• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

WiiU "Latte" GPU Die Photo - GPU Feature Set And Power Analysis

Status
Not open for further replies.

krizzx

Junior Member
Citation needed. AFAIK real FP16/FP32 HDR lighting was unattainable on consoles until the 360/PS3 era which possessed GPUs that had the DX 9.0c feature sets which made them capable of doing so. Even then, it's still computationally expensive.

To my knowledge, only 2 games used real HDR on the Wii. Dead Space Extraction(used it in a boss fight) and Cursed Moutain

http://image.gamespotcdn.net/gamespot/images/2009/151/958247_20090601_screen005.jpg

http://www.youtube.com/watch?v=oVnW254I2nY

I can't find a video or photo of cursed mountain using, but I remember I could see it on the water when the buildings cast a shadow. I could tell because of the way the light looked(it wasn't fluorescent looking like bloom) and it was curved/rounded kind of. It also bent around the corner a little. Its listed in the HDR wikipedia. Type in Athena. That was the name of the game engine

http://en.wikipedia.org/wiki/High_dynamic_range_rendering
 
Read the OP. It's been updated since you last read it!

320:16:8 -> 352 GFlops. But half the die is magical mystery hardware, who knows what secrets it may hold? That's what the "fixed functions" talk is about. The alternative would be some asymmetrical core configuration, which is possible (but not necessarily likely) and would mean more GFlops.

Where does that number come from? That's what I'm asking.

Also, it's been confirmed that none of this circuitry is redundant for fab yield?
 

NBtoaster

Member
FP10 HDR lighting was actually used extensively by Valve in most of their Source based engine games, starting from their Lost Coast benchmark up until Half Life 2: Episode 2 (which featured an updated Source engine). It's not just the 360 that features this gimped HDR solution.

I didn't say it's the only console that supports it, just that a lot of solutions are available.
 

AlStrong

Member
FP10 HDR lighting was actually used extensively by Valve in most of their Source based engine games, starting from their Lost Coast benchmark up until Half Life 2: Episode 2 (which featured an updated Source engine). It's not just the 360 that features this gimped HDR solution.

Valve used an integer format.
http://www.valvesoftware.com/publications/2006/SIGGRAPH06_Course_HDRInValvesSourceEngine_Slides.pdf
http://arstechnica.com/features/2005/09/lostcoast/3/
 

LeleSocho

Banned
It's the number of cores (ALUs) x the clockrate x flops per cycle

So, 320 x .550 x 2

I have yet to find someone that gives me a coherent explanation on why there are 320 ALUs aside "Digital foundry said it"... from what we know it can easily be 160 ALUs (20 every SIMD core).

twvRnpx.png

yeah i kinda of deserve it :p
 
I have yet to find someone that gives me a coherent explanation on why there are 320 ALUs aside "Digital foundry said it"... from what we know it can easily be 160 ALUs (20 every SIMD core).

From my understanding it's because the SIMD cores are larger than RV770 even on a smaller production node, so we know there's more, but not how much more, it's still speculation.


What's the explanation for that?

That's what it was on RV770
 

axisofweevils

Holy crap! Today's real megaton is that more than two people can have the same first name.
I don't understand much of what's going on, but I'm finding this thread really interesting. Thanks to Chipworks and to everyone involved!

Also, Eurogamer / Digital Foundry has gone down a lot in my estimation. Not crediting GAF and eagerly writing a "Wii U is last gen Trolololo" article without knowing what a lot of it does.
 

dwu8991

Banned
I don't understand much of what's going on, but I'm finding this thread really interesting.

Also, Eurogamer / Digital Foundry has gone down a lot in my estimation. Not crediting GAF and eagerly writing a "Wii U is last gen Trolololo" article without knowing what a lot of it does.

Yep, what they are basically saying is there won't be too many 3D 1080P / 60 fps games

Except Monster Hunter Ultimate 3 of course !
 
I don't understand much of what's going on, but I'm finding this thread really interesting. Thanks to Chipworks and to everyone involved!

Also, Eurogamer / Digital Foundry has gone down a lot in my estimation. Not crediting GAF and eagerly writing a "Wii U is last gen Trolololo" article without knowing what a lot of it does.

This is the biggest blow to them for me. Especially then using non final specs for Durango or Orbis as a comparison (even if it would make it look worse) is just signs of click baiting and rushing out an article for it.
 

NBtoaster

Member
I don't understand much of what's going on, but I'm finding this thread really interesting. Thanks to Chipworks and to everyone involved!

Also, Eurogamer / Digital Foundry has gone down a lot in my estimation. Not crediting GAF and eagerly writing a "Wii U is last gen Trolololo" article without knowing what a lot of it does.

GAF was credited for getting the photo released. They have their own speculation on what the rest of the chip is.
 

JordanN

Banned
Microsoft releasing Nintendo's own numbers had no effect at all. People already knew that the Xbox was the most powerful console and that the PS2 was the weakest. It was generally accepted that the GCN in the middle power-wise, which had no effect on sales because the PS2 still outsold the competition by huge amounts despite being a weaker machine than the Xbox.

Not only that but Nintendo actually kept the Gamecube specs up years after the supposed "Microsoft humiliation". So yeah, if they really gave a damn, they would of tried to erase all history of it.

http://web.archive.org/web/20030606221758/http://nintendo.com/systems/gcn/specifications.jsp
^
That's from 2003
 

ahm998

Member
From my understanding nothing make us know how much Gflops inside Wii U.

Why everyone think Wii U weak ? please explain why Gflops less than 350?

People said Wii U has ati 4870 and this is Graphics card info.

http://www.amd.com/us/products/desk...d-4870/Pages/ati-radeon-hd-4870-overview.aspx

Also they mention maybe ati 5450.

http://www.amd.com/US/PRODUCTS/DESK.../HD-5450-OVERVIEW/Pages/hd-5450-overview.aspx

Also they mention maybe ati e6760.

http://www.amd.com/us/products/embedded/graphics-processors/Pages/radeon-e6760-discrete-gpu.aspx

All these graphics card more than 500 Gflops.

Is Wii U graphics card mobile card?
 

wsippel

Banned
From Gipsel/ B3D:

This is odd. Only half the number of register files you'd expect from a 40 ALU cluster, but the register files are quite a bit larger. Which would make sense if the chip was 55nm, but Chipworks says it's 40nm, and the eDRAM appears to be too small for 55nm as well.
 

AzaK

Member
From Gipsel/ B3D:


This is odd. Only half the number of register files you'd expect from a 40 ALU cluster, but the register files are quite a bit larger. Which would make sense if the chip was 55nm, but Chipworks says it's 40nm, and the eDRAM appears to be too small for 55nm as well.

Confused,the bottom images have the same number of registers. What's what?
 

wsippel

Banned
Confused,the bottom images have the same number of registers. What's what?
Of course, because it's comparing one Brazos cluster to two Latte clusters. Now the number of registers meets expectations, but the size doesn't make sense in this case.
 

beje

Banned
Of course, because it's comparing one Brazos cluster to two Latte clusters. Now the number of registers meets expectations, but the size doesn't make sense in this case.

Isn't this normal if we're talking about different architectures or dies? Honest question.
 
here's the thing though, we stopped using fixed functions a long time ago, the chances that they handle any particularly advanced effects which would translate into measurable performance boosts are slim to none.
Fixed functions are nothing more than specialized parts of the chip doing some operations. For example, GC/Wii had fixed function self shadowing (and to emulate them, maybe the WiiU has that fixed function too). ¿Is self-shadowing something not used today?
The fact that fixed functions are not used on PC nowadays doesn't mean that Nintendo couldn't include some of them to perform some common operations.

Of course I'm not saying they are there for sure, but it's a possibility...
 

efyu_lemonardo

May I have a cookie?
So I just finished getting up to speed on all the new information, at least the parts of it I can understand. Major thanks to chipworks! Honestly not sure why you guys decided to be so generous :p
I guess you believed it would be worth it to open up the discussion and get more people analyzing this custom part. Still, really appreciate this level of openness with the community!
The more disturbing part, from what I understand, is we still don't know what constitutes more than half of the silicon(!)
And that's not even getting into other apparent discrepancies like the amount/density of register files, as pointed out above..

Question to those more knowledgeable than I:
Is the logic for the custom Broadcom hardware on this die as well?
If not, is it possible there is any specialized circuitry here that is being used for image compression or maybe even sharing of information between the rendering pipeline and video compression hardware to help reduce bandwidth or latency of the signal?
We know this part of the design was given special consideration by the teams involved.

Additionally, do we know anything new regarding the ARM cores? Such as their size or location on the die?
Anything on the audio hardware for that matter?

I know little in the area of of IC design, but it just seems surprising to me that the majority of the silicon is taken up by things we can't account for yet...
 

efyu_lemonardo

May I have a cookie?
Sorry for the double post, having a hard time editing from my phone..

Assuming the GPU is capable of 350+ GFLOPS as explained on the first page, and comparing that to Xbox360 performance and something like Monolith's demo, is it reasonable to say that in terms of graphics, the Wii U GPU is being put to fairly good use in the 'X' demo?
If that's the case, perhaps all that extra silicon on the die is being put to a different use, such as more general purpose computing?
Maybe certain logic we'd normally expect to find in the CPU has been moved over here? (I mean computational logic, not things like cache which is being discussed separately). Would perhaps explain Iwata's GPGPU remarks as well as possibly go along with the emphasis on avoiding memory bottlenecks that could be achieved by having certain computational elements closer to the GPU..
Is there some way something like that (ie. programmable rather than fixed function custom logic) could account for a significant area of the die?
 

beje

Banned
Isn't Nano Assault Neo also supposedly 60 fps 1080p?

They decided to leave it at 720p so the 60 fps could be completely locked and freed them resources for some extra effects. One game I didn't know to be 1080p on the Wii U was Skylanders, which is only 720p on PS360 (or at least that's what some gaffer posted yesterday in the new Skylanders thread) but I don't think it achieves 60 fps.
 

Donnie

Member
More modern, yes. More capable, no.

The Wii's shader could and did do HDR. The Xbox1 shader couldn't to my knowledge. The Wii's shader was just more difficult to use leading to it rarely ever being used.

I'm still wondering why on one has followed through on the comment that the GPU isn't based on an AMD GPU design that was made by the Jim Morrison from Chipworks.

XBox's shader architecture was more capable, its just that the difference in capabilities wasn't as big as some people made out, as long as enough effort was put in. Of course Hollywood (Wii GPU) had quite a lot of advantages over NV2a (XBox GPU), just that pixel/vertex shader flexibility/usability wasn't one of them. Still the most important thing that held back Wii's GPU was that Wii's method was so different to a modern shader architecture that it couldn't run the same style of code, meaning engines needed to be totally re-written. Just look at the Far Cry port from XBox to Wii, it was a mess despite Wii being more powerful, it all came down to the different pixel pipelines.
 

Rich!

Member
So, slightly unrelated - but has there been any news on Wii U exploits/hacking since the failoverflow Reggie photo?

Would this die scan get marcan and co any closer?

I thought I had a reliable source for this kind of news, but obviously not. Heh.
 

Thraktor

Member
From Gipsel/ B3D:


This is odd. Only half the number of register files you'd expect from a 40 ALU cluster, but the register files are quite a bit larger. Which would make sense if the chip was 55nm, but Chipworks says it's 40nm, and the eDRAM appears to be too small for 55nm as well.

I wouldn't read too much into it. Given they were rearranging the ALU clusters in the first place, and they've obviously put a lot of thought into memory hierarchies, they probably just decided to go with fewer, larger, SRAM blocks for the register files for some reason. The size comparison does seem a solid indication that we're looking at 40 ALUs per cluster, though.

I'm at work at the moment, but I have some more thoughts on the last couple of pages for later.
 

tipoo

Banned
I do wonder if that Marcan guy is just trying to stir up people. Knowing what he does about the hardware, comparing it to the Wii situation sounds incredibly ignorant and just generally dumb. The Wii was ORDERS OF MAGNITUDE weaker than the 360/PS3. We're talking numbers around 20x weaker while the Wii U seems to only be in the single digits. Anywhere from 3x-5x weaker, not to mention it actually has modern hardware. I just don't understand how anyone can compare the situations, assuming they know anything about the hardware of all the consoles

He did clarify:
Héctor Martín ‏@marcan42
Sigh. Guys, I said it was "kinda" like the Wii situation. Not "the ratio between the Wii and the 360 is identical to the Wii U vs. Durango".
 
Makes sense, thanks.

It also makes sense given that SOME Wii U multiplats are better. Trine 2 shows major improvements over the other console versions in it's graphics, sporting a lot of effects that were missing in PS3 and 360 versions. Game logic is identical. We know it's not the CPU that's given them more overhead to work with, because we know that's weaker, and the game shows no difference in physics/game logic etc.

if the GPU isn't somewhat more powerful then Trine 2 doesn't make any sense. Frozenbyte themselves have said it was the GPU that let them get more out of their engine, and if it was all coding magic rather than having more to work with, I'd imagine they'd be crowing about their achievements.

352 GFLOPs doesn't just 'sound right' or 'look right' it also lines up with things like Trine 2.
 
I'm looking at this chip and I'm also looking at the prowess of the games as a reflection of this tech, and I think to myself, "that seams about right."

Now, for those who say "just you wait for the awesomeness to come" - I wonder where they get that from (from a tech POV). I agree the more Nintendo gets their hands on this, the better the games will be - but I'm not seeing a huge jumps anytime soon. A couple franchises will gradually become better looking with time, but so will PS420 games. Every step of the way, the games will always look significantly better on the crop of new systems coming in the fall.

For those who imply this chip is garbage, I at first want to agree - but then remember Nintendo is clearly not gunning for technological superiority, rather sales of HD Nintendo IP.
 

wsippel

Banned

I'm pretty much convinced now that my gut feeling was correct and D is Starbucks. Starlet had 128kB SRAM as TCM, so Starbucks needs to have those as well. And that seems to be exactly the amount located in this block. It might seem a bit too large for an ARM926, but there should be other stuff like the crypto engine in there.
 
In now way am I saying the other 30% of the chip can bring it close to the PS4 and 720's performance, the Wii U's not going anywhere near their levels. All I'm saying is that it's more than the raw number being thrown around. As of this point, we still can't determine is performance.

All we can firmly say is that at bare-minimum it's 352 FLOPS.

When you say its 352 FLOPS do you mean Gigaflops? Excuse me, im not that technically savvy.
 

chaosblade

Unconfirmed Member
Now, for those who say "just you wait for the awesomeness to come" - I wonder where they get that from (from a tech POV).

I don't think anybody is saying this, and everybody is saying this:

A couple franchises will gradually become better looking with time, but so will PS420 games. Every step of the way, the games will always look significantly better on the crop of new systems coming in the fall.

This isn't about finding some sort of secret sauce that will magically make the WiiU as capable as Sony and MS next systems, it's figuring out what makes the WiiU tick since Nintendo isn't giving any information.
 

ozfunghi

Member
Between the 1GB gDDR3, 32MB eDRAM, 4MB eDRAM and 1MB SRAM... what about the beyond3D speculation that WiiU is memory/bandwidth starved and therefor has trouble with alpha textures, as seen in a couple of launch games and the "X" trailer? How likely is this still, or rather the result of developers still (needing to) getting the hang of it.

Anyone care to take a stab at this? B3D seemed rather convinced about it, but they seem to be convinced about a lot of stuff (they don't know).
 
I don't care about specs all that much. What matters are 1. The end user experience with the hardware and 2. The games. Both are bigger issues with the WiiU at the moment than some future tech the WiiU will have to compete with.
 
I'm pretty sure that every game on WiiU stugles with Alpha textures so it's a pretty solid hypothesis imo.

I mean look at this:Epic Mickey 2

It's clear there is a majour bottleneck somewhere, that isn't present on 360.
 

ozfunghi

Member
I'm pretty sure that every game on WiiU stugles with Alpha textures so it's a pretty solid hypothesis imo.

I mean look at this:Epic Mickey 2

It's clear there is a majour bottleneck somewhere, that isn't present on 360.

Sorry. This isn't the "proof" i'm looking for. The game looks hardly better than the Wii version, which runs perfectly. WiiU < Wii confirmed?
 
I'm pretty sure that every game on WiiU stugles with Alpha textures so it's a pretty solid hypothesis imo.

I mean look at this:Epic Mickey 2

It's clear there is a majour bottleneck somewhere, that isn't present on 360.

Yes clearly a rushed port of a unimpressive multiform game by an awful studio built for the architectures of the HD twins makes a solid hypothesis.

Just like how Bayonetta proves that the PS3 is garbage and can't maintain the frame rate the 360 can.

Not to mention the fact that that most of the horrid textures in X look like they are still alpha (development) textures with 64 x 64 resolution. Meaning?.... Place holders.

We need to actually see games built from the ground up on Wii-U to know if memory bandwidth is actually a problem. Not some awful port by a third party
 
But going by that comparison there's only 4 clusters, not 8. Where does the 320ALUs figure come from then?

No, going by that comparison there are 8 clusters. The important information to discern from the comparison was that the size difference means that most likely there are 40ALUs per cluster meaning 320ALUs.
 
Status
Not open for further replies.
Top Bottom