TheRedSnifit
Member
Thanks! And is there any way to determine what these "customized and unknown" features are? And what does all this mean in regards to power?
-
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
- Thread starter Thraktor
- Start date
- Status
shinra-bansho
Member
Sorry if dumb question, but is Hector Marcan saying he thinks all of the on-die RAM is 1T-SRAM or that only the additional RAM pools uncovered by this dieshot are 1T-SRAM?
Elfforkusu
Member
* most likely 352 GFlops (1.5x 360)
* heavily customized design, with about half the die spent on "special sauce", e.g. fast hardware implementations of "common" subroutines. What the special sauce is and how effective it will be for third party development is an enigma wrapped in a riddle shrouded by mystery. However, it will almost certainly boost graphical performance above the pure flops number, at least for developers who know how to use the hardware. (Nintendo first party titles)
and, here's some idle speculation:
* Nintendo might be really bad at documenting their stuff, since third party developers seem to be having trouble using the special sauce. Or maybe they consider it a trade secret? I don't know.
JordanN
Banned
Wouldn't they have come upon these results already via internal testing? I think it was Namco-Bandai who said they got information like the clockspeed that way.
That's basically what all the speculation is about. There are a few tech heads trying to figure it out. Guys that have some understanding about this type of thing, like Blu, AlStrong, Wsippel, Thraktor, Fourth Storm, Durante... Blu for one has worked/developed on the GCN iirc.
ambientmystic
Member
Whats the point of hiding away all those vital statistics from the 3rd party developers and their other partners? Who could've thought that was a great idea in the 1st place?
It doesn't bode well if even those who're supposed to take full advantage of the hardware don't even have the complete picture of what it can or cannot do.
Can't blame those 'lazy' 3rd parties for their poor porting efforts now can you?
metalslimer
Member
Thanks guys this has all been very interesting to read and much better analysis than that shitty DF article. It is sad that 3rd parties apparently don't have full documentation of the Wii U themselves (as if anyone would want to steal the specs lol). Nintendo continues to be their own worst enemy in having a good relationship with 3rd party developers.
lwilliams3
Member
Considering how much stuff was leaked for Durango and PS4 (even elaborate details of the "data movers" for Durango), it does unfortunately seem like Nintendo was justified when it comes to actually securing their info.
But if nobody knows what stuff is on there, what point is there to having the stuff on there to begin with?
Call me naive but I always find it to be a pretty healthy rule of thumb to take the general negative sentiment and take it down at least a single notch. I'm not defending anyone here but it does seem like people enjoy taking an incomplete picture, piling on assumptions, and building an entire narrative out of it.
metalslimer
Member
Who cares though? Was Nintendo scared that they would look bad if their hardware was eventually compared to the other two? What good does it do them? For Sony and MS it would actually make sense to hide the stats because they are going to actually launch close to each other.
I can't remember exactly who it was, but didn't we get a leak from someone making a Wii U game that they still didn't know the full ins and outs of the system?
shinra-bansho
Member
Why? By the time these things leak plans are so far set in motion that changes are unlikely.
How is that worth handicapping developers and causing justified ire from devs/pubs.
lwilliams3
Member
I didn't say that it was a good thing for those reasons.
To be honest, though, I believe one reason why there is no documentation for it is due to the system still getting tweaked on with these customizations for effectiveness and how they work. Matt implied that there was features that was still locked when the Wii U launched, and Li Mu Bai from Beyond3D stated that the most mature SDK arrived in November. Nintendo's dev teams may even still be trying to figure some of these things out.
Schnozberry
Member
This chip is a custom design that has quite a few patents granted on it, some of which weren't even granted until well into 2012. They were likely protecting their proprietary design from being reverse engineered before the patents went through. Their NDA would suggest quite a bit of paranoia on the subject. I don't think it has much to do with it being portrayed as "weak", as they have stated repeatedly in the past their goal isn't to compete in terms of raw horsepower.
Oh, I'm not denying that that's probably the case. But there's enough about their relationships - or lack of a relationships - with 3rd parties that we don't know that I, personally, can't make the leap from "this seems very short sighted on nintendo's part" to stories of nintendo refusing to help anyone but themselves because they're nintendo and nintendo just loves being incompetent and jerks to everyone not called nintendo. It's always more fun to try to create stories from facts to try to make companies look better or worse.
In this case, for me personally, the situation is peculiar enough to warrant suspicion that there's likely more to the story than "nintendo sux." Would an explanation change the situation with 3rd parties? Well probably not but I don't know what their actual relationship is right now with their 3rd party partners so I'm not going to try to make something up.
Though on its face, it does look pretty bad. This is where you might say I'm being a bit naive. Personally, when it comes down to whether a company is simply incompetent or made a series of complicated miscalculations, I'll always choose the latter, though they're not always mutually exclusive. There's almost always some amount of reasoning behind a mistake, even if the reasoning turns out to be flawed or misdirected. And if you actually read what I'm saying, you'll see I'm not trying to argue nintendo = good here, I'm saying nintendo = bad but there's probably more to it than meets the eye.
The Abominable Snowman
Member
In the case of the 360 and PS3, leaked data benefitted consumers immensely.
We got bigger HDDs, more RAM and more general functionality that couldn't be tacked on at a later date (Like Xenos' unconfirmed bump in CPU speed)
So if that's the justification, it's still OK by me.
Were they afraid that 3rd parties would see the specs and run? Never mind making it harder for third parties that do support them. Everyone was expecting that Wii U wouldn't come close to matching the other consoles. Now they have little third party support, basic performance information is out anyway and a bunch of poorly running software.
The other console specs leaking is more damaging because they are direct competitors in a better postion to match specs.
lwilliams3
Member
Actually, several insiders and I believe Gearbox did say that Nintendo did accepted feedback from third-party developers, and actually made some changes.
I stated what I think was the issue eariler, but I believe Schnozberry has a good theory on what was going on.
The Abominable Snowman
Member
Nintendo's competing against themselves. 'leaked' data would affect nothing. Thats all I'm saying.
The Abominable Snowman
Member
What do you guys think of Esrever's analysis of the die size of the GPU
http://forum.beyond3d.com/showpost.php?p=1703471&postcount=4532
http://forum.beyond3d.com/showpost.php?p=1703534&postcount=4533
http://forum.beyond3d.com/showpost.php?p=1703471&postcount=4532
Wii U die to scale superimposed onto llano's
The SIMD blocks that makes up the CUs. LLano has 40 SPs per block. Wii U is unknown. At 40nm, the blocks in the Wii U are smaller than the Llano ones at 32nm. It is highly unlikely that the Wii U has 40 SPs per block.
http://forum.beyond3d.com/showpost.php?p=1703534&postcount=4533
The 3 die shots I used to get my data, they are scaled to be relative size. Wii U is 11.88mm x 12.33mm actual, 4870 is ~16mm x 16mm and Llano is 13.78 mm x 16.54mm. I could not find a useful die shot of bobcat.
Here is the scaled picture of the unit sizes. I use photoshop to calculate the sizes and areas are probably within a small margin of error but if my numbers for the die sizes are correct( I can't really be sure size I just used the internet to find them), there should not be much error.
If we assume perfect scaling from 55nm to 40nm. Then R770 on 40nm should be smaller than what is inside the Wii U. If we again assume perfect scaling, Llano would be about 2x the area of the Wii U's block at 40nm and 2.3x the size of the R770 at 55nm.
Wishmaster92
Member
Probably hiding too lol.
shinra-bansho
Member
That would imply less than 320 SPs?
Schnozberry
Member
Llano also contains a dual or quad core AMD CPU. Isn't it kind of a poor comparison to try and match up die size with?
The Abominable Snowman
Member
Yes, his die area comparison, and the SRAM bank count, would imply less.
It's a different way of stating that the SIMD units look really weird for 40nm, like a few pages back.
I'm not thoroughly convinced that the EDRAM isn't stacked in some sort of weird 2.5D/3D 40nm/55nm organization and I'm just going to err on the side of caution here. I did think the large area of EDRAM was a bit too clean in regards to the crisp area around it.
He analyzed the SIMD area moreso than the total area of the chip. That shouldnt have been too far changed
Why would Iherre need to hide?
Wishmaster92
Member
I was jk, we haven't seen him at all today that's why, hes probably busy.
Schnozberry
Member
Can you elaborate?
Zoramon089
Banned
But they look too large for 20...his comparison doesn't seem to...well, they just seem different
Nintendo would probably prefer developers to take a dev kit and see if their code runs on it, rather than have a developer look at the spec sheets and write it off. Nintendo makes efficient hardware, but that doesn't come across with numbers. Code that shouldn't work on such specs on other platforms might work... If that IS the case the specs are irrelevant. Heck perhaps Nintendo can make a 12gb/s bus not be a bottleneck, but how would you convey that to people who would just like to judge a system based on a spreadsheet of numbers ( like most on this and similar forums...) those types would write off the system before reading the fine print regarding optimizations and such,
What's there for Nintendo to document when most developers know more about the basic hardware than Nintendo themselves? 
From: Nintendo
To: All current and aspiring Wii U developers
"Our console supports fully programmable shaders. Any documentation on how these damn things work would greatly be appreciated." - NCL
I kid! Seriously though, as much as people are down on Nintendo for not providing adequate documentation, that doesn't automatically mean devs weren't supported. Ok, so developers weren't sent a bunch of sh!t to read, but on the other hand, at least a couple Wii U developers have gone on record stated how extremely helpful and forthcoming Nintendo's support team was.
Broken Rules:
From: Nintendo
To: All current and aspiring Wii U developers
"Our console supports fully programmable shaders. Any documentation on how these damn things work would greatly be appreciated." - NCL
I kid! Seriously though, as much as people are down on Nintendo for not providing adequate documentation, that doesn't automatically mean devs weren't supported. Ok, so developers weren't sent a bunch of sh!t to read, but on the other hand, at least a couple Wii U developers have gone on record stated how extremely helpful and forthcoming Nintendo's support team was.
Broken Rules:
He said he was going to try and stay out of these types of threads.
DreadPirateRoberts
Member
24 ALUs confirmed!
Really though, this whole chip is just odd, especially since it's supposed to be radeon based.
The extra stuff could be nothing but other chips and supporting structures unrelated to core performance. Which, until I hear or see otherwise I will assume is the case.
Reallink
Member
Actually yea, I'd say they were scared. They were scared of threads like this one comparing FLOPs to PS360 and Pentium 3's--declaring it a Wii situation all over again (or worse). Lot of their rhetoric has focused on U being a powerful "HD" console that could court "core" third-party experiences and wouldn't be left wanting come Orbis/Durango. This is negative PR, and if it were allowed to circulate the months leading up to launch, it would have trimmed a decent number of units off what they managed to sell with everyone in the dark.
FYI, the 360 CPU received a massive downgrade/speed reduction from its original documentation (ie the kind of stuff that's leaking to sites now - the 2004 docs I have sitting here in a folder somewhere), and developers were quite pissed going from the G5-based alpha kits to the near-final silicon which got resulted in a large performance dip. It essentially ended up being 3 of the Sony PPE instead of what MS wanted from IBM.
Also, welcome back bg!
The Abominable Snowman
Member
Basically what they seem to be going off of there are 2 things:
AMD never in any designs they have access to used a register bank larger than 2KB. There's no need to. The register banks on the Wii U die look too large to be 2KB@40nm. So either the fab is different or its larger than 2KB
In addition the simd logical area is also smaller than other units that have 40 SP per core at 40nm. That points to either an odd number or different fab.
And thats in addition to having only 32 register banks for over 32 SP units
Interesting. I seem to remember a thread saying it had a dual-core configuration@2.5GHz, back when the PS3 purportedly didnt have a dedicated GPU and used a beefier Cell.
I also remember MS bumping up the speed of their own GPU when Sony announced theirs was running at 550MHz or vice versa
lightchris
Member
2.5 GHz in case of a dual-core G5/PPC970 would have been faster than Xenon in most cases (but also quite power hungry afaik). The PPC970 was more in the league of Athlon 64 in terms of IPC.
lostinblue
Banned
All cases minus whetstones, really.
Quite power hungry and worse yet, quite hot; not suited for the form factor of a console; those debug PowerMac's were having problems being closed in boxes without breathing space.
efyu_lemonardo
May I have a cookie?
Just so I understand, say Nintendo decided to include custom silicon dedicated to lighting: I can see several reasons why that would make sense for them, but how would you go about approximating the number of FLOPs required to achieve similar lighting using traditional shaders? Could you possibly give a ballpark estimate based on the Zelda Demo, Bird Demo, NintendoLand etc.? I'm not trying to force a quote out of you and hold you to it, just trying to understand how these things can be estimated.
I've been looking at that picture as well, mostly wishing it was a bit biggerI forgot to add that this old site has a picture of GC's GPU die though small.
http://www.segatech.com/gamecube/overview/
freezamite
Banned
Since the clock is lower, the transistor density can be higher. They also don't have to suport double precision floating point operations, so they can cut more area here.The Abominable Snowman said:
The memory banks are much bigger than "normal" so 40ALUs per block seems completely plausible, or at least, much more plausible than just 20 of them.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
WiiU will be manufactured way beyond 2015.
This is negative PR only in the eyes of the people looking for negative nintendo PR.
A good deal of the engines and titles running on Orbis/Durango will be running on tablets and smartphones. WiiU has no technological issue there.
I don't really understand why you are posting this. AlStrong already responded to this on B3D, both him, Blu, even that B3D regular who was claiming 160 SPU at first, all say it's more than likely 320. And this guy compares it to a chip it is not meant to be compared with. There was another comparison with a 40nm chip, that almost matched to a T. Maybe this guy can use his comparison to disprove that chips process/ALU-count too.
Good enough for me, lol
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
Apropos, I just spotted this:
For the record: I have not developed for the cube. I've developed for a bunch of desktop and embedded GPUs across a range of vendors and performance brackets, but I've never developed for one of the established console brands. That said, I've studied cube's SDK out of curiosity and I do keep a Debian-clad Wii for the sole purpose of tinkering with some of the platform's features.
tkscz
Member
I thought he meant the AMD architecture. Ok, from now on, we call the GameCube GC. I know it's coded as GCN on the bottom of every gamecube (I have one, I can see it), but this is getting confusing. So AMD architecture is GCN, the gamecube is GC (or NGC).
I just saw that post and oh boy would it open up another can of worms if it turns out to be true. It would probably make the flops count mean much less than before as well.
The Abominable Snowman
Member
Excuse me? Because its a different take on what we have, and when cross posted here noone had responded to it. No other poster pixel-counted the areas. And which chip was that?
He annotated the die shot
Nevermind, not sure if youre serious.
- Status
Similar threads
- Poll
- Poll