Rumor: Wii U final specs

This thread seems to be veering off target again. The IBM edram is most definitely not 14nm. That article is talking about future technology. Nor will there be 12 MB. It's 3 MB implemented as L2 cache and then the CPU can also access the edram on the gpu (32 MB), so that could be seen as basically an off-chip L3 if devs use it that way. But the cpu is confirmed at 45nm. That includes the L2. The GPU and its edram are all but confirmed at 40nm at this point.
 
Can someone please explain if this is a good thing or not? The eDRAM is 14nm? Is this related to the CPU in some way? Sorry for the multiple questions, I just want to be clear on what we are talking about.

You'll have to ask him:


Gregory Pitner
14nm Fin-FET based eDRAM Design/Verification at IBM
Research Assistant at Rensselaer Polytechnic Institute
eDRAM Development Co-op at IBM
http://www.linkedin.com/pub/gregory-pitner/39/580/4b1
 
What I find strange about this rumor is the choice of using z/Architecture.
This is used for IBM's mainframe computers.

I see something called zIIP mentioned:

In IBM System z9 (and successor) mainframes, the System z Integrated Information Processor (zIIP) is a special purpose processor. It was initially introduced to relieve the general mainframe central processors (CPs) of specific DB2 processing loads, but currently is used to offload other z/OS workloads as described below.

IBM publicly disclosed information about zIIP technology on January 24, 2006. The zIIP hardware (i.e. microcode, as the processors hardware does not currently differ from general purpose CPUs) became generally available in May, 2006. The z/OS and DB2 PTFs to take advantage of the zIIP hardware became generally available in late June, 2006.

Commercial software developers, subject to certain qualification rules, may obtain technical details from IBM on how to take advantage of zIIP under a Non-Disclosure Agreement.

Could that account for the boost in CPU performance?

Maybe what is going on with the WiiU CPU is that its made up of different types of processors:

Depending on the capacity model, a PU (processing units ) can be characterized as a Central Processor (CP), Integrated Facility for Linux (IFL) processor, z Application Assist Processor (zAAP), z10 Integrated Information Processor (zIIP), or Internal Coupling Facility (ICF) processor. (The specialty processors are all identical and IBM locks out certain functions based on what the processor is characterized as.) It is also possible to configure additional System Assist Processors, but most customers find the mandatory minimum SAP allocation sufficient.

In this case it has a zIIP and IFL, from what I can discern from the document.
Thats three processors, gives us the three cores, plus the ARM? Because I dont think that
six means six cores, I think it refers to RISC-like execution units.

For example:

The z196 microprocessor is a chip made by IBM for their zEnterprise 196 mainframe computers, announced on July 22, 2010. The processor was developed over a three year time span by IBM engineers from Poughkeepsie, New York; Austin, Texas; and Böblingen, Germany at a cost of US$1.5 billion. Manufactured at IBM's Fishkill, New York fabrication plant, the processor began shipping on September 10, 2010. IBM stated that it is the world's fastest microprocessor.

fabricated in IBM's 45 nm CMOS silicon on insulator fabrication process

It has four cores, each with a private 64 KB L1 instruction cache, a private 128 KB L1 data cache and a private 1.5 MB L2 cache. In addition, there is a 24 MB shared L3 cache implemented in eDRAM and controlled by two on-chip L3 cache controllers. There's also an additional shared L1 cache used for compression and cryptography operations.[1]
Each core has six RISC-like execution units,


So, take one of those cores, with six RISC execution units, add two helper processors and an ARM?
And you got that strange CPU that people have been spreading rumors about over the last year.




.
 
Since the GameCube era, Nintendo has been rather conservative with it's numbers. Maybe we'll see some better performance with the CPU just because it's not the awful Cell PPE architecture (BTW, Xenon could achieve a theoretical peak performance of 115GFLOPS. It propbably can't reach half of that [hell, I've heard as low as 30-40GFLOPS. John Carmak has even said that the Xbox 360 is only 10X more powerful than the Wii [and Broadway has an "actual" peak rating of 2.9GFLOPS]).

Xenon being 10x faster than Broadway is quite optimistic for typical game code, imo. Xenon reaches it's peak GFLOPs mainly through it's SIMD (VMX) units which are of limited use; and then it's in order. In general, GFLOPs tell you something about raw number crunching capatibilities: Nowadays, this mainly applies to GPU work but rarely for typical CPU work.
The performance differences between Xenos and Hollywood were much bigger in many respects, and I think that's what Carmack mainly had in mind in his statement.
 
OK, let me try this one last time and see how people react to these proposed specs:

CPU: IBM Tri-Core PowerPC "Enhanced Broadway" (2-way SMT, improved SIMD, ect)
@1.2GHz, Approx. 25-35GFLOPS
Approx TDP: 7 Watts

GPU : Custom AMD Radeon HD GPU
300 Shader units
360 MHz clock speed
Approx 400 GFLOPS
Approx TDP: 20 watts

RAM: 1GB GDDR3 for games (clocked at 1100MHz)
1GB DDR3 for background applications (clocked again at 1100MHz)

DSP: commodity priced generic chip @ 120MHz

Sound reasonable? If not, add your thoughts.

Made changes.
 
Has anyone (from press) taken a photo of the Wii and Wii U together? Would be interesting to see how much bigger it is than the Wii. It's so ace that the Wii is tiny :)
 
i thought it was supposedly slightly under 600mhz (or was it 500mhz) according to...what his face. Matt?

Yeah Matt said that 600 was a little too high.......hmm now I'm thinking about it was he talking about Jigglyflops or MegaHertz?
 
you guys are getting your DBZ metaphors mixed up, is it cell or is it buu?

oh shit, or is it super android 13??

The previous was a Cell reference.

Mine was a Buu reference. The only being that absorbed himself to unlock more power.
 
Wait, the eDRAM is only 13 MB? What happened to 32 MB? I thought that was (somehow) confirmed for months?
 
OK, let me try this one last time and see how people react to these proposed specs:

CPU: IBM Tri-Core PowerPC "Enhanced Broadway" (2-way SMT, improved SIMD, ect)
@2.2GHz, Approx. 35-45GFLOPS
Approx TDP: 7 Watts

GPU : Custom AMD Radeon HD GPU
400 Shaders
550MHz clock speed
Approx 500GFLOPS

Approx TDP: 28 watts

RAM: 1GB GDDR5 for games (clocked at 1100MHz)
1GB DDR3 for background applications (clocked again at 1100MHz)

DSP: ARM11-based @ 225MHz

Sound reasonable? If not, add your thoughts.

Wouldn't 400 shaders at 550MHz be more like 1.76 TF?
 
Wait, the eDRAM is only 13 MB? What happened to 32 MB? I thought that was (somehow) confirmed for months?

One is GPU eDram, which is 32MB, the other is talking about the CPU, where 13 is wrong, the CPU as has been talked about by other folks who are in the know, has 3MBs of it as L2 Cache. So the 32MB hasn't changed, it's talking about something else.
 
By shaders they're referring to scalar ALUs.

Yea, I was looking at the specs for Xenos and RSX, where the term shader is used to refer to vector5/vector4 ALUs, but noticed that on the wiki pages for modern GPUs it seems to refer to single APUs, going by the flops. not sure why...

Either way people are going some weird approximations with the flops in their speculations.
 
Yea, I was looking at the specs for Xenos and RSX, where the term shader is used to refer to vector5/vector4 ALUs, but noticed that on the wiki pages for modern GPUs it seems to refer to single APUs, going by the flops. not sure why...

Either way people are going some weird approximations with the flops in their speculations.

So flops calculations is easy with AMD, the formula is # of individual shaders * 2 * (clockspeed/100)

This imaginary 400 shaders x 2 x .550 = 440GFLOPs. a bit low, 480 shaders however, gives us 528GFLOPs maybe a bit low still, but much closer to that ~600GFLOPs rumor (however BGassassin believed it to exceed 600GFLOPs.

My imaginary shot in the dark is 640 shaders @ ~488mhz (DSP's clock speed x 4) giving us 624GFLOPs.

However soon we will likely know quite a bit more.
 
So flops calculations is easy with AMD, the formula is # of individual shaders * 2 * (clockspeed/100)

This imaginary 400 shaders x 2 x .550 = 440GFLOPs. a bit low, 480 shaders however, gives us 528GFLOPs maybe a bit low still, but much closer to that ~600GFLOPs rumor (however BGassassin believed it to exceed 600GFLOPs.

My imaginary shot in the dark is 640 shaders @ ~488mhz (DSP's clock speed x 4) giving us 624GFLOPs.

However soon we will likely know quite a bit more.

Looking at you're maths, shouldn't that be "clockspeed/1000"?
Or more accurately "clockspeed/1048576000"?
 
Do you think there's a chance that Nintendo would allow developers in the future to unlock the WiiU full power? Aka not being forced to stream to the Gamepad? Just having a game run completely with the TV and classic controller/ or only on gamepad?
 
Looking at you're maths, shouldn't that be "clockspeed/1000"?
Or more accurately "clockspeed/1048576000"?

yes sorry, clockspeed/1000 is correct. Wii U should be an interesting piece of tech, I'm looking forward to it.

Do you think there's a chance that Nintendo would allow developers in the future to unlock the WiiU full power? Aka not being forced to stream to the Gamepad? Just having a game run completely with the TV and classic controller/ or only on gamepad?

Considering you can play some games with a classic controller, this is already being done?
 
Do you think there's a chance that Nintendo would allow developers in the future to unlock the WiiU full power? Aka not being forced to stream to the Gamepad? Just having a game run completely with the TV and classic controller/ or only on gamepad?

Err... they can do that day 1 with the system. You never HAVE to use the gamepad, and many games are using it only for maps/extra buttons (IE nothing that pushes the system at all)
 
Do you think there's a chance that Nintendo would allow developers in the future to unlock the WiiU full power? Aka not being forced to stream to the Gamepad? Just having a game run completely with the TV and classic controller/ or only on gamepad?

I dont think Nintendo is forcing developers to stream to the gamepad now.
 
I dont think Nintendo is forcing developers to stream to the gamepad now.

Yes but I remember hearing (perhaps from (Crytek) that you have to design around the gamepad and that the full potential of just creating a game specifically for the TV hasn't been unlocked. Sure you can stream just to the TV with a lot if launch games, but not with any added hardware power or anything.
 
Do you think there's a chance that Nintendo would allow developers in the future to unlock the WiiU full power? Aka not being forced to stream to the Gamepad? Just having a game run completely with the TV and classic controller/ or only on gamepad?

It wouldn't matter much. The impact of the GamePad depends entirely on what developers choose to do with it. Just echoing what's happening on the main screen should have practically zero performance hit (literally zero if the scaling and compression for the GamePad is handled by a dedicated chip, which I believe it is).
 
Do you think there's a chance that Nintendo would allow developers in the future to unlock the WiiU full power? Aka not being forced to stream to the Gamepad? Just having a game run completely with the TV and classic controller/ or only on gamepad?
Developers never were forced to use the gamepad at all.
That's one of the reasons why Nintendo came up with the pro controller... For those developers that didn't want to deal with the gamepad and/or had creative droughts about how to use it properly.
 
OK, let me try this one last time and see how people react to these proposed specs:

CPU: IBM Tri-Core PowerPC "Enhanced Broadway" (single thread per core, 4-5 instructions per cycle, improved SIMD, ect)
@1.6 GHz, Approx. ? GFLOPS
Approx TDP: 7 Watts

GPU : Custom AMD Radeon HD GPU
320 Shaders
532 MHz clock speed
Approx 340 GFLOPS
Approx TDP: 28 watts

RAM: MEM1: 32 MB Renesas eDRAM (UX8GD), 532 Mhz, <2ns latency

MEM2: 2 GB DDR3 (1 GB available initially for games) clocked at 1064 Mhz
128-bit bus: 1064 Mhz (34 GB/s) connection to GPU, 532 Mhz (17 GB/s) to CPU

I/O: Dual Core ARM Cortex-A5 at 532 Mhz

DSP: 133 Mhz, possibly Cortex-M3 based

Sound reasonable? If not, add your thoughts.

Made my own changes. As far as shader count goes, anything between 320 and 400 seems reasonable. The R700 chips seem to scale in groups of 40, though. I like 320 shaders, and thus, 32 TMUs, for Wii BC purposes. A key point in getting that to work is replicating Wii's 1 MB on-chip texture cache. The texture memory on Flipper was broken up into 512 banks, 32 of which could be accessed simultaneously. Each of those 32 banks also had its own address bus. In my design for BC, rather than add more L1 texture cache (expensive SRAM), they would bypass it and use 1 of the 32 MB eDRAM for this purpose. So the eDRAM on Wii U would have to be similarly split up into many banks (at least 512) and have 32 address buses.
 
On the RAM front, I'd say you're right in that it's 2GB DDR3 at about 1GHz on a 128-bit bus. Regarding the split between CPU/GPU bandwidth, my guess is that there's a memory controller on the GPU die which handles RAM access for both chips (and eDRAM access for the CPU), and there's 34GB/s or so of bandwidth shared between the CPU and GPU. I don't see any reason for CPU to be limited in bandwidth compared to the GPU, considering they're on the same MCM.

Edit: Actually, I'm surprised we don't know the RAM specs already. The console's been in the wild for a few days now, it's not difficult to screw the case off, read the codes off the RAM chips and leak them online. Perhaps if someone reading this could give it a quick peek...
 
On the RAM front, I'd say you're right in that it's 2GB DDR3 at about 1GHz on a 128-bit bus. Regarding the split between CPU/GPU bandwidth, my guess is that there's a memory controller on the GPU die which handles RAM access for both chips (and eDRAM access for the CPU), and there's 34GB/s or so of bandwidth shared between the CPU and GPU. I don't see any reason for CPU to be limited in bandwidth compared to the GPU, considering they're on the same MCM.

Can it work that way? I based alot of the above off what they did on Gamecube, as it seems they kind of used that as a matrix and then started replacing units with new compatible ones piece by piece. I'm with you on the memory controller being on the GPU, but it seems most IBM cores use a 2:1 ratio of core:memory clock at the lowest. So maybe a 798 Mhz connection to CPU(keeping the multipliers I used intact), but, theoretically speaking, could it run that fast if the GPU is only 532 Mhz and the memory controller is on the GPU?

Edit: Yes, I've been waiting for that particular information to emerge any day now. I hope I don't have to do it myself.

Edit 2: Actually, another way for the CPU to gain faster access to the RAM without upping the clocks is if they had a fatter bus between the CPU and GPU. We know little about that connection, but considering the proximity of the chips, perhaps it isn't that ludicrous. But then you start thinking about balance and wonder, "Does the CPU Nintendo used even require that much bandwidth?"
 
Can it work that way? I based alot of the above off what they did on Gamecube, as it seems they kind of used that as a matrix and then started replacing units with new compatible ones piece by piece. I'm with you on the memory controller being on the GPU, but it seems most IBM cores use a 2:1 ratio of core:memory clock at the lowest. So maybe a 798 Mhz connection to CPU(keeping the multipliers I used intact), but, theoretically speaking, could it run that fast if the GPU is only 532 Mhz and the memory controller is on the GPU?

Well, for one thing, my assumption has been for a while that the GPU:RAM:CPU clocks would be in a ratio of 1:2:4, for this precise reason (and also because clocks in that ratio would make sense for each component). As far as a memory controller on the GPU is concerned, it is possible to have different frequencies for different components on the same die, the most obvious example being Intel's Sandy Bridge, which has CPU, GPU and memory controller all on one die, all operating at different frequencies. Obviously it's simpler with clean multiples, which I'd expect Nintendo to go with in any case.

Edit: Yes, I've been waiting for that particular information to emerge any day now. I hope I don't have to do it myself.

Edit 2: Actually, another way for the CPU to gain faster access to the RAM without upping the clocks is if they had a fatter bus between the CPU and GPU. We know little about that connection, but considering the proximity of the chips, perhaps it isn't that ludicrous. But then you start thinking about balance and wonder, "Does the CPU Nintendo used even require that much bandwidth?"

We'll get a full teardown from someone like ifixit within a day or two of the 18th, I'm sure.

The bus between CPU and GPU could be arbitrarily wide, really. Even just considering memory, I could see there being a 128-bit bus for the DDR3 and another anywhere up to 512-bit for the eDRAM. Whether the CPU needs 250GB/s+ of bandwidth to the eDRAM I don't know, but it's feasible in an MCM.

Edit: Of course, it would make more sense just to have a single bus for both eDRAM and DDR3 and let the memory controller work it out, but we could still be looking at several hundred GB/s of bandwidth between the CPU and eDRAM.
 
This may be old news, but some of you with pre-orders may end up disappointed this weekend. Nintendo's having manufacturing trouble with the gamepad, specifically. The amount of consoles going out is less than what was planned and they're going to have a very difficult time satisfying the holiday rush for the console.
 
This may be old news, but some of you with pre-orders may end up disappointed this weekend. Nintendo's having manufacturing trouble with the gamepad, specifically. The amount of consoles going out is less than what was planned and they're going to have a very difficult time satisfying the holiday rush for the console.
That runs contrary to what we've been recently hearing actually. Well, in regards to pre-orders, when it comes to non-preordered systems then unless you're lucky with the waitlist or willing to wait in a line for a long time you're probably screwed.
 
This may be old news, but some of you with pre-orders may end up disappointed this weekend. Nintendo's having manufacturing trouble with the gamepad, specifically. The amount of consoles going out is less than what was planned and they're going to have a very difficult time satisfying the holiday rush for the console.

StevieP
Doesn't actually understand technology or have insider info.

Seems reliable.
 
StevieP
Doesn't actually understand technology or have insider info.

Seems reliable.

Quote my tag if you must, I'm just passing on a bit of information that I came across. Take it or leave it. The situation to pre-orders is information in regards to my own country (Canada) where there may not be enough consoles to fill every pre-order made across all chains. Frankly I expected this (the first priority for NCL has always been USA). The gamepad manufacturing thing was new to me, though. I don't know what to make of that little piece of information, or how accurate it is. But I was told specifically "they're having an issue with the gamepad, not the console".

Edit: Plinko - see in regards to my "Canada" comment.
 
This may be old news, but some of you with pre-orders may end up disappointed this weekend. Nintendo's having manufacturing trouble with the gamepad, specifically. The amount of consoles going out is less than what was planned and they're going to have a very difficult time satisfying the holiday rush for the console.


What?! Nintendo having trouble with manufacturing the gamepad therefore I may not get my pre-order. Jeebus, when will this company learn. Pre-order cancelled........'

j/k

Seriously though, have your heard anything more specific about the problem? I wouldnt think the gamepad to be that difficult to manufacture.
 
This may be old news, but some of you with pre-orders may end up disappointed this weekend. Nintendo's having manufacturing trouble with the gamepad, specifically. The amount of consoles going out is less than what was planned and they're going to have a very difficult time satisfying the holiday rush for the console.

You'd think we'd have heard this well before today, and several stores have already stated they know they're getting more than what was preordered.
 
Made my own changes. As far as shader count goes, anything between 320 and 400 seems reasonable. The R700 chips seem to scale in groups of 40, though. I like 320 shaders, and thus, 32 TMUs, for Wii BC purposes. A key point in getting that to work is replicating Wii's 1 MB on-chip texture cache. The texture memory on Flipper was broken up into 512 banks, 32 of which could be accessed simultaneously. Each of those 32 banks also had its own address bus. In my design for BC, rather than add more L1 texture cache (expensive SRAM), they would bypass it and use 1 of the 32 MB eDRAM for this purpose. So the eDRAM on Wii U would have to be similarly split up into many banks (at least 512) and have 32 address buses.

I think the DSP will be 121.5mhz like Wii since thats what the first leak mentioned. Also wow at the lowball GPU GFLOP prediction. That could be the lowest I've ever seen here, lower even than USC-fan :O
 
Quote my tag if you must, I'm just passing on a bit of information that I came across. Take it or leave it. The situation to pre-orders is information in regards to my own country (Canada) where there may not be enough consoles to fill every pre-order made across all chains. Frankly I expected this (the first priority for NCL has always been USA). The gamepad manufacturing thing was new to me, though. I don't know what to make of that little piece of information, or how accurate it is. But I was told specifically "they're having an issue with the gamepad, not the console".

Edit: Plinko - see in regards to my "Canada" comment.

If I don't get my Wii U, I blame you.
 
I think the DSP will be 121.5mhz like Wii since thats what the first leak mentioned. Also wow at the lowball GPU GFLOP prediction. That could be the lowest I've ever seen here, lower even than USC-fan :O

The issue with the 121.5MHz DSP rumour, although I'm not disputing the validity of it at the time, is that it came out before Nintendo would have done full thermal testing on the final hardware, which would mean before clock speeds were finalised. It's entirely possible that 121.5MHz was the target speed for the DSP at the time, but it got changed (either upwards or downwards) as they fine-tuned the various clock speeds in the system.
 
I think the DSP will be 121.5mhz like Wii since thats what the first leak mentioned. Also wow at the lowball GPU GFLOP prediction. That could be the lowest I've ever seen here, lower even than USC-fan :O

Well, the predictions don't serve any sinister agenda, I can assure you that. It's close to the 1.5 rumors we've been hearing and don't forget, the eDRAM will help them get more out of those shaders. I've read reports of CPUs being effectively 4 times faster in real world calculations with eDRAM implemented.

I've also been revisiting the rumor of some type of fixed function capabilities, which the 320 programmable ALUs would, in effect, supplement. What if Nintendo added in some sort of T&L unit similar to Flipper's but suped up? Say it could do some basic calculations for things like lighting and normal maps so that the programmable shaders could be freed up for some of the crazier effects. Can anyone with some technical know-how tell me if that idea is completely insane or not?

The issue with the 121.5MHz DSP rumour, although I'm not disputing the validity of it at the time, is that it came out before Nintendo would have done full thermal testing on the final hardware, which would mean before clock speeds were finalised. It's entirely possible that 121.5MHz was the target speed for the DSP at the time, but it got changed (either upwards or downwards) as they fine-tuned the various clock speeds in the system.

Well said. That's exactly the way I've been thinking. It would make little sense for the clock speeds of the entire system to be bound by multiples of the DSP of all things. I'm sure whatever DSP they chose has some degree of flexibility in clockspeed.
 
The issue with the 121.5MHz DSP rumour, although I'm not disputing the validity of it at the time, is that it came out before Nintendo would have done full thermal testing on the final hardware, which would mean before clock speeds were finalised. It's entirely possible that 121.5MHz was the target speed for the DSP at the time, but it got changed (either upwards or downwards) as they fine-tuned the various clock speeds in the system.

Yeah the clocks could have changed. Just think that until we hear of such a change the info in that leak is the most solid thing we have to base a prediction on.
 
IGN are saying Ninja Gaiden has worse graphics on the Wii U than the other consoles.

Sloppy ports ahoy! Unless you're trying to flip this thing for a profit, or if you really need Nintendoland, I have no idea why anyone would buy this at launch. It might very well turn out to be a promising system, but the level of effort in some of these ports is disconcerting.
 
Sloppy ports ahoy! Unless you're trying to flip this thing for a profit, or if you really need Nintendoland, I have no idea why anyone would buy this at launch. It might very well turn out to be a promising system, but the level of effort in some of these ports is disconcerting.

Team Ninja releasing a game that's not a visual stunner in the launch window is very unlike the Team Ninja of old.

Plus with a framerate just as bad?

No AA? WTF?
 
Top Bottom