Support NeoGAF

[Horse Armour]

I don't think that it even matters, however it's being done it looks suburb in-game. That screenshot has all kinds of aliasing and artifacts that aren't present while playing the game and it still looks great.

I'd like to see how some of these development techniques translate to other art styles before I get too excited though.

You can't, because many polygons are rendered smaller than individual pixels (micropolygons). That's the point I guess.

He's stating that based on the poly number on screen that the Wii U can push a lot more polygons than previously thought or that there is tesselation going on. It's impossible to tell how many polys are in that screen so his suggestions are completely baseless/wishful thinking.

 

[krizzx]

He's stating that based on the poly number on screen that the Wii U can push a lot more polygons than previously thought or that there is tesselation going on. It's impossible to tell how many polys are in that screen so his suggestions are completely baseless/wishful thinking.

That is not what I was stating.

 

[lostinblue]

Either the Wii U GPU can output far more polygons than we realized, or this is using tessellation.

Or it is simply as high polygon as it needs to be for 720p output.

The theoretical throughput is not greatly increased over PS360, but memory is (and geometry takes RAM space also, big reason why tesselation on 360 was so scarcely used too) and on top of it all efficiency probably increased due to newer generation tech and Nintendo's own effort to reduce bottlenecks and latency. It really doesn't need to be pulling some impressive number there, really. (and it's not like we can count them)

It does look good, but Mario games are driven by the art, not the opposite, hence they tend to strive for including pretty good assets in... And that helps a lot. Like saying even Mario Galaxy assets are nothing you can sneeze at; pretty high poly for what they are, but it's the art that makes them shine. It seems to be the same thing here, too.


As for some details in there, stuff like the protuding metal screws, bricks and the like in the bridge could be using steep parallax mapping. It's used in these situations and using them in a complimentary manner seems like Nintendo modus operandi providing the performance tradeoff is better than just throwing polygons at it.

 

[Saint Gregory]

That is not what I was stating.

You kind of did, maybe not full JordanN but close.

I will say that there doesn't seem to be much consistency in game performance analysis on the Wii U so far. Last gen games run like crap and it automatically leads to speculation that the Wii U is a last gen powered system. Stuff like 3DW comes out and looks and performs incredibly and that leads to speculation about Nintendo magic or pixies or something.

Why is it so hard to just accept that there's more under the hood than we can understand by just looking at a die-photo? The Wii U is not in the same neighborhood of what the other systems are providing but we've seen enough from exclusive games to conclude it's not in the last gen ghetto either.

 

[Horse Armour]

That is not what I was stating.

What do you mean? Based on your post, that is exactly what you were stating. This is a good example of why your assertions are difficult to take seriously.

 

[HTupolev]

Either the Wii U GPU can output far more polygons than we realized, or this is using tessellation.

Or you vastly overestimate the amount of polygons needed to make large bubbly shapes look convincing.

There are PS360 game models with absolutely pathetic polycounts that manage to make the rounded, bubbly parts of the models look great. Look at FFXIII; faces are impressively smooth even up close, with models that are only 7K-10K polygons in total.

It's the fine-grained objects with steep curvatures that really start to take a toll. Things like realistic hands with the fingers individually modelled, or accessories. This is a big part of why Halo: Reach has character model polycounts that are 3-4 times as high as FFXIII's, with "blockier" overall model components, and yet still manages to look like it could use more polygons.

 

[fred]

You kind of did, maybe not full JordanN but close.

I will say that there doesn't seem to be much consistency in game performance analysis on the Wii U so far. Last gen games run like crap and it automatically leads to speculation that the Wii U is a last gen powered system. Stuff like 3DW comes out and looks and performs incredibly and that leads to speculation about Nintendo magic or pixies or something.

Why is it so hard to just accept that there's more under the hood than we can understand by just looking at a die-photo? The Wii U is not in the same neighborhood of what the other systems are providing but we've seen enough from exclusive games to conclude it's not in the last gen ghetto either.

Yup, I've been saying for a while that there's a lot more going on under the hood that we're unaware of. Like I've said a few posts up my money would be on fixed functions of some sort. What we've seen of Super Mario 3D World, Mario Kart 8, X, Bayonetta 2 and SSBU shouldn't be possible with a bog standard 160 shader part.

I'd personally say that HDR lighting, depth of field effects and tessellation are likely candidates for fixed functions.

The best example to look at is the Bayonetta 2 demo and the Gomorrah boss fight to demonstrate how capable Latte is. Like I've said above that shouldn't be possible on a bog standard 160 shader GPU at 60fps with v-synch enabled on one screen let alone two. As krizzx has mentioned (although with a different title) it has to be either using tessellation or is pushing a ridiculous amount of polygons.

I don't give a monkey's how efficient the 160 ALUs are, what we've seen of Bayonetta 2 in particular shouldn't be possible if it's a bog standard GPU.

 

[Xanonano]

I must say that the roundness Latte pulls off in this game is superb.

Either the Wii U GPU can output far more polygons than we realized, or this is using tessellation.

He's stating that based on the poly number on screen that the Wii U can push a lot more polygons than previously thought or that there is tesselation going on.

That is not what I was stating.

Sigh...

 

[StevieP]

Or you vastly overestimate the amount of polygons needed to make large bubbly shapes look convincing.

There are PS360 game models with absolutely pathetic polycounts that manage to make the rounded, bubbly parts of the models look great. Look at FFXIII; faces are impressively smooth even up close, with models that are only 7K-10K polygons in total.

It's the fine-grained objects with steep curvatures that really start to take a toll. Things like realistic hands with the fingers individually modelled, or accessories. This is a big part of why Halo: Reach has character model polycounts that are 3-4 times as high as FFXIII's, with "blockier" overall model components, and yet still manages to look like it could use more polygons.

You posted a bullshot

 

[Thunder Monkey]

Sigh...

Yeah... just so happened to be exactly what he was saying.

Even for those of us that have been modeling in some way or another for the past decade it's kind of a bitch to nail down poly counts. I can look and take a guess that Bowser train face is at least 10,000 polygons... but that's a pretty safe bet.

If it wasn't for fans ripping models and devs being forthright we would not know the poly counts of damn near anything. I mean I made a Santa Claus model for a Christmas card animation, he's nicely rounded in all the spots necessary to create an illusion and the model is only 15,000 quads.

Just please guys, stop trying to be the poly counters. It's a crapshoot for someone that's been modeling a decade. You might be able to guess within reason, but in the end its still a guess. Some devs are really efficient modelers, others are not.

Drake from Uncharted 2 is reportedly a 40,000 poly model, Ayane from NG3 with a certain outfit is 120,000 polygons. The lower poly model looks better. We are masters of illusion and the quality of the illusion is at the mercy of the illusionist.

 

[knerlington]

I guess X, some shinen game or Zelda U will be the first time we will really see the potential of the console.
Especially X looks already way beyond anything PS360. if all the stuff from the trailers is really realtime, that is.
but then again: monolith soft did xenoblade which looked unbelievable (for a wii game) and did some amazing stuff.

Way beyond huh? Hate, but still love to break it to you. X doesn't come close to being as technically advanced or good looking as God of War. 3 or Ascension. Neither does anything on 360 for that matter. 360 which still has Halo 4 which I think is pretty much in the same league as Killzone 3. Both of which also looks much better than X. I could go on with titles such as The Last of Us, Motorstorm Apocalypse (whooping 1280x1080 native res along side crazy amounts of physics, rock solid framerate, great effects and models, scale etc.), Beyond Two Souls.

Not trying to bash the Wii U, but to say it's beyond PS360 when it actually has a hard time competing (and failing to keep up) is plain wrong.

 

[fred]

Way beyond huh? Hate, but still love to break it to you. X doesn't come close to being as technically advanced or good looking as God of War. 3 or Ascension. Neither does anything on 360 for that matter. 360 which still has Halo 4 which I think is pretty much in the same league as Killzone 3. Both of which also looks much better than X. I could go on with titles such as The Last of Us, Motorstorm Apocalypse (whooping 1280x1080 native res along side crazy amounts of physics, rock solid framerate, great effects and models, scale etc.), Beyond Two Souls.

Not trying to bash the Wii U, but to say it's beyond PS360 when it actually has a hard time competing (and failing to keep up) is plain wrong.

Super Mario 3D World, X, Bayonetta 2, Mario Kart 8 and SSBU are all beyond anything that the PS3 and 360 are capable of. You have to remember that most of those titles are 720p native, 60fps and have v-synch enabled. You may be able to have those games recreated on the PS3 and 360 but you'll need to make concessions to do so. Turn off the v-synch and put up with screen tearing and probably framerate drops too...or maybe even half the framerate altogether to keep things smooth.

For me personally Bayonetta 2 is the standout title we've seen so far, it looks like it's pushing a ridiculous amount of polys during that Gomorrah boss fight, there's no screen tearing at all and it appears to be locked at 60fps. The Gomorrah boss model is huge, quite complex as far as the scales/skin is concerned, the city skyscrapers are there, Bayonetta herself is over 100K polys if I remember correctly..?

 

[Mr. Wonderful]

I really don't notice the aliasing in 3D World at all, and I'm a snob about that kind of stuff. Seriously, no complaints about the graphics.

 

[lwilliams3]

Way beyond huh? Hate, but still love to break it to you. X doesn't come close to being as technically advanced or good looking as God of War. 3 or Ascension. Neither does anything on 360 for that matter. 360 which still has Halo 4 which I think is pretty much in the same league as Killzone 3. Both of which also looks much better than X. I could go on with titles such as The Last of Us, Motorstorm Apocalypse (whooping 1280x1080 native res along side crazy amounts of physics, rock solid framerate, great effects and models, scale etc.), Beyond Two Souls.

Not trying to bash the Wii U, but to say it's beyond PS360 when it actually has a hard time competing (and failing to keep up) is plain wrong.

Something "looking" better is a bit subjective in the first place. A lot of the games that you mentioned would have different priorities and things that they can "cheat" on over what open-world genre X would be classified as. The developers working on X is obviously focused on scope, draw distance, texture/normal mapping, and very large objects.

 

[Zarx]

I must say that the roundness Latte pulls off in this game is superb.

Either the Wii U GPU can output far more polygons than we realized, or this is using tessellation.

There is more than one way to skin a cat as they say. They could be using something like Bézier Patches like Quake did back in the day.

http://en.wikipedia.org/wiki/Bézier_surface

 

[NBtoaster]

The Bowser head looks pretty high poly, but the rest of the train and the background look very low complexity. The shading doesn't look terribly advanced either.

You posted a bullshot

Not relevant when discussing polycounts.

 

[HTupolev]

You posted a bullshot

You could certainly call the results of the screenshot utility "bullshots", but that's irrelevant. I was simply trying to show what the model looked like, so I looked through my screenshots and found one which demonstrated both the hands and the lack of polygon density on some curved areas despite the "high" polycounts (i.e. other expensive parts of the model, like hands, were emphasized).

There is more than one way to skin a cat as they say. They could be using something like Bézier Patches like Quake did back in the day.

Funny you mention that, as it's an implementation of dynamic tessellation. It's not like they could just tell the a triangle-based GPU to handle bezier parameters!

 

[ab.aeterno]

Yeah... just so happened to be exactly what he was saying.

Even for those of us that have been modeling in some way or another for the past decade it's kind of a bitch to nail down poly counts. I can look and take a guess that Bowser train face is at least 10,000 polygons... but that's a pretty safe bet.

If it wasn't for fans ripping models and devs being forthright we would not know the poly counts of damn near anything. I mean I made a Santa Claus model for a Christmas card animation, he's nicely rounded in all the spots necessary to create an illusion and the model is only 15,000 quads.

Just please guys, stop trying to be the poly counters. It's a crapshoot for someone that's been modeling a decade. You might be able to guess within reason, but in the end its still a guess. Some devs are really efficient modelers, others are not.

Drake from Uncharted 2 is reportedly a 40,000 poly model, Ayane from NG3 with a certain outfit is 120,000 polygons. The lower poly model looks better. We are masters of illusion and the quality of the illusion is at the mercy of the illusionist.

Indeed. As I've said before, I believe in this very thread when this exact same conversation happened previously, it's folly to assume supremely high poly models because a model looks rounded.

There are a number of ways to create the same effect with lower poly models, ways that are far better for performance. Why render a million poly model when you can bake the detail of that model into a much lower poly one using normal mapping or similar and get similar looking results at a lower visual cost? Combined with something like phong shading and you can make very smooth surfaces.

 

[HTupolev]

Why render a million poly model when you can bake the detail of that model into a much lower poly one using normal mapping or similar and get similar looking results at a lower visual cost? Combined with something like phong shading and you can make very smooth surfaces.

Partly because in most current implementations, the silhouette will give you away. I think Halo 4's Cortana has a pretty stark contrast between the complex shading across the surfaces and the actual geometric smoothness (screenshots courtesy of Fyrewulff):

However many polygons you think the SM3DW train looks like it has, barring any surface-extruding funny business, it's got enough at the edges of the curved surfaces that most of them look "smooth enough" (from that distance, from that angle, at 720p).

 

[OryoN]

I think the whole point with krizzx's Mario 3DW pic was to underscore how "perfectly" round/smooth that particular model is at places where it's meant to be. Even in many high-end games today, it's still easy to spot angular parts of meshes that are supposed to be more rounded. They are many rounded surfaces in the pic, and none of them are lacking, or showing any signs of being poly-starved. So his believe was that either they threw a lot of polygons at the model, or implemented some form of tessellation(I'd imagine the former was the more likely and simpler route). The notion itself makes sense, but he worded it in a rather tasteless way.

Anyway, whatever. I came to talk about other things...


As you may have seen in another thread, Project Cars dev revealed how they are using the eSRAM in XB1:

“Our engine uses a light pre-pass style rendering approach and after experimenting with a number of different variations we found it was more efficient to use eSRAM to hold the deferred render targets.

http://gamingbolt.com/project-cars-...l-use-mitigates-ps4s-unified-memory-advantage

I'm guessing that this may be a hint at exactly how they are using the eDRAM(Mem1) in Wii U. That would make sense, as Shin'en also revealed that this is how they are using Mem1 in their second-gen engine, even stating that this method for deferred rendering works well on Wii U(@ 60fps too, for their engine!).

If Project Cars is using Mem1 in this manner, it would suggest that the devs are really putting this high bandwidth, low latency memory pool to effective use, as the architecture calls for. I say that because render targest would not be the only way they are using it. Their logs had also revealed that they are also using it for other things also:

HUD Mirror rendering optimised for WiiU – uses it own aliased targets for phase1/2/3, removing a texture copy. Increased Mirror resolution + all in MEM1

http://www.virtualr.net/project-cars-weekly-build-recap-5

Question: How are they able to use Wii U's eDRAM for so many different things simultaneously?
I was under the impression that render targets alone could occupy most of the eDRAM, but it seems likely that it's enough for that, in addition to other things as well. I'm anxious to see how the Wii U version turns out, since it is by far the least powerful of the Project Cars target platforms. A good port should give us a better idea of the hardware's potential.

 

[NBtoaster]

Question: How are they able to use Wii U's eDRAM for so many different things simultaneously?
I was under the impression that render targets alone could occupy most of the eDRAM, but it seems likely that it's enough for that, in addition to other things as well. I'm anxious to see how the Wii U version turns out, since it is by far the least powerful of the Project Cars target platforms. A good port should give us a better idea of the hardware's potential.

Being only 720p would help.

 

[HTupolev]

Question: How are they able to use Wii U's eDRAM for so many different things simultaneously?

32MB is a pretty healthy render target pool when you're aiming for 720p. Even a large 160bpp deferred G-buffering scheme would fit in less than 18MB at 1280x720, leaving room for other things.

Even the Xbox 360's 10MB eDRAM can fit some (boring) framebuffering schemes at 720p, without needing to tile.

 

[tipoo]

Question: How are they able to use Wii U's eDRAM for so many different things simultaneously?
I was under the impression that render targets alone could occupy most of the eDRAM, but it seems likely that it's enough for that, in addition to other things as well. I'm anxious to see how the Wii U version turns out, since it is by far the least powerful of the Project Cars target platforms. A good port should give us a better idea of the hardware's potential.

A 720p framebuffer was what, 8MB? For double or triple buffering you can do the math, still a bit left for other stuff.

 

[pottuvoi]

A 720p framebuffer was what, 8MB? For double or triple buffering you can do the math, still a bit left for other stuff.

You do not store front buffer or the secondary back buffer in eDRAM.
If you have traditional forward rendered game you can store 720p back buffer and Z-buffer in ~7.3MB.

 

[fred]

Have just had a quick go on Super Mario 3D World after getting it for Christmas and the screenshots and pics really don't do it any justice. It's fucking gorgeous!!!

So here's a question for you guys: It's 720p native and 60fps...and by the looks of it v-synch is also enabled. And v-synch cuts the framerate in half, doesn't it..? So how the fuck have they managed that..? Does that mean it would run at 120fps if v-synch wasn't enabled..? :Oo

 

[lightchris]

And v-synch cuts the framerate in half, doesn't it..? So how the fuck have they managed that..? Does that mean it would run at 120fps if v-synch wasn't enabled..?

No, VSync only cuts the framerate in half (or more precisely: in a divisor of the refresh rate) when it drops below the refresh rate and double buffering is used. Otherwise it doesn't have much impact on performance.

 

[HTupolev]

So here's a question for you guys: It's 720p native and 60fps...and by the looks of it v-synch is also enabled. And v-synch cuts the framerate in half, doesn't it..? So how the fuck have they managed that..? Does that mean it would run at 120fps if v-synch wasn't enabled..? :Oo

No.

V-sync just means that a new frame will only be made into the frontbuffer during the time that the display is transitioning between two frames.

If the game is using a classic double-buffered scheme (has only a single frontbuffer and a single backbuffer), then the GPU will stall after completing a frame when waiting for a screen refresh (the backbuffer is complete so you can't write the new frame in that location, and you don't want to start writing to the frontbuffer since it's currently being sent to the output). This result is that if your performance cannot smoothly maintain the current harmonic fraction of the screen refresh rate, you drop to the next lower one. With a 60Hz display, if you're just barely below what's necessary to maintain 60fps, you drop to 30fps. If you cant maintain 30, you drop to 20. If you can't maintain 20, then 15, then 12, then 10, etc.

Triple-buffering provides an extra buffer for the GPU to render to, preventing it from stalling. In times when performance drops below a given harmonic fraction, this results in better framerates than double-buffering. The cost is that you can get obnoxious variable latency behaviours (ugh, deep pipelines), and you have a larger memory allocation requirement (since there are more buffers).

But no, vsync absolutely does not cut your framerate in half. Even double-buffered vsync only "cuts it in half" if you're targeting the screen refresh rate and you drop below that.

 

[Wishmaster92]

In regards to this thread on how the wii u hardware was downgraded, what is the probability that the wii u wouldn't be able to run the 2011 tech demos anymore?

IIRC the Zelda tech demo was 720p with no aa.

 

[Powerwing]

In regards to this thread on how the wii u hardware was downgraded, what is the probability that the wii u wouldn't be able to run the 2011 tech demos anymore?

IIRC the Zelda tech demo was 720p with no aa.

This rumor got debunked by Shinen :
http://playeressence.com/shinen-debunks-wii-u-hardware-downgrade-devkits-replacements-got-faster/

 

[Ishida]

In regards to this thread on how the wii u hardware was downgraded, what is the probability that the wii u wouldn't be able to run the 2011 tech demos anymore?

IIRC the Zelda tech demo was 720p with no aa.

I think the Wii U is perfectly capable of running that. The Zelda Demo doesn't look better than games we already have like God of War: Ascencion. And the Wii U is clearly more powerful than the HD Twins.

I fully expect the next Zelda game to look like that, although I don't know if they will use that particular art style (I hope so).

 

[ArchangelWest]

I think the Wii U is perfectly capable of running that. The Zelda Demo doesn't look better than games we already have like God of War: Ascencion. And the Wii U is clearly more powerful than the HD Twins.

I fully expect the next Zelda game to look like that, although I don't know if they will use that particular art style (I hope so).

I don't think it's fair to say things like that when God of War: Ascension is largely fixed view, and the gamer cannot control the camera at all, whereas in Zelda games, and in that 2011 tech demo, the user had complete control, zooming in and out, as well as any direction, change of lighting at a whim, etc.

God of War games look so great in part because of almost everything being pre-rendered in a way that doesn't allow for visual changes or angle of view that aren't scripted. They are on rails games where the gamer controls whether or not the train is moving forward. The Wii U is able to output visuals like that which aren't restricted in that way.

I agree with you on the hope in art style though. They probably won't do it, but hope is what it is.

 

[OG_Original Gamer]

I don't think it's fair to say things like that when God of War: Ascension is largely fixed view, and the gamer cannot control the camera at all, whereas in Zelda games, and in that 2011 tech demo, the user had complete control, zooming in and out, as well as any direction, change of lighting at a whim, etc.

God of War games look so great in part because of almost everything being pre-rendered in a way that doesn't allow for visual changes or angle of view that aren't scripted. They are on rails games where the gamer controls whether or not the train is moving forward. The Wii U is able to output visuals like that which aren't restricted in that way.

I agree with you on the hope in art style though. They probably won't do it, but hope is what it is.

Even comparing it to Bayo 2 doesn't fit with how often that game camera changes and runs at 60 compared to God of War Ascension.

I have to admit, was wondering if the rumor is true despite Shinen comments. Suggesting the downgrade occurred before devs got kits. Problem is Shinen was one of the first to get kits.

 

[phosphor112]

Don't know if I should post this here or the espresso thread. Both don't really tell us too much more than we already know.

 

[prag16]

Don't know if I should post this here or the espresso thread. Both don't really tell us too much more than we already know.

Wonder if there's anything to the RV770 mention. Because it seemed most settled on something more like RV740 as more comparable.

 

[Argyle]

Don't know if I should post this here or the espresso thread. Both don't really tell us too much more than we already know.

I hope that now people can stop posting about how they think the old GameCube GPU is behind all the magic of the Wii U...Looks like there are two GPUs there and they look pretty separate to me.

 

[phosphor112]

I hope that now people can stop posting about how they think the old GameCube GPU is behind all the magic of the Wii U...Looks like there are two GPUs there and they look pretty separate to me.

...huh?

 

[NBtoaster]

Wonder if there's anything to the RV770 mention. Because it seemed most settled on something more like RV740 as more comparable.

It obviously doesn't perform anywhere close to PC RV770 hardware (4830-4870). They probably just haven't looked to far into the GPU side of things.

 

[Argyle]

...huh?

The GameCube GPU ("GX") is in a separate functional block from the Radeon based Wii U GPU ("GX2"). In other words it doesn't look like the GameCube GPU is the secret sauce a bunch of posters in this thread were talking about (all that nonsense about GameCube fixed function hardware making the lighting awesome on Wii U games, etc.)

 

[Fourth Storm]

That's just a diagram someone made up. It's got a few inaccuracies, such as the amount of L2 cache in the main CPU core (it should be 2 MB and not 1). Plus, "Starbuck" is not an official code name used by devs for the ARM core.

 

[Apophis2036]

In regards to this thread on how the wii u hardware was downgraded, what is the probability that the wii u wouldn't be able to run the 2011 tech demos anymore?

IIRC the Zelda tech demo was 720p with no aa.

That is Twilight Princess assets running at 720p with a few fancy lighting effects, I think Nintendo blow that tech demo away for the final game. It was also made in a short space of time by a very small team. If they started making the game that Summer and it arrives after Summer 2015 then they will have spent 4 years developing it so it should look incredible going on Nintendo's other WiiU games.

The main thing that will limit how good the new Zelda looks is budget more than anything IMO, if it's on the scale of TP then you are talking $100 million and I'm not sure Nintendo will spend that kind of money on a single project, esp a single project of a series which last game sold just 3.5 million copies :-(.

 

[atbigelow]

That graphic is more talking about the memory structure than GPU components. Nobody has been able to figure out where the Wii U replicates the GCN/Wii GPU (either it's a shim or special hardware).

As for the Radeon 770, we know some of the tech is based off it and that it's a "Radeon" chip. But it's nothing like an off-the-shelf part.

 

[blu]

I hope that now people can stop posting about how they think the old GameCube GPU is behind all the magic of the Wii U...Looks like there are two GPUs there and they look pretty separate to me.

I would definitely not read too much into that diagram. As Storm noticed (good eye), it did not even get the L2 config right.

 

[Apophis2036]

With regards to this rumoured 'downgrade' didn't Ideaman and someone else confirm just before launch that the CPU and GPU were boosted, 1GHz to 1.2GHz and 400MHz to 550MHz ?...

 

[phosphor112]

The GameCube GPU ("GX") is in a separate functional block from the Radeon based Wii U GPU ("GX2"). In other words it doesn't look like the GameCube GPU is the secret sauce a bunch of posters in this thread were talking about (all that nonsense about GameCube fixed function hardware making the lighting awesome on Wii U games, etc.)

I don't think the GX denotes an actual gpu but the sanctioned mem pools dedicated to the Wii functions.

That's just a diagram someone made up. It's got a few inaccuracies, such as the amount of L2 cache in the main CPU core (it should be 2 MB and not 1). Plus, "Starbuck" is not an official code name used by devs for the ARM core.

It's a diagram from fail overflow seems odd that they'd have a mistake like that.

 

[prag16]

I would definitely not read too much into that diagram. As Storm noticed (good eye), it did not even get the L2 config right.

With the amount of time spent on analyzing the die shot by some of our savvy users, I can't imagine something like the equivalent of the full blown Wii GPU being included on die would have been missed.

And phosphor is probably right. That block probably just denotes memory structure; not indicating anything more than that.

 

[Ty4on]

I must say that the roundness Latte pulls off in this game is superb.

Either the Wii U GPU can output far more polygons than we realized, or this is using tessellation.

That is impressive now? The PS360 could both render a lot of polygons. In GT5-6 and Forza 3-4 the highest LoD models had several hundreds of thousands of polygons. Here's what a GT5 model looks like when prerendered on a PC and here's a Forza 4 model rendered by the Xbox 360. Both were used in game (not just photomode/menus) at 720p60 with AA.

3DW looks great, but it is in no way a technical show piece from what I've seen.

 

[fred]

The GameCube GPU ("GX") is in a separate functional block from the Radeon based Wii U GPU ("GX2"). In other words it doesn't look like the GameCube GPU is the secret sauce a bunch of posters in this thread were talking about (all that nonsense about GameCube fixed function hardware making the lighting awesome on Wii U games, etc.)

Nobody has said anything about the TEV Unit from Flipper being the 'secret sauce'. If Latte is using fixed functions (and I believe it is) then the TEV Unit will have to have evolved considerably to prevent the Wii U from having the same problem that the Wii had with ports. Hollywood gave the Wii a nonstandard rendering pipeline making ports impossible. We know this isn't the case with the Wii U because developers have found it easy to port PS3 and 360 games to the console.

And if it isn't using fixed functions and Latte is indeed a 160 ALU GPU then I'd love someone to explain how Bayonetta 2 is possible when the power draw is so low and why the ALUs are also twice the size they should be. Fixed functions of some description explain the size of the ALUs and the low power draw.

It doesn't matter how efficient the shaders are, what we've seen of Super Mario 3D World, Bayonetta 2, Mario Kart 8, X and SSBU shouldn't be possible on a bog standard 160 ALU GPU.

You can call it secret sauce, Nintendo magic, Navi fairy dust, Pikmin hard at work or whatever you want but there's plenty going on under the hood that we're completely unaware of. Again, if you have an alternate theory to fixed functions making all this possible that also explains the huge ALUs and low power draw then we're all ears.

 

[Narroo]

That is impressive now? The PS360 could both render a lot of polygons. In GT5-6 and Forza 3-4 the highest LoD models had several hundreds of thousands of polygons. Here's what a GT5 model looks like when prerendered on a PC and here's a Forza 4 model rendered by the Xbox 360. Both were used in game (not just photomode/menus) at 720p60 with AA.

3DW looks great, but it is in no way a technical show piece from what I've seen.

To be fair, those car games tended to look a lot better than other games in different genres. Not sure about the specifics of that though.

 

[NBtoaster]

It doesn't matter how efficient the shaders are, what we've seen of Super Mario 3D World, Bayonetta 2, Mario Kart 8, X and SSBU shouldn't be possible on a bog standard 160 ALU GPU.

You can call it secret sauce, Nintendo magic, Navi fairy dust, Pikmin hard at work or whatever you want but there's plenty going on under the hood that we're completely unaware of. Again, if you have an alternate theory to fixed functions making all this possible that also explains the huge ALUs and low power draw then we're all ears.

Those games simply aren't that impressive from a tech standpoint and don't indicate anything special going on with the hardware.

 

[NBtoaster]

can you explain what ISNT impressive about X? just curious

Can you explain what is? It has pretty good draw distance but foreground detail isn't overly complex.