Wii U hardware discussion and investigation *rename

[Exophase]

Gipsel, this is Andy Glew's presentation that originally made me very confused (since he insisted that GPUs actually used this, and because he used the same "SIMT" term):

http://parlab.eecs.berkeley.edu/sites/all/parlab/files/20090827-glew-vector.pdf

Here he says that all GPUs are using this type of SIMT and stresses that it's not just a robust SIMD with masking. The description on pages 25/26 is what I thought nVidia GPUs were using (having read about the ISAs on AMD GPUs and descriptions from posters here I didn't think it was more than SIMD w/predication, didn't know about the ability to create wavefronts dynamically).

I think he was just wrong about what GPUs at the time were doing and I took it on good faith that someone in his position wouldn't be...

 

[Gipsel]

AMD actually cleared up the terminology a bit with GCN (and the clear cut architecture actually makes it simple to do) and they state explicitly that what they are doing is SIMD with lane masking and have drawn parallels to vector processors. As I mentioned, that may change at least partially in the future (going more into the MIMD direction), but that is not the current situation.

Btw., as said that dynamic wavefront "creation" (forked wavefronts are not really newly created, they put the current PC on a stack [using scalar registers] and reuse the existing one) appears to be a bit of a kludgy addon. While it enables to circumvent a class of problems with arbitrary control flow, you still lose the parallelism as the execution is potentially completely serialized (i.e. it has the potential to become very slow).

 

[Cyan]

1 Wavefront= 64 Threads.

1 Wavefront is completed in 4 cycles (16 threads per cycle).

In the old AMD architecture 1 thread=1 VLIW5.

Then:

192 Threads= 3 Wavefronts.
1 Cycle=48 Threads
48 Threads*5=240 Stream Processors?

Well, in the GPU it seems that we have 160 Stream Processors:

32 Threads per cycle= 128 Threads.

This means that the other 64 other threads are memory operations?

It looks like the WiiU's EDRAM + its GPU-centric design are giving some promising results.

A game called Fast Racing Neo is going to produce and feature hyper real graphics and uses 4k and 8k textures.

http://playeressence.com/fast-racing-neo-uses-4-8k-textures-produce-hyper-real-graphics-says-shinen/

The developers want to impress from a technical standpoint.

I wonder though what's the point in using 4K and 8K textures to impress if no game runs at 4k or 8k these days.

 

[(((interference)))]

So... Nintendo was intentionally incompetent?

I thought that was already established beyond reasonable doubt lol.

 

[Grall]

I wonder though what's the point in using 4K and 8K textures to impress if no game runs at 4k or 8k these days.

To cover all sides of a multi-faceted object in high res texturework you need large textures...

 

[Shifty Geezer]

It looks like the WiiU's EDRAM + its GPU-centric design are giving some promising results.

A game called Fast Racing Neo is going to produce and feature hyper real graphics and uses 4k and 8k textures.

That's bollocks talk. There's no such thing, and Fast Racing Neo won't look any more realistic than other racers out there, which don't even look real, let alone 'hyper real'. Shinen are great are getting the Wii U fans excited at their box and eager for games, but let's not use PR talk as a reference in a technical hardware thread.

As for 4k and 8k textures, that's just a structuring format. One 4k texture provides the same amount of detail as 4 2k textures. As you don't have to have one object per texture, that texture can be reused across models. That'd reduce drawcalls which would be an optimisation, although consoles typically don't struggle with drawcalls. An 8k texture could cover the entire environment with low-res texturing, for example, far worse than streaming in lots of 512 pixel textures, or be an entire texture for one face in crazy detail that you can zoom really close to without pixelating.

 

[creaks]

That's bollocks talk. There's no such thing, and Fast Racing Neo won't look any more realistic than other racers out there, which don't even look real, let alone 'hyper real'. Shinen are great are getting the Wii U fans excited at their box and eager for games, but let's not use PR talk as a reference in a technical hardware thread.

As for 4k and 8k textures, that's just a structuring format. One 4k texture provides the same amount of detail as 4 2k textures. As you don't have to have one object per texture, that texture can be reused across models. That'd reduce drawcalls which would be an optimisation, although consoles typically don't struggle with drawcalls. An 8k texture could cover the entire environment with low-res texturing, for example, far worse than streaming in lots of 512 pixel textures, or be an entire texture for one face in crazy detail that you can zoom really close to without pixelating.

Well, I agree with the 'hyper realistic graphics' term... In the context you used. To be fair, the context they were talking about was specifically regaurding the old wii ware 40Mb size limit. But... arent you dismissing abyss a little lightly?

We know they are using bc1 compression, so its diffuse only for the 4-8k's. Knowing shin'en they will have small but sharp, probably generated specular, bump, and other layers on top of it streamed in when it gets close enough.

I dont think using an 8k texture for the entire environment would be a bad idea at all for a game like fast racing, which takes place on a hover track or something around 50 feet off the background environment. Youd never get the camera near the texture, you wouldnt have those godawful repeating texture blocks over and over and over and over and over *pukes*. And you wouldnt have to worry as much about managing speed with concern to texture streaming. It might actually be.... fast.

But thats just my immediate thought, shin'en/abyss will probably do something demoscene overkill like provide 4k/8k textures on every asset for no real reason or purpose, except because they can, particularly since their old demoscene buddies vd dev just went looney and built a complete modern game engine package completely out of assembly.

 

[Shifty Geezer]

Well, I agree with the 'hyper realistic graphics' term... In the context you used. To be fair, the context they were talking about was specifically regaurding the old wii ware 40Mb size limit. But... arent you dismissing abyss a little lightly?

No. 1) There's no such thing as hyper realistic graphics. That's a PR term to get fanboys excited over 'looks pretty' graphics. 2) I only pointed out that an announced use of 4 and 8k textures on its own doesn't mean much. That can be anything from 4k per shoelace to 4k for the entire scene. Use of 4k and 8k textures isn't anything impressive or to get excited about; it's how those textures are used. And remember Wii U has a 1 GB limit (despite having 2 GBs RAM...) so that's going to put a significant cap on the variety of textures. Effectively Wii U can achieve what PS360 did with twice the resolution assets, whether with more textures of the same res or bundled into large res textures.

I dont think using an 8k texture for the entire environment would be a bad idea at all for a game like fast racing...

It depends. Is the environment a tunnel or cave or a massive open world? How close does one get to the surfaces? The IQ comes from texel mapping which is a function of texel density, not total number of texels.

 

[creaks]

No. 1) There's no such thing as hyper realistic graphics. That's a PR term to get fanboys excited over 'looks pretty' graphics. 2) I only pointed out that an announced use of 4 and 8k textures on its own doesn't mean much. That can be anything from 4k per shoelace to 4k for the entire scene. Use of 4k and 8k textures isn't anything impressive or to get excited about; it's how those textures are used. And remember Wii U has a 1 GB limit (despite having 2 GBs RAM...) so that's going to put a significant cap on the variety of textures. Effectively Wii U can achieve what PS360 did with twice the resolution assets, whether with more textures of the same res or bundled into large res textures.

It depends. Is the environment a tunnel or cave or a massive open world? How close does one get to the surfaces? The IQ comes from texel mapping which is a function of texel density, not total number of texels.

Eh, I dont feel you adressed what I said very well... Like you were responding to someone who said something different. Is there really any point to dedicating a numbered response to saying there is 'no such actual technical thing as hyper realistic graphics'? I mean really? Did I give off the impression I thought shin'en was using the hyper realistic graphic shaders only available via sega's exclusive graphics core next level blast processing? Id hope I dont come across as quite that loopy. You bring up 360 games, when the 'pr marketing' was in comparison to wiiware games capped at 40 megabytes. I think its perfectly fair to use the term 'hyper realistic graphics', despite it not actually being an actual tangible 'thing' when comparing to a 40 megabyte downloadable wii game, and the restrisctions inherit within. If they were talking about 360/ps3 games, or really ANYTHING but wii ware, I would absolutely agree with you. They were answering a question specifically comparing to the restrictions of wii ware.

Im not particularly worried about abyss under delivering on their hype in the visuals department. I mean, NOTHING they have actually 'teased' has really been about anything 'pretty'... The closest they got was 'hyper realistic graphics' vs wiiware. Everything technical theyve talked about in regaurds to their wii u game has just been stuff thats real purpose is to clear up resources and overhead.... Like an old man picking up pennies... Its actually been rather frusterating that they are holding out on any real goods. They are rather good at what they do though. They can put an n64 to shame with an amiga. They generously used ambient occlusion on wii. They are old ones from the pcdemo days, its what they do, and they love nothing more than doing it without power. The games themselves are hit or miss... But they always look good.

If the original fast racing on wii is anything to go by, you wont be getting very close to the gound based background geometry, as the racing track is elevated quite high off the ground.

 

[Shifty Geezer]

You bring up 360 games, when the 'pr marketing' was in comparison to wiiware games capped at 40 megabytes. I think its perfectly fair to use the term 'hyper realistic graphics'.

I very much disagree. Firstly, as a term it'll be used without context, such as the post on this thread about Shinen using Hyper Realistic Graphics. Secondly it's still nonsensical. "More realistic" or "higher quality" is the correct term. Those are relative terms to the previous standard (use of qualifiers 'more' or 'higher'). "Now we don't have a 40 MB limit, we can use much higher quality textures. We can use more realistic textures." Use of 'hyper realism' is just PR speak to get people excited, using an absolute adjective, and the fact people will parrot a meaningless term with some emotional backing goes to show it worked. This is a thread about the hardware in Wii U, and the comment '4k textures and hyper realism' has been presented as some sort of evidence that Wii U can do something-or-other.

It can do 4k and 8k textures, is all the info that comment conveys.

 

[creaks]

I very much disagree. Firstly, as a term it'll be used without context, such as the post on this thread about Shinen using Hyper Realistic Graphics. Secondly it's still nonsensical. "More realistic" or "higher quality" is the correct term. Those are relative terms to the previous standard (use of qualifiers 'more' or 'higher'). "Now we don't have a 40 MB limit, we can use much higher quality textures. We can use more realistic textures." Use of 'hyper realism' is just PR speak to get people excited, using an absolute adjective, and the fact people will parrot a meaningless term with some emotional backing goes to show it worked. This is a thread about the hardware in Wii U, and the comment '4k textures and hyper realism' has been presented as some sort of evidence that Wii U can do something-or-other.

It can do 4k and 8k textures, is all the info that comment conveys.

Sounds like a bit much to fit in a tweet.

And I think you might be blowing this out of proportion, and doing a bit of projection...

So far, a PERSON has taken it out of context on this forum. Im sure there there are others, the internet is a large place, but this is far from the all encompassing everyone generalization. And Im sure there are going to be people around to put the statement back in its context like you did.

But I think you are projecting a bit much on the nefarious pr marketing genious department expertly taking advantadge of of over enthusiastic rubes.

They dont have a pr department. They cant even afford to make hq promo art for advertisements. They are just five dudes who dont really speak English all that much, let alone use the language to surgically manipulate.

I agree with your end statement. I do. It just means it can do 4k-8k textures... Which... We already knew. Which is why I didnt bring it here a day or so ago when I saw it. However I really dont think they implied anything more than that, certainly not purposefully, and certainly not if you pay attention to the conversation, They are talking about wiiware after all.

But I think at this point we are just arguing semantics... Over a point we agree on.

 

[AlNom]

Seems they're using 4-8K for BC1 textures:

https://twitter.com/DavidGX/status/403386691022360576

 

[Commenter]

Is there any technical reason why Nintendo couldn't lower the amount of memory reserved for the OS and up the space that games can use, such as 1.5Gb/ 512 Mb split, in the same way Sony lowered the reserved ram for OS on PS3?

 

[BRiT]

Is there any technical reason why Nintendo couldn't lower the amount of memory reserved for the OS and up the space that games can use, such as 1.5Gb/ 512 Mb split, in the same way Sony lowered the reserved ram for OS on PS3?

I'm frankly amazed at the sure lack of services and features the WiiU OS provides.

No technical reason why a competent development team couldn't do that. I'll leave it up to the reader to decide if Nintendo is or is not competent.

 

[bgassassin]

No prob, bg. I hope gipsel can weigh in, but let me take a stab at this just to see if I'm getting things straight myself. From what I gather, if a wavefront (64 threads/work items) is executed over four cycles, then each cycle 16 threads are executed by a SIMD engine. This is one thread for each of "ALUs" as Nintendo lists them (group of 5 shaders).

I've read that other AMD GPUs have a limit of 8 "work groups" per SIMD engine and that each work group is 1 or 4 wavefronts. Thus, if this is also true of Latte, it would seem that total, there can be up to 64 wavefronts in flight at any given time. 192 seems to be a global limit in the command processor.

As for the TMU to ROP ratio - some quick guesses. ROPs were needed to for the additional fillrate of the Gamepad? More TMUs would be bottlenecked by low MEM1 bandwidth? TMUs in Xenos were possibly bottlenecked to some degree? Or splitting the SIMD engines would require more complexity, which Nintendo vetoed? That would be wavefront size of 32. Or they just had a set silicon budget and two more sets of TMUs and L1 caches would take up to much space (this probably in combination with some of the other factors)

If it has just two full size SIMDs (so 16x5 ALUs and 64 elements wide wavefronts), it is actually limited to less. IIRC, the GPR allocation has an allocation of 4 float4 registers (and a few [4 by default I think] per SIMD are usually reserved for clause temporary registers). That means you can't keep more than 63 wavefronts in flight on each SIMD (would be 126 in total) at the absolute maximum. And this is only true for absolutely minimal shaders requiring not more than 4 registers (and there could be even a minimum allocation of 8 registers, don't remember and would have to look it up). That's why I think 192 is very generous. Latte will never be limited by this number (also evident by the fact that a RV770 with 10 SIMDs was fine with 256 wavefronts for the whole chip). But maybe it made no sense to invest time and money to scale the "ultra threaded dispatch processor" further down. Or there are additional small gains under some circumstances to have wavefronts waiting in there without any resources (GPRs, LDS) of the SIMD allocated to them. But nothing jumps to my mind (as the data for the wavefronts have to be stored somewhere). No idea.

4 ROPs may have been a bottleneck for 1080p titles?

Gipsel's post finally made it click.

And yeah the only thing I could think of was to increase the pixel fillrate.

 

[Melqart]

Is there any technical reason why Nintendo couldn't lower the amount of memory reserved for the OS and up the space that games can use, such as 1.5Gb/ 512 Mb split, in the same way Sony lowered the reserved ram for OS on PS3?

There shouldn't be any technical reason they can't. 1GB seems like a waste of RAM considering their UI and background services. What they have now should easily be able to fit in 512MB or less (360 only uses 32MB IIRC). They probably wanted to play it safe at first (they did something similar with the 3DS and opened up a little extra hardware after a while).

 

[Grall]

"Playing it safe" in this case though (reserving 1GB for Nintendo's barebones crappy OS) is like making a bicycle path 20 meters wide just in case you need to swerve for a baby carrier or something like that...

 

[TheLump]

"Playing it safe" in this case though (reserving 1GB for Nintendo's barebones crappy OS) is like making a bicycle path 20 meters wide just in case you need to swerve for a baby carrier or something like that...

I wouldn't call it "barebones". but yeah it's certainly doing nothing to require 1GB of RAM right now. My guess is it's for some future features which they're working on.

Nintendo are many things, but they certainly aren't wasteful with their software or hardware. It's unlikely it will go unused in the long run - one way or the other (ie for o/s or unlocked to devs)

 

[creaks]

Well, the overview page of espresso has been leaked.

https://marcansoft.com/transf/espresso_intro.png

Not too surprising. Not really surprising at all.

Its, as established a power pc based off the broadway power pc architecture, with improvements to flop performance and data transfer capability. The standard Nintendo enhancements to the ppc line to make them 'nintendo ppc'.

Since there is obviously no new additions like an actual simd unit, the nintendo custom fpu likely just got a simple upgrade, and obviously not an overhaul, probably a little wider, coupled with the increased clock and cache's is nets that improved flop performance an espresso core has over broadway.

L2 cache sizes are already known, however we now have confirmation that they bumped up the l2 cache's from 2 way associative in broadway and gekko to 4 way in the espresso cores. Makes sense, I dont really see any reason to go higher than 4 way with 512 Kb- 1Mb cache's, miss should be dang near identical even compared with full.

No such info for l1, but I see no reason for them to go beyond the 8 way present in broadway.

Really wish the rest of this would surface.

 

[haihoo]

And remember Wii U has a 1 GB limit (despite having 2 GBs RAM...) so that's going to put a significant cap on the variety of textures. Effectively Wii U can achieve what PS360 did with twice the resolution assets, whether with more textures of the same res or bundled into large res textures.

Strictly speaking the Wii U can achieve more than "twice" the resolution assets. The PS360 has 512MB RAM but the OS reserved RAM there too. Wii U games can use a full 1GB of RAM for games.

"Playing it safe" in this case though (reserving 1GB for Nintendo's barebones crappy OS) is like making a bicycle path 20 meters wide just in case you need to swerve for a baby carrier or something like that...

The Wii U has a full featured browser with tabs. You can use it while playing (you can use the Miiverse and eShop too, but I think they are less demanding). This is where most of this 1GB OS RAM is going to. But I definitely expect that a future update will make its possible to use more than 1GB RAM for games. The amount of RAM the OS used on PS360 has been shrunk too after some OS optimizations.