Wii U hardware discussion and investigation *rename

Status
Not open for further replies.
AlStrong said:
1tri/clk. Not gonna see uber geometry.
Click to expand...

That would put the Wii U's GPU's at theoretical* 550 million poly/sec. A little bit higher than the 360's theoretical limit, more than twice the PS3's (though it used Cell to help for that), but less than a third of PS4/Durango's 1600 million poly/sec (800Mhz @ 2 tri/clock).

What factors contribute to how close a GPU can reach its polygon-rendering limit while running a game?

*theoretical as in if polygon rendering was all the GPU was doing.
 
GPU's rarely get anywhere near the theoretical triangle limit, there are too many other factors that contribute.
Most GPU's can only hit the peak number with a very small number of interpolated values, trivial vertex shaders, and tiny triangles.
It has some value in how quickly degenerate, clipped and back faced triangles are thrown away, but it's really not a very interesting number.
 
function said:
I think the Wii U does pretty well given its power consumption and mature processes, but overall it's an incredibly weak machine.
Click to expand...
This is the position of the majority on these boards, but "weak" is a relative measure. Compared to what, exactly? Expectations? Previous Nintendo generation? Current competition? Future competition?
It draws 33W from the wall (less than half of the HD-twins at similar processes), is capable of running their multiplatform games OK even at launch, all while maintaining backwards compatibility, and offering some additional capabilities.

Compared to my 4TFLOP Ivy bridge gaming PC, it is a featherweight for sure. But to simply conclude that a featherweight boxer is "weak" compared to the heavyweights doesn't show much appreciation for the sport.
Why is it designed as it is? What are the benefits and drawbacks? What are the implications for multiplatform titles? What, in absolute terms is it capable of? How should it be approached from a programming standpoint to yield optimum results? How....
There are tons of valid points to discuss. It isn't done though.

I'm guessing 320 shaders, with the possibility of less. Any more than 400 and I will eat my hat. I would need to go out and buy one first, but I would, and then I would eat it.
Click to expand...
Even though I rejoice in the images this creates for my inner eye, I doubt you'll need to go shopping. But if you do, I think there's a lot of us who would demand picture proof! :)
 
Entropy said:
This is the position of the majority on these boards, but "weak" is a relative measure. Compared to what, exactly?
Click to expand...
Typically use of the term regards a new console is weak relative to what could have been released as a new console given a more gamer-centric performance target (sod the power draw, load up on high performance parts and dazzle the world!).
 
Entropy said:
This is the position of the majority on these boards, but "weak" is a relative measure. Compared to what, exactly? Expectations? Previous Nintendo generation? Current competition? Future competition?
It draws 33W from the wall (less than half of the HD-twins at similar processes), is capable of running their multiplatform games OK even at launch, all while maintaining backwards compatibility, and offering some additional capabilities.

Compared to my 4TFLOP Ivy bridge gaming PC, it is a featherweight for sure. But to simply conclude that a featherweight boxer is "weak" compared to the heavyweights doesn't show much appreciation for the sport.
Why is it designed as it is? What are the benefits and drawbacks? What are the implications for multiplatform titles? What, in absolute terms is it capable of? How should it be approached from a programming standpoint to yield optimum results? How....
There are tons of valid points to discuss. It isn't done though.
Click to expand...

Actually I've been quite vocal about what I mean by "weak"! It just gets lost amongst the noise though (can't complain, some of the noise is mine).

I think the Wii U is weak compared to the general trend of expensive, under the tv gaming boxes that it is and will be competing against. It's also weak relative to needs of third party software over its lifetime - and this is particularly important and a particularly fair criticism as it is Nintendo themselves who did so much to ensure that this was an area where the machine would be scrutinised. They got EA and others onto their stage at E3 to glow about the box and it's capabilities, they got Reggie to go on National TV in America and lie his tits off about how good multi-platform games were on the Wii U compared to the 360, and they've had other execs praise the system in vague terms (that got Nfans dripping) to make it sound awesome.

The system is also weak compared to fan expectations. A trip over to the always entertaining NeoGaf shows that even now new recipes of special sauce are being brewed up, such as the hidden sauce of super efficient VLIW5 and the magic bandwidth of its DDR3 (there's some clown claiming that the 360's memory BW is split 50:50 between read and write).

Interestingly from a tech perspective the WiiU is also weak compared to super cheap mass produced APUs. The Wii U is just super weak compared to anything you might play $50 games on, with the exception of the PS360, which are so old they need a nurse to bathe them in situ using a sponge. And even they can beat the Wii U up on occasion.

Of course, you could say that the WiiU is powerful compared to other systems that play Mario, and also that water based calculator that I once read about some ancient civilisation using (at least the controller batteries wouldn't run out after 3 hours).

I'll still end up with a Wii U at some point if only for Zelda, but don't expect em to be happy about it. :p

Even though I rejoice in the images this creates for my inner eye, I doubt you'll need to go shopping. But if you do, I think there's a lot of us who would demand picture proof! :)
Click to expand...

Well Azak already has the salt ready ... :eek:
 
I really wonder what Nintendo did to ensure backwards compatibility. I guess it's possible that Latte is no R700 at all, but a modified Flipper with Radeon parts bolted on top - or vice versa. I don't think Nintendo would waste die space on stuff that isn't used in native mode.
 
Hard to see how they could bolt anything modern to something as custom and outdated as the old flipper. It didn't even target any directx version at all, even those that preceeded shaders. Anyway, nintendo has outright stated that wii (meaning gamecube really) hardware is NOT included in wuu.

...Well, other than the basis of the CPU cores it would seem. :LOL:
 
Grall said:
Hard to see how they could bolt anything modern to something as custom and outdated as the old flipper. It didn't even target any directx version at all, even those that preceeded shaders. Anyway, nintendo has outright stated that wii (meaning gamecube really) hardware is NOT included in wuu.

...Well, other than the basis of the CPU cores it would seem. :LOL:
Click to expand...
Yes, they said they didn't include it for legacy purposes, but made new hardware that was fully compatible. Which means all those unique things like the TEV and EMBM units have to be in there somewhere, and can probably be used in Wii U titles as well.
 
If you look very carefully at the bottom right of the die, you can just about make out the special sauce that Nfans are going to spend the next five years desperately trying to see in screenshots of exclusive games.
 
So anyway, yeah, looks like the Wii U might be weaker than anyone (except me, hurray for me) imagined it could be.

Looking at the die shot, it seems that there are 8 blocks of (probably) 20 shaders for a total of 160. Directly above it there are what looks like 3 repeated blocks that I would guess are TMUs, so 12 of them. Then two blocks below the shaders, so probably 8 ROPs. The arrangement of TMUs to shader blocks seems odd, but that's going by the few other die shots I've looked at.

32MB of edram, with what looks like some other smaller blocks of embedded memory above it, maybe for BC purposes?

Anyway, what does everyone else think they see?

Edit: the distance between what I *think* are the ROPs and the edram probably indicates that Wii U ain't got no magic ROPs (like the 360) with free MSAA and transparency, as I think you'd want the ROPs right next to the edram rather than routing a swollen bus right across the chip.
 
Last edited by a moderator:
function said:
Directly above it there are what looks like 3 repeated blocks that I would guess are TMUs, so 12 of them.
Click to expand...
There is a fourth TMU block to the left of the other three, just laid out a bit differently. You can recognize that the memory cells are exactly the same, just in different positions (and the border is a bit unclear). Actually, if you look closely, you will see that all 4 TMU blocks are laid out differently. They really squeezed it in there. And with everything newer than the R600 generation, the number of TMUs is directly coupled to the number of SIMD units. They have to match. Wii U has 4 SIMDs with 4 TMUs each (16 TMUs total).

Edit:
And after thinking about it, I would say that are even full SIMDs in there (i.e. 320 SPs in total), even if they appear to have reached a a quite high density for that. But the clock target was low, which may have enabled a denser layout.
My reasing is as follows: One SIMD engine (without TMUs, which also appear quite small, but that's another story) measures about 3.0 mm². I don't remember seeing a good measure (or any at all) of an AMD GPU made at 40nm (I would expect the Wii U GPU is produced at TSMC, too), so we can just assume some scaling from 55nm (as GF's 32nm SOI is probably even harder to compare). A SIMD engine (just the 80 SPs without redundancy, LDS and TMUs) of RV770 measured about 6.4 mm². It's hard to come up with such a bad scaling to 40nm, that half of it would still be ~3.0 mm². It is more probable in my opinion, that they achieved higher than twice the density given the maturity of 40nm and the low clock target.
But as I said, the TMUs look quite small too and i fail to spot the LDS. Maybe someone else can offer more insight.
 
Last edited by a moderator:
Cheers! I was expecting 4 TMU blocks and three surprised me.

So 1 TMU block to 2 shader blocks, like Luigi (RV710).
 
function said:
Cheers! I was expecting 4 TMU blocks and three surprised me.

So 1 TMU block to 2 shader blocks, like Luigi (RV710).
Click to expand...
:?:
RV710 had two half size SIMDs and therefore 8 TMUs. As I said, starting from the R700 generation, each CU/SIMD has to have exactly 4 TMUs. The Wii U has 4 SIMDs (maybe even full size) which are just laid out in a unusual way to save the last fraction of a mm² on the die. It is basically a halved RV740 (8 SIMDs) with eDRAM on Die and some other changes (64 Bit DDR3 instead of 128Bit GDDR5, different external interfacing to the CPU and some southbridge functions).

Edit:
Without eDRAM, the die would measure around 100mm². As a comparison, the mentioned RV740 measures 137mm² and has twice the compute resources, a significantly higher clock target, twice the memory interface width at a higher standard (128bit GDDR5 PHY measures up to 20mm²) and lacks basically 4 years of experience at 40nm. It appears not impossible to fit half of it together with the eDRAM and some chipset functions in a ~150mm² die, especially if Nintendo opted to gimp it even further (no LDS, some restrictions with the amount of usable texture formats, whatever).
 
Last edited by a moderator:
Status
Not open for further replies.

Similar threads

Rys
Reopening Architecture and Products + rules and guidelines changes
3 4 5
Replies
90
Views
13K
Shifty Geezer
Shifty Geezer
Back
Top