Yes, but hopefully they go with an APU style chip like Xbox 360's Slim, that should allow it to hit even higher wattage, as you would have more room to cool the single chip, instead of having 2 heaters to cool. Which would also make sense with them not having parts cool enough to run yet, and having down clocked dev kits to hit the performance of what they can currently fit in a Wii U unit. (since they wouldn't be combo units yet)
Wii U Speculation Thread of Brains Beware: Wii U Re-Unveiling At E3 2012
- Thread starter bgassassin
- Start date
- Status
Nightbringer said:
I don't know if they will use exactly this kind of card as a base, but I am sure as hell that it will have something between 320 and 480 shader units @500-600 MHz (so between 320 and 580 GFLOPS)
You also have to keep in mind that most of Flippers/Hollywoods die size was eaten by the 1T-SRAM IIRC.
Nintendo seems to go with a proven process and should therefore have great yields, so die level integration might be possible. That also has some nice additional benefits the XCPU can't even use, like much lower latencies and enormous bandwidths. And it decreases the overall transistor count. With good yields and due to less complex cooling, it should be cheaper overall, while also offering a better performance.z0m3le said:
wsippel said:
Really hoping for this, it is sort of getting my hopes up, but I've been a gamer for a long time, and can deal with a little Nintendo disappointment every 5 years or so. As much as I love my 360, I will be Wii U only next gen, unless third parties decide to ignore Nintendo's console again.
THIRD PARTIES: Gamers don't buy boxes, they buy games. If you make em, they will come. - I bought my Xbox for Halo2, all my friends bought their Xbox 360's for Gears of War, I bought it for Blue Dragon and Mass Effect.
disap.ed said:
I don't understand how you can "know" this for a fact?
Most rumors point to R770, which I believe doesn't go below 640 shader units, the HD4850 for example has 800 shader units, hits 1000Gflops and is a chip @ 55nm. I think they are probably using a HIGHLY modified chip but it's performance target is probably between the HD4730 and HD4850.
disap.ed said:
Nice I never made the comparison, but that info is also not going to lead you to a solid case, as GPUs are built for PC's that don't worry about power requirements very much, they don't care about power usage below 100 watts.
Mobile GPUs do in fact worry about power usage, and the HD4000m series has 800 shaders, though I don't know the TDP of them, they can pack far more shaders into a 50watt TDP, the HD6900m series for instance has a chip at 50w TDP with 960 shaders, and the 5800m series goes up to 800 shaders inside 50watts, some as low as 35 watts with 800 shaders can be found.
Wii U will use a Modified GPU, that will most likely cut some corners for power consumption. AMD did the same thing for the 360, and Nintendo has a group working with AMD to hit just where Nintendo is targeting their power to be.
So just had a look at some newer notebook chips and there is in fact one card that would be quite nice, the Mobility Radeon 6830M:
800 Shader units / 40 texture units / 16 ROPS @ 575MHz => 24W ... Wow!
http://www.notebookcheck.net/AMD-Radeon-HD-6830M.43739.0.html
The 6870M is only clocked higher (@675 MHz) and already uses 50W, so clock rate will very likely be under 600W (Maybe even lower than XB360s 500 MHz, Hollywood's clock x 2 would be 486MHz)
So I have to apologize for my former post, it seems it would be quite possible to have 640 - 800 shader units. But we still have to keep in mind that we are speaking about Nintendo, so cost is always a matter and also that they most probably will again use embedded 1T-SRAM (if they want to implement Wii's 27MB of 1T-SRAM this will also be around 200 million transistors I guess)
800 Shader units / 40 texture units / 16 ROPS @ 575MHz => 24W ... Wow!
http://www.notebookcheck.net/AMD-Radeon-HD-6830M.43739.0.html
The 6870M is only clocked higher (@675 MHz) and already uses 50W, so clock rate will very likely be under 600W (Maybe even lower than XB360s 500 MHz, Hollywood's clock x 2 would be 486MHz)
So I have to apologize for my former post, it seems it would be quite possible to have 640 - 800 shader units. But we still have to keep in mind that we are speaking about Nintendo, so cost is always a matter and also that they most probably will again use embedded 1T-SRAM (if they want to implement Wii's 27MB of 1T-SRAM this will also be around 200 million transistors I guess)
disap.ed said:
No worries about apologizing, we are all getting a little nuts waiting for more info, that's why I have committed to only talking about Wii U in this one thread.
I'm thinking inside of an APU, they can make it a pretty big chip and cool it well, I wonder if you could share the embedded ram between the GPU and CPU in this case? is it possible? (I know the 360Slim doesn't do this, but that was more a compatibility issue right?) or would it be better to go with a llano type set up, where the CPU uses it's L2 and L3 cache, and the GPU uses it's own cache.
If they can share a large L3 cache say 32MB, then wouldn't that reduce size and TDP?
You have nothing to apologize for, since there's a major flaw with considering notebook chips. The 6970M is basically an underclocked desktop 6850. The 6850 costs about $160. However, the 6970M costs over $400. A big part of this is the fact that mobile parts are usually much higher binned than desktop parts. Higher binning means lower yields and higher prices. So, there's pretty much zero chance that Nintendo will use a notebook chip.disap.ed said:
That said, there's also a flaw in logic involving the desktop cards. That flaw, funny enough, is that they are cards. The TDP and measured power consumption applies no just to the GPU, but everything else on the card, most notably the memory. To make this even worse, TDP stands for "Thermal Design Power," so it's more related to heat than power consumption usually, but not all the time, and it's pretty much always inflated due to the fact that no two chips are ever exactly the same (which is of course why binning exists). In other words, this line of reasoning really doesn't get anywhere in the end.
Long story short: PC tech is fucking confusing.
BurntPork said:
I believe his point was that you couldn't fit a GPU with more than 480 shaders inside 50watt TDP, and has found out that AMD produces a GPU with 960 shaders @ 50w TDP.
Again, Wii U won't be using an off the shelf part, it will be a custom part, built around the R700 series GPUs, it is completely possible that they are working around getting the most power under a certain wattage with this card, and price of any GPU drops by up to half, when buying in large bulk... For reference the Wii's GPU cost about $40 a piece in bulk, Wii U's could in fact be just as pricey as a 6970m, though I doubt it would be based on a mobile part, but might borrow some power saving design choices.
bgassassin
Member
z0m3le said:
I'm adding this to the OP. There seemed to be a lot of praise about that control method so my interested has piqued.
z0m3le said:
According to the rumored specs given to wsippel, it sounds like the current dev kits have an underclocked 4830.
From EDGE magazine site:
EDGE: Did you make any assumptions about Nintendos new hardware before Wii U was announced? Did it tally?
Mark Rein: Oh, wow. A really sticky situation. We were in the enviable position of not having to make assumptions, lets just say that. Im pretty impressed with the Wii U. It looks like a great device and I think itll do really well.
EDGE: Did you make any assumptions about Nintendos new hardware before Wii U was announced? Did it tally?
Mark Rein: Oh, wow. A really sticky situation. We were in the enviable position of not having to make assumptions, lets just say that. Im pretty impressed with the Wii U. It looks like a great device and I think itll do really well.
AndyMoogle
Member
MS do NOT have that 4MB limit anymore. It's been gone for at least 8 months now. With that in mind, I wouldn't really believe anything Gustav Halling has to say.antonz said:
Zoramon089
Banned
McHuj said:
Both the 3DS and the GC had 3rd party input on hardware...why would you doubt Epic, probably the largest middleware provider in the industry today, had some level of input into the development of Nintendo's new system?
It has nothing to do with that. I'm just saying that yields are the reason Nintendo isn't using 28nm for the GPU despite the fact that TSMC has already started producing them and it would be a HUGE benefit.z0m3le said:
I don't get why Rein is so excited, while that guy from DICE seems to pretty much hate it. Perhaps it's simply that DICE wants to stay PC-only and push graphics to the ultimate limit, but EA won't let them?Maxrunner said:
BurntPork said:
COULD, (could, mind you) be the reason hardware isn't ready yet, this console releases next year and 28nm is coming out with the HD7000 series this year, so it will be produced in mass quantities by the end of this year.
BurntPork said:
The DICE guy just wants more ram, he is a PC dev, and BF3 was built from the ground up, for PC and is being ported to the consoles. He still says that Wii U will be the most powerful console BY FAR, his capitals not mine.
I think Wii U has over 1GB memory, but possibly less than 2GBs, if it were 2, they wouldn't have any need to worry about PS360s successors, but Nintendo likes to limit devs... I have my fingers crossed, but if we get 1.5gb I will be happy.
I think it might simply be because the Wii U falls drastically short of cutting edge PCs. That's usually not the case for new consoles outside of the original Wii, correct? That's likely what devs like DICE want in the next gen of consoles (though it's highly unlikely they'll get it)
I doubt it. Chances are that they'd still have a lot of trouble making enough, and it would end up being more expensive anyway.z0m3le said:
Possible. Actually, after that Crytek guy wanting 8GB, I think he may even be expecting 4.z0m3le said:
Again, possible, but could he really have been hoping for a high-end card? I'm sure that devs have some common sense.guek said:
It makes even less sense considering that Epic said that they wouldn't support next-gen systems if they can't run Samaritan. I am so confused.
guek said:
Yes, I agree with this post, most people expect it, but they don't realize the power draw of current top end GPUs, the fastest in the world right now is the 580GTX, but that is a 250Watt card, and even at 20nm (won't happen till late 2014 or 2015) You'll only cut that power down by 40%, even Nvidia said that something like Epic's newest showcase won't be possible on a single GPU for a few years.
Close platforms have their advantage, but that advantage quickly goes away as tech pushes forward. For instance, in 2007 GPU's could outperform consoles released in 2005 and 2006 on windows platforms.
Sorry I'm sometimes dense and try not too use too much induced information to parse posts. I read what is there and try to make minimal assumptions, because these can be false. As I said before I not necessarily arguing the points, but much more the presentation.
1. (underlined above by itself) Your conclusions have to be convincing though. That depends on the evidence that you supply and the knowledge of your audience. It's what would differentiate a weak induction from a strong one. IMO the evidence was not strong enough for the conclusion.
2. (bold above) A new build of what? The rendering engine? sure fine I'll give you that. But for what I consider a game engine that would make up maybe 20% of it. I wouldn't call 20% constructing something new. Now again, remember, I'm not arguing that they won't build a new engine. Something that has changed with the new hardware that is going to be much more fundamental is the level of parallelism exposed. Multiple CPU threads and lots of CPU cores (depending on how much these are exposed to general compute/stuff that isn't synthesizing an image). Effectively making use of parallelism is a much, much bigger beast than the evolution of shader model from 0 to 3. Why? 1. Because in the case of shaders they can still mostly fit into the concept of geometry transformation and fragment shading, which you have even with model 0. So you are not changing the fundamental operations and the pipelining concept your engine might be based on could still be perfectly relevant: just update it a bit to add binding of the programmable parts. 2. a paradigm shift from having mostly sequential execution to algorithms that most be conceived to expose more parallelism to be effective is currently a big issue with software development. It requires a different way of thinking about things, and producing such algorithms is much harder. Sorry I don't have time to find specific resources for that (talks at AMD's Fusion 11 conference might be good -- C++ AMP -- and you can always start with (wikipedia).
Sorry I fail to see how the following sentences support the task being "hard". If you're saying you have to do something as opposed to nothing to get it working, then I wouldn't call that "hard". Also, particularly when it come to assets, I can say for a fact that that is a gigantuan amount of work that goes into building content pipelines. These depend on two things: 1. the producer and 2. the consumer. In our case 1. the producer is a various set of tools like Maya for models/animations, Audio composition software and what have you. These software suites have a much more long-lived development cycle and typically change incrementally. 2. The consumer is the game engine that needs to be able to understand the assets and process them properly. Although the refresh cycle of these engines might be more dramatic than the producer software, what assets are and how they are described usually doesn't. Also no matter what your rendering special effects are, a mesh is a mesh so to say, and it's description is fairly stable. 90% of a game (if not more) is assets and optimizing the asset management, making sure you get the most out of your artists is important. All of this is to support the following proposition: developers are likely to separate out how an asset is managed from other components of the engine, such as the rendering engine. And that is to support: they are likely to reuse, without loss of functionality or "power" their asset pipeline. So finally, I would rather say that the task of using "old assets" with a "new engine" is likely not "hard".[/QUOTE]bgassassin said:
And you are also missing my point (most likely because I don't express it properly enough). I'm not saying the capabilities of UE2 vs UE3 aren't different and that running a game on one vs the other is not going to "gimp" it. I was talking about how fundamentally different Engine6 is as opposed to Engine5 (using UE as an illustration, not that I'm going to spend half a day googling and contrasting their internals): say 90% of the Engine6 uses the same code as Engine5 will you call it "new"? The most obvious changes from a users point of view are the graphics, but that is only one component of the engine. The point is that you don't know what the capability of Nintendo's engine is, nor how much needs to change for it to completely exploit the capabilities of the new hardware.bgassassin said:
Thanks, that's all I wanted to see, is something to support claims. I wouldn't say the evidence is convincing to me, but that is a different story. At least with these I can put your other comments into proper context.bgassassin said:
Yes.bgassassin said:
Shit, I think how anal I am stems from having to deal with too many academic papers that are trying to pass BS by you.bgassassin said:
1. (underlined above by itself) Your conclusions have to be convincing though. That depends on the evidence that you supply and the knowledge of your audience. It's what would differentiate a weak induction from a strong one. IMO the evidence was not strong enough for the conclusion.
2. (bold above) A new build of what? The rendering engine? sure fine I'll give you that. But for what I consider a game engine that would make up maybe 20% of it. I wouldn't call 20% constructing something new. Now again, remember, I'm not arguing that they won't build a new engine. Something that has changed with the new hardware that is going to be much more fundamental is the level of parallelism exposed. Multiple CPU threads and lots of CPU cores (depending on how much these are exposed to general compute/stuff that isn't synthesizing an image). Effectively making use of parallelism is a much, much bigger beast than the evolution of shader model from 0 to 3. Why? 1. Because in the case of shaders they can still mostly fit into the concept of geometry transformation and fragment shading, which you have even with model 0. So you are not changing the fundamental operations and the pipelining concept your engine might be based on could still be perfectly relevant: just update it a bit to add binding of the programmable parts. 2. a paradigm shift from having mostly sequential execution to algorithms that most be conceived to expose more parallelism to be effective is currently a big issue with software development. It requires a different way of thinking about things, and producing such algorithms is much harder. Sorry I don't have time to find specific resources for that (talks at AMD's Fusion 11 conference might be good -- C++ AMP -- and you can always start with (wikipedia).
AndyMoogle
Member
I guess we won't see Epic on consoles until next-next gen then.BurntPork said:
But seriously, they'll optimize and remove features to make it run on next gen consoles. Simple as that.
mckmas8808
Mckmaster uses MasterCard to buy Slave drives
Luckyman said:
Why do people say this? What is Apple doing that makes people think this?
Grampa Simpson
Banned
I'd even take it a step further. He isn't planning on porting it over to the Wii U.NateDrake said:
I read that he's taking a "wait and see" approach with it. He seems more excited about it as a gamer than than as a developer.NateDrake said:
Also, I'm not saying that devs don't want to work on it; they just don't seem to be as excited about it as you'd expect them to be with a new console. It seems more like "Hey! Another console to port stuff to!" than "Finally! A powerful new console that takes away some of the annoying limits of the current-gen!"
BurntPork said:
ugh
Gabe Newell said:
Really, you need to stop with the negative nancy bullshit, because there are plenty of positive reactions. Yes, there are not so positive or more tepid reactions as well, but consider this:
Peter Moore said:
Peter fucking Moore didn't know very well how powerful the thing was at E3. Coupled with the fact that hardware is not yet finalized and how much actual information some devs are privy to when making comments, positive statements are a lot more telling than negative ones because if anything, the hardware is not going to get worse as new dev kits are shipped out.
Zoramon089
Banned
Junior members...
- Status