• 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.
I'm sure he's bitter. Two years ago he was a regular poster in the WiiU topics. Never had any problems with him. Then last year, with the first evidence that WiiU was not what bgassassin had us believe, he started trolling it. He was banned for a while, i can only assume that had something to do with it, but i'm not sure. But i have no problem with him not liking the WiiU, thinking it is not powerful or whatnot. I do have a problem with broken logic and changing goalposts every time you're proven wrong. Launch games were made on poorly documented hardware, there was a dev saying only one core of the CPU was available, ports were made by small teams (the guys from Darksiders said with how many they were, it was an insanely low number, i think it was 4 or 5 people), usually outsourced, new to the hardware... Obviously, games would see drastic improvements over time. His notion that the hardware design is almost identical to that of the 360 is just too crazy. What have we been doing the past year in this thread of not scratching our heads that the setup was unlike anything anybody had thought of.

Log4Girlz vs DragonSworne, battle of the jaded Wii U supporters turned haters
 

SerodD

Member
This was on gamefaqs, anything to it?
Wii U CPU:
- Tri Core IBM 45nm PowerPC750CL/G3/Power7 Hybrid
- Core 0; 512KB Core 1; 2MB Core 2; 512KB L2 Cache
- Clocked at 1.24Ghz
- 4 stage pipeline/not Xenon-Cell CPU-GPU hybrid
- Produced at IBM advanced CMOS fabrication facility
- Can use eDRAM as cache(!?)
- Dual Core ARM for background OS tasks clocked at 1Ghz with 64KB L1 cache per core

Wii U Memory:
- DDR3 2GB, 1GB for OS, 1GB for Games (12.8/Gbps is speculation)
- eDRAM 32MB+4MB Memory/VRAM/Cache(UNIFIED, 4MB for gamepad)
- Clocked at 550Mhz
- CPU for Cache and GPU for VRAM use eDRAM
- eDRAM acts as unified memory, similar to AMD’s hUMA/HSA by function/behaviour(?!)
- Little to no latency between CPU and GPU with eDRAM Cache/Memory/VRAM

Wii U GPU:
- VLIW 5/VLIW4 Radeon HD 5000/6000
- DirectX 11/Open GL 4.3/Open CL 1.2/Shader Model 5.0
- Supports GPGPU compute/offload
- Customized, Using custom Nintendo API codenamed GX2(GX1 Gamecube)
- Clocked at 550Mhz
- Produced at TSMC
- Uses eDRAM as VRAM
DirectX can only be used by Microsoft thus silicon/hardware/feature is removed replacing with transistors for performance and maintaining full Open GL 4.3/Open CL 1.2/Shader Model 5.0/GPGPU compatibility

Wii U Note;
- Can’t use DirectX 11 except Open GL 4.3 that is ahead DX11
- Easier to program, no bottlenecks causing trouble like on Xbox 360 and PlayStation 3
- more efficient hardware with no bandwidth bottlenecks
- Most early ports and even now use 2 cores and 3rd is barely used #fact
- Wii U CPU has much higher operations per cycle than Xbox 360/PlayStation 3
- It is maybe in fact most efficient performance per watt in the world in terms of 45/40nm chips
- Power7 architecture allows shave off 10% of texture size without loss of quality(special compression)

It is ahead of 7th generation, developers will learn to handle and optimize for Wii U properly. I was digging on the internet for various information’s for a month and all you read is what I compiled and gathered. It can handle 1080p because it has 3.5 times more eDRAM than Xbox 360 also has no severe memory/bandwidth bottlenecks that Xbox 360/PlayStation 3 have all over the system with their FSB that is choking. Xbox 360′s GDDR3 that has theoretical 22-28Gbps is reduced to 10Gbps thanks to piss poor FSB.

Wii U has same amount of eDRAM as Xbox One, remember that and also Wii U’s CPU utilizes Power7 architecture a bit so maybe just maybe Wii U’s CPU can use eDRAM as CPU cache as GPU uses it as VRAM so in a way something like AMD’s hUMA/HSA is possible on Wii U.

I have to ask, since I don't know much about hardware, is this good news? Does it mean the console might be a little better than previously thought? Weren't some developers saying it only ran the open GL equivalent of DirectX 10.1?
 
I have to ask, since I don't know much about hardware, is this good news? Does it mean the console might be a little better than previously thought? Weren't some developers saying it only ran the open GL equivalent of DirectX 10.1?

Bg would more than likely squash this rumor or what you want to call it.
 
This was on gamefaqs, anything to it?
Wii U CPU:
- Tri Core IBM 45nm PowerPC750CL/G3/Power7 Hybrid
- Core 0; 512KB Core 1; 2MB Core 2; 512KB L2 Cache
- Clocked at 1.24Ghz
- 4 stage pipeline/not Xenon-Cell CPU-GPU hybrid
- Produced at IBM advanced CMOS fabrication facility
- Can use eDRAM as cache(!?)
- Dual Core ARM for background OS tasks clocked at 1Ghz with 64KB L1 cache per core

Wii U Memory:
- DDR3 2GB, 1GB for OS, 1GB for Games (12.8/Gbps is speculation)
- eDRAM 32MB+4MB Memory/VRAM/Cache(UNIFIED, 4MB for gamepad)
- Clocked at 550Mhz
- CPU for Cache and GPU for VRAM use eDRAM
- eDRAM acts as unified memory, similar to AMD’s hUMA/HSA by function/behaviour(?!)
- Little to no latency between CPU and GPU with eDRAM Cache/Memory/VRAM

Wii U GPU:
- VLIW 5/VLIW4 Radeon HD 5000/6000
- DirectX 11/Open GL 4.3/Open CL 1.2/Shader Model 5.0
- Supports GPGPU compute/offload
- Customized, Using custom Nintendo API codenamed GX2(GX1 Gamecube)
- Clocked at 550Mhz
- Produced at TSMC
- Uses eDRAM as VRAM
DirectX can only be used by Microsoft thus silicon/hardware/feature is removed replacing with transistors for performance and maintaining full Open GL 4.3/Open CL 1.2/Shader Model 5.0/GPGPU compatibility

Wii U Note;
- Can’t use DirectX 11 except Open GL 4.3 that is ahead DX11
- Easier to program, no bottlenecks causing trouble like on Xbox 360 and PlayStation 3
- more efficient hardware with no bandwidth bottlenecks
- Most early ports and even now use 2 cores and 3rd is barely used #fact
- Wii U CPU has much higher operations per cycle than Xbox 360/PlayStation 3
- It is maybe in fact most efficient performance per watt in the world in terms of 45/40nm chips
- Power7 architecture allows shave off 10% of texture size without loss of quality(special compression)

It is ahead of 7th generation, developers will learn to handle and optimize for Wii U properly. I was digging on the internet for various information’s for a month and all you read is what I compiled and gathered. It can handle 1080p because it has 3.5 times more eDRAM than Xbox 360 also has no severe memory/bandwidth bottlenecks that Xbox 360/PlayStation 3 have all over the system with their FSB that is choking. Xbox 360′s GDDR3 that has theoretical 22-28Gbps is reduced to 10Gbps thanks to piss poor FSB.

Wii U has same amount of eDRAM as Xbox One, remember that and also Wii U’s CPU utilizes Power7 architecture a bit so maybe just maybe Wii U’s CPU can use eDRAM as CPU cache as GPU uses it as VRAM so in a way something like AMD’s hUMA/HSA is possible on Wii U.

So much incorrect here. Enough is enough. It's amazing, yes, but nobody outside the dev community knows more about Wii U's innards than some of the people in this thread. It's a DX 10.1 equivalent part. No power 7 architecture. It's not a TSMC part. People need to just...stop with the nonsense.
 

AzaK

Member
I have to ask, since I don't know much about hardware, is this good news? Does it mean the console might be a little better than previously thought? Weren't some developers saying it only ran the open GL equivalent of DirectX 10.1?
Correct. It's not DX 11 equivalent. And it's SM 4.1 (IIRC)
 

The_Lump

Banned
So much incorrect here. Enough is enough. It's amazing, yes, but nobody outside the dev community knows more about Wii U's innards than some of the people in this thread. It's a DX 10.1 equivalent part. No power 7 architecture. It's not a TSMC part. People need to just...stop with the nonsense.

What other nonsense has been flying around? Did I miss some juicy garbage?! ;)

Yeah that list is mostly guff. A lot is just purged from the various speculation threads here and elsewhere.

The only similarity to power 7 is that the CPU can access the eDRAM as a cache according to Shin'en (which I think was a major boon for power7 at the time??), but it certainly wasn't "based" on it or anything like though.

The list reads like a 'Best Of WiiU Speculation 2013' compilation or something. Not worth posting here.
 

prag16

Banned
So much incorrect here. Enough is enough. It's amazing, yes, but nobody outside the dev community knows more about Wii U's innards than some of the people in this thread. It's a DX 10.1 equivalent part. No power 7 architecture. It's not a TSMC part. People need to just...stop with the nonsense.

I thought you'd know as well as anyone at this point that saying "10.1 equivalent part, period" isn't quite the whole story, no? Don't feed into the very misinformation cycle you're trying to combat. Just saying...
 
A

A More Normal Bird

Unconfirmed Member
We know that it has at least some DX11 equivalent features due to the logs from Slightly Mad Studios for Project CARS, don't we..? Unless I've missed something..?

I thought you'd know as well as anyone at this point that saying "10.1 equivalent part, period" isn't quite the whole story, no? Don't feed into the very misinformation cycle you're trying to combat. Just saying...
AMD R700 DX10.1 equivalent hardware + closed box & custom API = DX11 featureset, pretty much.
 
What other nonsense has been flying around? Did I miss some juicy garbage?! ;)

Yeah that list is mostly guff. A lot is just purged from the various speculation threads here and elsewhere.

The only similarity to power 7 is that the CPU can access the eDRAM as a cache according to Shin'en (which I think was a major boon for power7 at the time??), but it certainly wasn't "based" on it or anything like though.

The list reads like a 'Best Of WiiU Speculation 2013' compilation or something. Not worth posting here.

True, haha. As for other nonsense, it's not so much in this thread as elsewhere on the net these days (check out the Latte Wikipedia page for example).

We should also probably specify that MEM1 isn't really a cache. It doesn't move data to and fro automatically like a cache does. For the CPU, I believe Shin'en called it a "scratchpad." POWER7's eDRAM is on-die L3 cache, so I don't think there's much comparison. The CPU's L2 is eDRAM, but that's also a different function from POWER7, so I find most mentions of that architecture do more harm than good.

AMD R700 DX10.1 equivalent hardware + closed box & custom API = DX11 featureset, pretty much.

Yeah, pretty much this. Although I wonder how the GX2 updates are coming along. Last I heard, compute shaders weren't fully enabled.
 
What's been bugging me is SMS use of multi threaded rendering. Was this not available before system launch and has now been enabled in GX2?

Another thing that crossed my mine is Espresso having a core with eDRAM and its purpose being,"decompression".
 

Oblivion

Fetishing muscular manly men in skintight hosery
I've been trying to follow this thread, but it's been so confusing for the past 10-20 pages. What is the deal with the 160 shader number everyone keeps throwing around? Is it lower than what we though it was supposed to be? What was the number people were expecting it to be around?


And lastly, do you guys think Nano Assault looks better than Super Stardust HD?
 
I've been trying to follow this thread, but it's been so confusing for the past 10-20 pages. What is the deal with the 160 shader number everyone keeps throwing around? Is it lower than what we though it was supposed to be? What was the number people were expecting it to be around?


And lastly, do you guys think Nano Assault looks better than Super Stardust HD?
Fourth Storm was pretty sure about that number for awhile now, while others weren't as certain but considered that as a possibility. Only a few people posted, afaik, didn't take what fourth said in consideration.

The issue some people, including me at a certain point, don't think that a GPU with 160 shader units shouldn't be showing the results that we are seeing in some of the newer games and/or the size of certain parts of the die seems to bigger we would have expected it to be.
 

Xun

Member
It may have been mentioned, but does SM3DW have anti-aliasing?

Some shots seem to have it, whilst others seem to lack it.
 

krizzx

Junior Member
We know that it has at least some DX11 equivalent features due to the logs from Slightly Mad Studios for Project CARS, don't we..? Unless I've missed something..?

I still don't see why people keep trying to nail it as DX anything.

The API it uses is GX2 and I thought that was based on OpenGL. The feature set avilable in it could be anything for all we know.

So far, I've never seen a change log list it as using a DX10 implementation of something for anything. Its always listed as using DX11 features specifically. There also that one Project C.A.R.S. where it listed the PS4 and Wii U version being updated with the exact same exact update at the same time. Would be odd for them to be using DX10 on the PS4, or PS4 features on the supposedly less featured Wii U. This is all that can be confirmed from what I've come across.

Anything aside form that is pure subjective guessing in my opinion.
 

prag16

Banned
I still don't see why people keep trying to nail it as DX anything.

The API it uses is GX2 and I thought that was based on OpenGL. The feature set avilable in it could be anything for all we know.

So far, I've never seen a change log list it as using a DX10 implementation of something for anything. Its always listed as using DX11 features specifically. There also that one Project C.A.R.S. where it listed the PS4 and Wii U version being updated with the exact same exact update at the same time. Would be odd for them to be using DX10 on the PS4, or PS4 features on the supposedly less featured Wii U. This is all that can be confirmed from what I've come across.

Anything aside form that is pure subjective guessing in my opinion.

The hardware from which it's derived was originally released during the 10.1 "era".

So while saying "it's a DX 10.1 part" isn't wholly inaccurate, it's also a little disingenuous just leaving it at that.

No AA AFAIK.

Looking closely, the shots don't appear to have any AA. Doesn't mean many of them don't look extremely nice.
 

gamingeek

Member
I know this isn't the thread for it, bit if anyone's interested I uploaded ton of NFS U screens here.
http://thevgpress.com/forumtopics/t...ow-in-hd_2476_3.html&perPage=20#comment179323

wiiu_screenshot_tv_01d1dpz.jpg

wiiu_screenshot_tv_015kqb6.jpg

wiiu_screenshot_tv_01ikr3z.jpg


This game is fairly depressing. I didn't even think it looked that good when I played it. But can you imagine if all Wii U ports used high resolution textures like this? It took Criterion 10 minutes work to flip a PC textures switch in their pipeline. What's more depressing is that this is an open world game you scream across at 190 mph and there are no framerate problems, no loading problems. The world loads once when you start the game and you can drive anywhere with no further loading. If they can do this in open world then imagine what smaller games in smaller environments could look like? Bah.
 
I know this isn't the thread for it, bit if anyone's interested I uploaded ton of NFS U screens here.
http://thevgpress.com/forumtopics/t...ow-in-hd_2476_3.html&perPage=20#comment179323

wiiu_screenshot_tv_01d1dpz.jpg

wiiu_screenshot_tv_015kqb6.jpg

wiiu_screenshot_tv_01ikr3z.jpg


This game is fairly depressing. I didn't even think it looked that good when I played it. But can you imagine if all Wii U ports used high resolution textures like this? It took Criterion 10 minutes work to flip a PC textures switch in their pipeline. What's more depressing is that this is an open world game you scream across at 190 mph and there are no framerate problems, no loading problems. The world loads once when you start the game and you can drive anywhere with no further loading. If they can do this in open world then imagine what smaller games in smaller environments could look like? Bah.

Top and bottom images seem to be void of any tesselation (?).

But I really can't tell about the central one with rocks. Seems like it's using some sort DX11 feature to bump the... bumps ?

Extreme tech noob in here. Reply with caution and simple wording. Thanks.
 

gamingeek

Member
My understanding of tessellation is that it adds geometry to the rocks? So like the bricks would jut out, which they don't. No idea on the rocks but some get more love than others.
 

prag16

Banned
My understanding of tessellation is that it adds geometry to the rocks? So like the bricks would jut out, which they don't. No idea on the rocks but some get more love than others.

I think you're thinking of displacement mapping.

Tessellation is for smoothing out blocky objects making poly counts appear higher than they actually are.

I think.
 

ozfunghi

Member
Yeah... it's like a real time subdivision modifier. Devs don't use it a lot because it's hard to control. Maybe you'd want to have it applied to a certain part of a mesh, but not the other part(s) (not everything is supposed to be rounded off). At least that's how i understand. It is also used in combination with normal displacement maps i believe, for terrain and rocks, to get better results and better detail from the map. Maybe that's where the confusion comes from.

Edit: see http://www.youtube.com/watch?v=-uavLefzDuQ
 

tipoo

Banned
So much incorrect here. Enough is enough. It's amazing, yes, but nobody outside the dev community knows more about Wii U's innards than some of the people in this thread. It's a DX 10.1 equivalent part. No power 7 architecture. It's not a TSMC part. People need to just...stop with the nonsense.

Indeed. Every once in a while someone comes by and posts what some website compiled as the "official" Wii U specs list, but glaring problems with their information prove they don't know anything special that we don't. Another issue is it says the 4MB embedded memory is used for the gamepad, but there's 2MB eDRAM and 1MB SRAM in split pools. Them being used for the gamepad would be interesting though.

The hastag fact sure adds to their credibility :/

The only similarity to power 7 is that the CPU can access the eDRAM as a cache according to Shin'en (which I think was a major boon for power7 at the time??), but it certainly wasn't "based" on it or anything like though.

Did they actually say the 32MB eDRAM? Because the CPU has eDRAM on its own, that's a feature from Power7 right there. The Xbox ONe CPU can't actually directly access the 32MB eSRAM on the GPU, it has to copy to main memory first, so the memory is completely dedicated to GPU. Do we know which case is true here? If it's also true that the eDRAM can be accessed at 32(or whatever)GB/s, I'm not sure splitting it would be ideal anymore, as you'd want the GPU to have all the bandwidth if it's that little.

I've been trying to follow this thread, but it's been so confusing for the past 10-20 pages. What is the deal with the 160 shader number everyone keeps throwing around? Is it lower than what we though it was supposed to be? What was the number people were expecting it to be around?

Before we had the picture on the first page, yes, 160 was considerably lower than we were expecting (iirc before the launch and teardowns, people were citing 320 shaders as poor, 480 as decent, 5** as good, etc). but since we got the picture, most reasonable people were saying 160 all along, and there were only a few 320 shader holdouts. Then we got more confirmation of 160.
 

fred

Member
Yeah... it's like a real time subdivision modifier. Devs don't use it a lot because it's hard to control. Maybe you'd want to have it applied to a certain part of a mesh, but not the other part(s) (not everything is supposed to be rounded off). At least that's how i understand. It is also used in combination with normal displacement maps i believe, for terrain and rocks, to get better results and better detail from the map. Maybe that's where the confusion comes from.

Edit: see http://www.youtube.com/watch?v=-uavLefzDuQ

Watching that video makes me think that Super Mario 3D World and Bayonetta 2 are both using tessellation. Could explain how everything looks so round in the former and how everything looks so high poly in the latter. I've thought that Bayonetta 2 has been using it since we first saw it.

What do you all think..?
 

NBtoaster

Member
It might have a limited form of post-AA but a lot of edges get no treatment. Aliasing in the background is hidden by thick DOF and bloom anyway.
 

HTupolev

Member
Really? I can a lot of aliasing in some shots, but many don't seem to have any apparent aliasing.

Take this shot for example:

wiiu_screenshot_tv_01n4l3u.jpg
That's weird actually.

TtaJuJV.png


Some edges that look reasonably clean are either aligned perfectly with the pixel grid, or else are highly rotated relative to it. Edges that don't fit the criteria, like on the lower left of the block, sometimes appear to have no antialiasing.

But then there are edges like the upper left of the block... I'm not sure what's going on there. Would be nice to have a WiiU in front of me so I could see what it looks like in motion.
 
That's weird actually.

TtaJuJV.png


Some edges that look reasonably clean are either aligned perfectly with the pixel grid, or else are highly rotated relative to it. Edges that don't fit the criteria, like on the lower left of the block, sometimes appear to have no antialiasing.

But then there are edges like the upper left of the block... I'm not sure what's going on there. Would be nice to have a WiiU in front of me so I could see what it looks like in motion.

It doesn't make sense to me.
 

OryoN

Member
Nintendo is known to use various FX only where it's absolutely necessary, or for gameplay & functionality purposes. I don't know we can make a case for AA not being absolutely necessary at all times, but certainly, lacking it wouldn't be as bad in some areas, as it may in others. So using AA sparingly - to keep a high performance - isn't out of the question. It's not uncommon for Nintendo to use this approach(though, doing it with AA may be a first). If you remember Luigi's Mansion 2: Dark Moon, you'll notice they didn't just toss in real-time projected shadows in the entire game. Certain areas/rooms have it, while others don't.
 

Doczu

Member
Nintendo is known to use various FX only where it's absolutely necessary, or for gameplay & functionality purposes. I don't know we can make a case for AA not being absolutely necessary at all times, but certainly, lacking it wouldn't be as bad in some areas, as it may in others. So using AA sparingly - to keep a high performance - isn't out of the question. It's not uncommon for Nintendo to use this approach(though, doing it with AA may be a first). If you remember Luigi's Mansion 2: Dark Moon, you'll notice they didn't just toss in real-time projected shadows in the entire game. Certain areas/rooms have it, while others don't.

Yup, just like Silent Hill Origins on the PSP (and propably on the PS2 too). Open spaces had the simplest light/shadow effects, but the closed spaces in the otherworld had some rooms with real-time shadows on enemies and objects). Beautiful little game it was.
 

krizzx

Junior Member
Top and bottom images seem to be void of any tesselation (?).

But I really can't tell about the central one with rocks. Seems like it's using some sort DX11 feature to bump the... bumps ?

Extreme tech noob in here. Reply with caution and simple wording. Thanks.

Its probably just parallax mapping in the center.




Thanks for pointing this out. Once again, we have a dev directly saying the GPU has DX11 features.
 

Oblivion

Fetishing muscular manly men in skintight hosery
Before we had the picture on the first page, yes, 160 was considerably lower than we were expecting (iirc before the launch and teardowns, people were citing 320 shaders as poor, 480 as decent, 5** as good, etc). but since we got the picture, most reasonable people were saying 160 all along, and there were only a few 320 shader holdouts. Then we got more confirmation of 160.

How many shaders does Xenos and RSX have?
 

Log4Girlz

Member
Ass Creed 4

http://www.eurogamer.net/articles/digitalfoundry-assassins-creed-4-next-gen-face-offace-off

Wii U:


The Wii U version consistently falls 6-10fps behind the 360 and PS3 releases, resulting in a jerky, often uneven update and a profound, unsatisfying reduction in controller response. Combined with the GamePad's twitchy analogue sticks, the low frame-rate makes frantically running across rooftops and chasing down targets a frustrating affair, the net result being that we'd often have to repeat some missions multiple times - something that wasn't much of a problem on 360, and with no issues at all when playing at 60fps on the PC version of the game.

Ouch.
 

tipoo

Banned
How many shaders does Xenos and RSX have?

They were counted differently back then. The 48 shaders in the Xenos would be something like a 240 shader part now. The 240 coming from (48*5 vector units). Whereas now they just give the total count from the start.
 

Oblivion

Fetishing muscular manly men in skintight hosery
Apart from Ryse, does it have any not-ass looking games?

True.

They were counted differently back then. The 48 shaders in the Xenos would be something like a 240 shader part now. The 240 coming from (48*5 vector units). Whereas now they just give the total count from the start.

240? WTF? Please tell me that shaders are somewhat like clockspeeds in that they're not really good indicators of a GPU's power.

I mean, up until this shader business, the GPU was supposed to be the one thing that was definitively superior in all aspects compared to Xenos.
 

tipoo

Banned
240? WTF? Please tell me that shaders are somewhat like clockspeeds in that they're not really good indicators of a GPU's power.

I mean, up until this shader business, the GPU was supposed to be the one thing that was definitively superior in all aspects compared to Xenos.

They aren't comparable across generations, I'm sure the 160 are more efficient than the 240. However with the clock speeds I'm also sure they're in the same ballpark, it's not like they'll be well over 1.5x as efficient, GPUs have gained most of their performance over these years by going with more and more shaders and bandwidth, the efficiency of singular units hasn't risen that much.

Also, incidentally, both have 8 ROPs.

http://en.wikipedia.org/wiki/Xenos_(graphics_chip)
 

tipoo

Banned
Digital Foundry is finally doing Wii U exclusives, here's 3D World

http://www.eurogamer.net/articles/digitalfoundry-vs-super-mario-3d-world

Last year's New Super Mario Bros U already gave us a taste of what a higher resolution can do for Nintendo's designs but 3D World provides an even clearer example of this. Assets appear to have been created from the ground up for the Wii U this time around and image quality is sharp and clean, operating with a full 1280x720 framebuffer. Basic edge-detection, not unlike what was utilised in NSMBU, handles anti-aliasing duties but coverage can be a bit spotty at times - results are surprisingly inconsistent on a per-stage basis, with certain areas appearing remarkably free of aliasing while others seem quite jaggy. Thankfully, the smart visual design generally overcomes most instances where this would prove bothersome. While we would like to see Nintendo embrace a more sophisticated approach to anti-aliasing in the future, it's hard to complain about the end results here.
 

fred

Member
You've also got to remember that if this 160 shaders info is true then there has to be something else going on under the hood that we're unaware of, I'm still firmly in the Fixed Functions camp. It explains why the ALUs are almost twice the size that they should be and how it can run something like the Bayonetta 2 demo drawing so little juice.
 

tipoo

Banned
You've also got to remember that if this 160 shaders info is true then there has to be something else going on under the hood that we're unaware of, I'm still firmly in the Fixed Functions camp. It explains why the ALUs are almost twice the size that they should be and how it can run something like the Bayonetta 2 demo drawing so little juice.

I think the "something" is new shader models, the probably much better tesselator, and double the RAM of the last gen consoles. Honestly I don't see why people are finding 160 without some magical fixed function thing so hard to believe, as you just have to look at what completely unexpected things came out of the Geforce 7800-Radeon X1800 like chips in the PS360. Who would think the 7800 could run God of War 3 or the x1800 Halo 4, or either of them Revengeance at mostly 60FPS?

Also, the die size thing again, different fabrication processes target different optimizations, they could have optimized for low power draw with the layout, or high yields, etc rather than what the original GPU did which could make them fatter. And not only that, different fabrication plants have different sizes for everything, Intels 22nm is smaller than most other peoples 20nm, etc etc.

This is reminiscent of what happened with the Wii, a bunch of theories on why it was better than expected but little that panned out.
 

AmyS

Member
They were counted differently back then. The 48 shaders in the Xenos would be something like a 240 shader part now. The 240 coming from (48*5 vector units). Whereas now they just give the total count from the start.

Yeah, Xbox 360's Xenos had a unified shader architecture: 48 Vec5 units, so 240 shader ALUs or whatever you wanna call them. The 8 ROPs were on the EDRAM die as part of the GPU package, before the 360 Slim came along where the CPU, parent GPU shader core and EDRAM daughter die were merged into a single SoC.


PS3's RSX was based on an older Nvidia design (even though PS3 launched a year later) derived from the NV47 aka G70 / GeForce 7800 (although RSX was not a full-spec 7800). NV47 was Nvidia's 2005 refresh of the NV40 / GF 6800 from 2004. RSX was a custom variant of that refresh for PS3. It did not have a unified shader architecture (Nvidia did not have USA until G80 / GF 8800 in fall 2006, too late to be used in RSX for PS3). RSX had 8 vertex-shader pipelines and 24 pixel-shader pipelines. 8 ROPs. The full G70 / GF 7800 for PCs had 16 ROPs, 256-bit memory bus, thus clock for clock double RSX's fillrate, and greater bandwidth since RSX used a 128-bit memory bus.

Both Xenos and RSX had different strengths & weakness but overall Xenos was considered to be 'better' by most devs and was certainly more efficient than RSX because devs could allocate as many of Xenos' unified shaders/ALUs toward either vertex or pixel shading operations. Plus there was the extremely high-bandwidth internal bandwidth from EDRAM to the ROPs. The EDRAM's external bandwidth to the shader core was reasonably high. While these cannot be added up, the effect is that it took much of the burden of graphics rendering off the main external 128-bit memory bus to the GDDR3 memory.

Devs later learned to employ CELL to assist RSX, especially with vertex shader stuff since RSX only had 8 vertex-shader pipes, AFAIK it could not use the 24 pixel-shader pipes for any vertex stuff etc. Xenos could allocate any of its shader-ALUs for either type of operation.

I do not understand much about Wii U's GPU. It might only have 160 shader ALUs but they're certainly more capable and more efficient than Xenos' or RSX's Latte is at least a shader model 4.1 equivalent GPU, probably with stuff that goes beyond 4.1. -RSX was shader model 3.0, while Xenos went a bit beyond 3.0.
 

tipoo

Banned
Yeah, to be precise the Xenos was x1800-ish throughput but with HD 2000 series like features in some regards even well before the latter chip launched, and the RSX was a 7800-ish base but with a few things cut down to 7600 level hardware. I aways thought it was a shame the Cell wasn't paired with the Xenos instead last gen, as any advantage it had was spent trying to make up for the crappy RSX, although even then the Cell was hard to use. Unified shaders were a nice efficiency jump for everyone, reducing unused compute hardware.

I do think the 160 shaders could match or surpass the old 240, but excluding Nvidias old "big" shader design approach (which they canned for a similar "small" shader approach as AMD) the individual units havn't gotten all that much more efficient since unified shaders first hit, rather shrinks allowed more per die size and watt.
 

krizzx

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

WiiU_screenshot_TV_01061.jpg


Either the Wii U GPU can output far more polygons than we realized, or this is using tessellation.
 
How on earth are you able to tell how many polys are in this screen?
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.
 
Status
Not open for further replies.
Top Bottom