *rename

Status
Not open for further replies.
Bolted-on CPU and GPU, maybe on a silicon interposer or otherwise (in the past, pentium pro was connected to its L2 and pentium II was connected to its L2), is a solution I believe to be credible and the eDram would be on the CPU solely because IBM knows how to make 32MB of eDram.

IBM can do off-die L4 and AMD seemed to have one or two pictures of a MCM prototype so, MCM it could be - multi chip module. You still have single memory pool.

I believe it's either this or the SoC.
 
Bolted-on CPU and GPU, maybe on a silicon interposer or otherwise (in the past, pentium pro was connected to its L2 and pentium II was connected to its L2), is a solution I believe to be credible and the eDram would be on the CPU solely because IBM knows how to make 32MB of eDram.

IBM can do off-die L4 and AMD seemed to have one or two pictures of a MCM prototype so, MCM it could be - multi chip module. You still have single memory pool.

I believe it's either this or the SoC.

Would be really interesting if Wii U used either an MCM or a SoC.

My next question would be, what solution would provide a better bandwidth environment?
 
You would only choose MCM over SoC because of practical considerations - R&D costs, mixing processes, building an experience - I'm sure AMD will want off-die memory, with Intel doing the same. Xbox 360's daughter die was a pretty similar affair.
A SoC is better but an MCM at least is on one package and sits below one cooler.
 
What are you smoking? If the edram is a scratch pad then I'll take that over the 360's edram any day of the week. At least that way it has more advantages in terms of rendering than the 360's approach. It will allow devs to use the memory in ways they want, for the different types of buffers they want to have super high bandwidth for. Not just the frame buffer would be of benefit.


Megadrive, if the machine is a SoC and the edram is hooked to the GPU, and it's a scratch pad then think of that as a good thing.
Well the whole thing is that it doesn't seem for now that the gpu has really high bandwidth access to that pool of memory. Otherwise an obvious and easy use of that would be AA which is nowhere to be seen for now at least.
 
Hollywood was an MCM, wasn't it? The package contains three separate dies.

Yeah, totally forgot about Hollywood for a moment there.


edit: I was looking for the Japanese Nintendo Direct where Iwata mentions 2 GB RAM, anyone know? I'm looking on Youtube for it...

edit: nevermind, I found it here
 
Last edited by a moderator:
Well the whole thing is that it doesn't seem for now that the gpu has really high bandwidth access to that pool of memory. Otherwise an obvious and easy use of that would be AA which is nowhere to be seen for now at least.


I beleive some of the recent hands on impression from the WiiU experiences have mentioned that certain games were showing AA (nintnedoLand for one, is apparently now looking like the Japanese screen shots AA included) Take this as you will though, as they're only forum user's impression so may be biased/untrue.
 
Tweets also said it was POWER7 in Wii U. That particular example doesn't say Wii U's CPU is using eDRAM either.

Actually it does, and its pretty clear.
Lets take a look at the IBM news release and see why:
Here is the quote from IBM's own website:

IBM's unique embedded DRAM, for example, is capable of feeding the multi-core processor large chunks of data to make for a smooth entertainment experience.

key word is "the" as in the multi-core processor that IBM defined as
The all-new, Power-based microprocessor for the Nintendo WiiU.

It does not say "(a) multi-core processor"
Which would then imply any multi-core processor not necessarily including the WiiU cpu.

Also, you have to take into consideration the context of the paragraph.
Its starts off with:

The all-new, Power-based microprocessor will pack some of IBM's most advanced technology into an energy-saving silicon package that will power Nintendo's brand new entertainment experience for consumers worldwide.

They clearly state that the CPU will pack advanced technology, not that the CONSOLE will be packed with IBM's most advanced technology. They go further and state what that technology is:

IBM's unique embedded DRAM, for example,

Its their main selling point of the CPU.
It is a clear statement of fact that is in no way vague.

Now if you told me that their statement was talking about the GPU, then I would agree that IBM statement was vague, confusing and open to interpretation.

The facts stated:
The CPU will be a NEW multi-core processor.
The CPU will be placed into a energy-saving silicon package
The CPU will be fed large chunks of data from IBM's unique embedded DRAM.

Why is that vague and not correct?
The only info that the article does not address is
How many cores? What kind of cores?
How much eDRAM?
What kind of energy-saving silicon package?

These are the questions that will lead to the understanding the architecture of the WiiU.
The article does provide a few more clues regarding those questions:

First, Size and location:
IBM plans to produce millions of chips for Nintendo featuring IBM Silicon on Insulator (SOI) technology at 45 nanometers (45 billionths of a meter). The custom-designed chips will be made at IBM's state-of-the-art 300mm semiconductor development and manufacturing facility in East Fishkill, N.Y.


Second, the technology that will make it "power-efficient":
Built on the open, scalable Power Architecture base, IBM custom processors exploit the performance and power advantages of proven silicon-on-insulator (SOI) technology. The inherent advantages of the technology make it a superior choice for performance-driven applications that demand exceptional, power-efficient processing capability – from entertainment consoles to supercomputers.

Notice how IBM linked consoles to supercomputers at the end of the article.
Due to the scaleability of the Power Architecture.

From EDA360:
Tim Fahey wrote to tell me that IBM’s Jeopardy!-playing computer named Watson was powered by SOI. (I wrote about Watson last week, see “Being human is the ultimate app. IBM’s Watson plays Jeopardy!”) Tim heard about IBM’s use of SOI for Watson’s processors on a teleconferencing call with IBM just this morning. A little Googling turned up this new White Paper from IBM that tells all about Watson’s 32-bit SOI processors. They’re Power7 server processors fabricated in IBM’s 45nm SOI process technology. Each processor runs at 3.55GHz and Watson uses a total of 2880 processors in the 90 clustered IBM Power 750 servers that comprise the computer. In this case, SOI is contributing to both the speed and the reduced power consumption of the processors.


But that doesn't mean the WiiU is sporting Power7 cores. No.
And that's why they had to retract the TWEET calling it a POWER7
Here is what IBM had to say:

the POWER line of chips is found only in high-performance servers and workstations from IBM, but the PowerPC chips are found in everything from embedded devices to supercomputers (you may already own one, especially if you're a gamer, run a home network, or drive a late-model car).

So, no matter how similar in design the WiiU CPU is to the POWER7 CHIP, it can't be called a POWER7, unless IBM has changed their own rules.

So when we look at that ENGADGET article:
Nintendo's new console, the Wii U, was finally unveiled to the world today at E3 2011, and we got a glimpse of its graphical prowess at the company's keynote. Details were scarce about the IBM silicon Nintendo's new HD powerhouse was packing, but we did some digging to get a little more info. IBM tells us that within the Wii U there's a 45nm custom chip with "a lot" of embedded DRAM (shown above). It's a silicon on insulator design and packs the same processor technology found in Watson, the supercomputer that bested a couple of meatbags on Jeopardy awhile back.

People jumping to conclusions that they were talking about POWER7 being WiiU's CPU was not correct. But the two chips are similar as they both use eDRAM, their SOI design, and other processor technology, and the fact its produced at 45nm.

Secondly, what did IBM say about the WiiU processor? Its NEW. A key point I missed myself when I assumed the article was talking about the POWER7.
The POWER7 was the latest, but was not a new processor, neither is BROADWAY.
 
People jumping to conclusions that they were talking about POWER7 being WiiU's CPU was not correct. But the two chips are similar as they both use eDRAM, their SOI design, and other processor technology, and the fact its produced at 45nm.

Secondly, what did IBM say about the WiiU processor? Its NEW. A key point I missed myself when I assumed the article was talking about the POWER7.
The POWER7 was the latest, but was not a new processor, neither is BROADWAY.

People didn't jump to POWER7 conclusions from any article.

IBM said straight out (in Twitter) that it's Power7:

David Hansen‏@TheDavidHansen
@IBMWATSON is it true that they're sticking you inside of #WiiU?

IBM Watson‏@IBMWatson
@TheDavidHansen #WiiU uses same #power7 chips.

Later they said "WiiU has 45nm custom chip ... a SOI design & packs same processor tech found in #IBMWatson"
 
Are you responding to me? If so, can you elaborate?
I missed your post, it could have saved you the big post you wrote as an answer to Shifty.

What I meant is that, Nintendo may have asked for a big pool of Edram to be accessible to the CPU it's not exclusive with the fact that (may be BC consideration aside) it doesn't make much sense.

ERP, a resident developer with team management experience, so a person with a opinion well grounded in reality, stated that the GC memory sub system was over designed.
Ie the presence of an L2 cache kind of (mostly) nullified the need for the pool of fast, low latency embedded T-sram.

That is still true, read for example that post of Sebbbi, especially that part:
How much this advantage affects your code base is another debate. As a console programmer, I naturally have a completely different view towards this issue as many PC (or server) software programmers. In games there's huge amount of (mainly vectorized) batch processing happening every frame (viewport culling, matrix multiplies, particle processing, etc). All this batch processing can take 50%-80% of your frame time (depending on game type of course). This kind of code is often highly optimized and uses (often manual) data prefetching heavily, and thus doesn't hit cache (or pipeline) stalls that much. This is also something that is visible in BD benchmarks. It fares well in many PC based software, but not so good in games ported from consoles.
He pretty much states the same thing ;)

So they are no contradiction between IBM pr and Shifty statement.

For the power7 part (in your latter port), the tweeter account acknowledge that it was not a power7 but a power processor.
They had no choice, imo, as they were pushing further than what Shifty called word smithing.
No matter how you look at it, broadway or enhanced broadway would not have the same ISA as even a PPC47x, which doesn't have the same ISA as power7 (2.05 vs 2.06).
So the statement was grounded in nothing technical at all, they could not hold that line any longer as people sooner than latter would have called their talk lies. IBM is not a company that would tolerate that be it on tweeter, they cut it. And it's better to cut failed "word smiting" before it turns in PR incident.
 
Last edited by a moderator:
People didn't jump to POWER7 conclusions from any article.

IBM said straight out (in Twitter) that it's Power7:

David Hansen‏@TheDavidHansen
@IBMWATSON is it true that they're sticking you inside of #WiiU?

IBM Watson‏@IBMWatson
@TheDavidHansen #WiiU uses same #power7 chips.

Later they said "WiiU has 45nm custom chip ... a SOI design & packs same processor tech found in #IBMWatson"


The POWER7 speculation started with that ENGADGET article that was prior to any IBM tweets.
 
PPC750 is already OoO in part
and his pipeline can be clocked up to 1.5/2Ghz in 45nm

Very very very limited OOO capability. The FPU has delayed execution which can cover the latency of a D$ hit, but nothing else.

If the PPC 750 is considered OOO the so is the ARM Cortex A8 (and it isn't).

Going another interation with an ancient in-order micro architecture is very unlikely, IMO. Especially considering IBM has a proper OOO replacement, the 476FP

Cheers
 
The POWER7 speculation started with that ENGADGET article that was prior to any IBM tweets.
All speculation due to the presence of the word "Watson" in the (original PR following WiiU unveiling) PR ;)
 
Last edited by a moderator:
I missed your post, it could have saved you the big post you wrote as an answer to Shifty.

What I meant is that, Nintendo may have asked for a big pool of Edram to be accessible to the CPU it's not exclusive with the fact that (may be BC consideration aside) it doesn't make much sense.

If Nintendo asked IBM to make the eDRAM accessible, it wouldn't have beeen for the CPU,
as thats what IBM has been using their eDRAM for, it would have been for the GPU.

We are assuming the GPU has access to eDRAM based on the design of the current consoles. Which is a logical. But, lets not fly against the face of facts that the WiiU CPU, like the POWER7, has "alot" of eDRAM.

My only question is, is it shared with the GPU, or is it seperate. And if its shared,
what's stopping Nintendo from having IBM & AMD design their chips to have access to 32MB, 256MB, or even one GB of eDRAM?

Not IBM, they claim they can offer more than 128MB of eDRAM. AMD has for, 2011- 2012, the goal of having their CPU and GPU to share fully coherent memory.

Here is what AMD said about GPGPUs not to long ago:

Existing APIs for GPGPU are not the easiest to use and have not had widespread adoption by mainstream programmers.

Wouldn't this be a problem for IWATA and team Nintendo who has declared that WiiU is using a GPGPU? Can you imagine the fail for going through the trouble of customizing a chip to be used as GPGPU and not having developers make use of it. Even now we are hearing developors complain about weak CPU, or unknown architecture. Nintendo would have had to be prepare for this, luckily they are working with AMD:

In HSA we have taken a look at all the issues in programming GPUs that have hindered mainstream adoption of heterogeneous compute and changed the hardware architecture to address those. In fact the goal of HSA is to make the GPU in the APU a first class programmable processor as easy to program as today's CPUs. In particular, HSA incorporates critical hardware features which accomplish the following:

2. HSA Memory Management Unit: This allows all system memory is accessible by both CPU or GPU, depending on need. In today's world, only a subset of system memory can be used by the GPU.

3. Unified Address Space for CPU and GPU: The unified address space provides ease of programming for developers to create applications. By not requiring separate memory pointers for CPU and GPU, libraries can simplify their interfaces

4. GPU uses pageable system memory via CPU pointers: This is the first time the GPU can take advantage of the CPU virtual address space. With pageable system memory, the GPU can reference the data directly in the CPU domain. In all prior generations, data had to be copied between the two spaces or page-locked prior to use

5. Fully coherent memory between CPU & GPU: This allows for data to be cached in the CPU or the GPU, and referenced by either. In all previous generations GPU caches had to be flushed at command buffer boundaries prior to CPU access. And unlike discrete GPUs, the CPU and GPU share a high speed coherent bus
http://www.anandtech.com/show/5847/...geneous-and-gpu-compute-with-amds-manju-hegde
 
Actually it does, and its pretty clear.
Lets take a look at the IBM news release and see why:
Hold your horses. I replied explicitly:
Me said:
That particular example doesn't say Wii U's CPU is using eDRAM either.
to the quote:
IBM said:
Yes, Power-based Watson and the Wii U both use IBM’s small, stable and power-efficient on-chip embedded DRAM (eDRAM) on SOI, which is capable of feeding multi-core processors large chunks of data.

Referencing the PR from IBM's website isn't changing that above quote which is the one I said doesn't say Wii U's CPU is using eDRAM. ;)

Now that said, I'm getting lost by what your argument is and what your referencing this PR statement to prove. The content in it is questionable. Firstly, there's the key issue regards PR statements that they are written by copy-writers and PR people, and not technicians, so the accuracy of the content beyond corporate stats can't be relied upon without good scrutiny.

eg. IBM's unique embedded DRAM, for example, is capable of feeding the multi-core processor large chunks of data to make for a smooth entertainment experience.

That sentence is pure fluff. Large chunks of data through the CPU isn't what makes for "a smooth entertainment experience." In fact a lot of times it's small chunks streamed that are needed, requiring system bandwidth rather than large CPU caches.

We also have fluff like this:

"We're very proud to have delivered to Nintendo consistent technology advancements for three generations of entertainment consoles,"

Technological advancements for three generations? Last year's advancement was virtually nothing other than manufacturing.

Second, the technology that will make it "power-efficient":

Notice how IBM linked consoles to supercomputers at the end of the article.
Due to the scaleability of the Power Architecture.
That's more PR fluff. That paragraph is talking about IBM's custom CPU services in all sectors, which spans everything.
IBM said:
Built on the open, scalable Power Architecture base, IBM custom processors exploit the performance
...That is, IBM custom processors all use the Power base, and these processors are good from the smallest to the largest computer. That's an advert that whatever your processing needs, IBM are happy to design a chip. The actually relevance to Wii U is unspecific.

People jumping to conclusions that they were talking about POWER7 being WiiU's CPU was not correct. But the two chips are similar as they both use eDRAM, their SOI design, and other processor technology, and the fact its produced at 45nm.
PS3 and Xenon use SOI and 45nm.

The info we have, as I see it, is

  • IBM are building the CPU
  • it's a custom design following their standard Power Architecture developments with whatever customisations Nintendo want
  • it has eDRAM
We don't know the application of the eDRAM. There are three options:

  1. It's for the CPU
  2. It's for the GPU
  3. It's for both
1. This makes very little sense, and as the only remarks supporting the idea of the eDRAM being for the CPU seem to come from a PR article that's clearly lacking in-depth understanding of games, there's not much reason to ask for this in a CPU.



2. There's no mention of the GPU in the PR so far, suggesting the eDRAM isn't for the GPU. But then I'm not placing a lot of faith in these PR comments. ;) Would IBM want to post 'we are incorporating AMD's graphics hardware into the chip'? I don't honestly know.


3. Same sort of issue as 2. Where does the GPU fit in?

So we are left with the conundrum of which vital piece of information is being left out. Is the GPU being included on die but not being talked about? Or is the eDRAM for the CPU and the GPU's on a separate chip? The arguments in contention are PR vs. technical knowhow. I'm siding with technical knowhow on this one. That, or the eDRAM in question is actually just a couple megs of cheap L2 cache and Wii U hasn't even got eDRAM for its GPU, given that the 32 MB number has never been substantiated and appears to just be a self-reinforcing rumour. But the EG article, which is based on devs giving info rather than PR copy-writers, says there's eDRAM on the GPU, so I have to believe that at this point as it's the architecture that makes sense in every way.
 
Wouldn't this be a problem for IWATA and team Nintendo who has declared that WiiU is using a GPGPU? Can you imagine the fail for going through the trouble of customizing a chip to be used as GPGPU and not having developers make use of it. Even now we are hearing developors complain about weak CPU, or unknown architecture. Nintendo would have had to be prepare for this, luckily they are working with AMD:

I can't imagine the fail, because the Wii U GPU is not customized for GPGPU.
In fact, GPGPU originally appeared just because you could do it, in the beginning you would draw two triangles that cover a virtual rectangular surface and just run pixel shaders.

Then your links and quotes bring very tasty features, and make using GPGPU elegant, but they're expected for the Xbox 720 if the tech it uses is current when it's out :p
 
How technically feasible/expensive would it be for IBM to develop a custom PowerPC 470 variant with additions to make it fully (or almost fully) backwards compatibility with Broadway?
 
Status
Not open for further replies.
Back
Top