The latest rumour (via IGN) has it using an R700 chip aka 4xxx. The high-end of those chips tops the 5770 range. I’m not too sure how much power the PC’s hardware abstraction steals from the GPU, but if it’s using a 4870-esque chip, this Project Cafe console is probably going to manage the PC-equivalent of a 5870 or 470 GTX. That ain’t too shabby. Today anyway. Late 2012 is another story.
The thing I saw said a ‘revamped R700’. If that’s accurate, looks like SenorBeef gets his wish. 
As you say, that could certainly top the 5770 in terms of raw power, but we would still be sadly lacking in hardware tessellation and a few other DirectX 11 goodies (depending, I suppose, on exactly what ‘revamped’ means in this context). Still, games designed with that card as their minimum spec (and 1080p as their maximum resolution) would certainly be capable of taking most people by surprise with how good they look, I’d think.
God knows what chicanery will be used to get video from the graphics card to the controllers’ screens. Maybe that’s where the revamping comes in.
I love the idea of the controller with a touchscreen. It would be awesome to play Zelda with all the items available or a map on the controller.
Well Nintendo wouldn’t be using DX 10 ro 11 anyway, as it’s an MS proprietary API.
Likely the games would start with some sort of openGl adaptation and 3-4 years later start writing more and more custom code at a lower level.
The R700 rumor doesn’t tell us the actual power of the chip.
I’m assuming they’ll use a smaller manufacturing process (because otherwise they’d need the equivalent of a large HD 4870 fan + heatsink to keep it cool, and a hefty PSU to keep it fed.
So we don’t really know what the final performance will be yet.
But unless it’s based on a budget design of the R700 (I doubt it) it will definitely be better than the other two consoles.
Doesn’t the HD 4870 run the PC version of Dead Space at like 100+ FPS at 1080p with 4x AA? While the other consoles struggle at 720p, lower rez textures, no AA and 25-30 fps?
That’s a significant jump for THEM, nevermind Nintendo.
Then you’re going to love Ocarina of Time 3D:
Yeah I know - I’m just using those terms as a stand-in for that hardware feature set. No matter how fast a Dx10 compatible card is, there are certain things it won’t be able to do, that even a slower Dx11 card could do - in terms of stuff that most people care about, this is pretty much just tessellation. In my opinion hardware tessellation is rather a big deal though - it would potentially be the defining fancy graphic thing of a generation, not unlike how normal maps were in the last generation. Geometry shaders, which were introduced in Dx10, allowed lots of cool new things, but don’t really seem to be a game changer (no pun intended) in terms of visual appeal.
Oh and, the 4870 was my most recent desktop card, and though I never played Dead Space it could happily churn out Left 4 Dead at 1920x1440x16 with everything maxed and it always remained playable (it was a bit smoother at 1920x1440x4 and I couldn’t see much difference - that’s what I usually played at). By comparison, getting a game to run on the 360 at 720p with any form of anti-aliasing is a serious challenge - observe the fact that Halo 3 ran at a lower resolution than the official Microsoft requirement (they got special permission). This is why the idea of 3D on the PS3 seems so crazy to me - it’s hard enough getting a game to run at HD on those consoles without doubling the frame rate requirement. So anyway, yeah, a 4870 would perhaps not quite be a traditional generation-sized jump over the 360 and the PS3, but it would definitely be a noticeable arse-kicking.
Also, for some reason what you just said about using a smaller manufacturing process just game me a hilarious mental image of a Wii with the side cut open and one of those giant enthusiast ATI cards (like the 5970) sticking out the side. Maybe with a separate plug going to it. I’d buy that. 
I don’t see what giant magical revolution tessellation is, isn’t it just tiling? I mean, I guess it could make for more complex wallpapers and floors. Maybe even have some slightly better textures, but it’s fundamentally based on a repeating pattern, which games already do to conserve memory, making it more efficient wouldn’t really be a big jump.
What am I missing?
Graphics tessellation is when the video card auto-magically converts the polygons into even smaller polygons so it looks a lot better and smoother. It’s like getting a 300,000 poly model for the price of a 30,000 poly model. (I totally made these numbers up.)
It’s kind of a big subject, but basically it’s what Palooka said - it means the hardware can magically turn a low-polygon model into a high-polygon model. For anyone who wants to know more though, I present the following long-arse description. 
It can magically turn a low-polygon model into a high-polygon model, but it doesn’t just take a triangle and make it into four triangles that look the same as the original one though - it performs automatic smoothing. I don’t know what algorithm they use but one common mesh smoothing algorithm is called Catmull-Clarke - see the picture in the wikipedia page for it here to get an idea of what it looks like.
Okay so already, that’s kind of cool. No more dubiously hexagonal gun barrels in first person shooters, right? But it gets better. Now that we have all that vertex density, we can apply a displacement map. You know what heightmapped terrain looks like? If not: A heightmap is essentially a big greyscale image, and the whiteness of each pixel represents the height at that point on the map. So you can use an image to create, say, a mountain range. Cool stuff - people have been doing this for ages. Displacement mapping uses the same technique on regular geometry. Take a 5,000 poly mesh, smooth it using the tessellation unit so that it’s an 80,000 poly mesh, then apply a greyscale texture to it and suddenly you have a pretty good replica of a very high-density mesh. To see what this can look like, cop the surface detail in this picture. That’s a screenshot from the Unigine Engine demo - you can check it out on youtube to see it in motion (or download the demo, if you think your graphics card is 'ard enough).
Anyway, at this point you might be thinking “Woo, shitloads of polygons make stuff look better. Big surprise. But we still have to actually render the polygons, so why can’t we just beef up polygon counts on models in older games?” Good question. There are two reasons why hardware tessellation is way cooler than this.
-
Rendering lots of polygons isn’t the problem. It’s actually storing lots of polygons. Take some dude from Unreal Tournament. He has about 5,000 triangles. Let’s hand-wave that and pretend it equates to about 5,000 vertices.
Even if we make very optimistic assumptions about how much space is required to store a vertex, that model sitting in video memory (as it must do, to be rendered with any kind of speed) takes up a fair bit of space. In my experience, finding the memory to put all your models and textures is as big a problem as finding the time to render everything in the frame. And in terms of time, the problem is often fill rate or some other thing - not vertex transform time. Obviously as you crank up polygon count eventually you’ll start taking too long to transform and shade the vertices, but you’ll run out of VRAM first. However, tessellation sidesteps this issue. You can store the low-resolution version of a model in memory, and tessellate it (afresh, each frame - sounds crazy, I know) for rendering. Suddenly you can crank up the poly count and take full advantage of the vertex processing, without taking a hit to memory usage. Awesome. -
Automatic LODding. LOD stands for ‘level of detail’. For generations (heh), graphics programmers and game artists have been using several versions of a model for different distances. You have a piece of scenery, say, and you store a high-poly version, a mid-poly version, and a low-poly version. You cunningly swap in the lower poly version when the player is far away and can’t tell anyway, they are none the wiser, and you save the vertex processing cost (I know I said before that vertex processing isn’t a bottle neck - it would be if not for techniques like this though). This turns out to be a really worthwhile saving because when you think about it, if the player is walking through say a field of sunflowers, only a tiny percentage of the flowers are close to the player - the bulk of the visible flowers are far away. However, with tessellation, you can very smoothly and automatically skip/achieve the same thing - the hardware decides how many polys to render a model at, not the artist, so there’s no ‘pop’ when you cross the LOD threshold and the models swap. So that screenshot from Unigine I posted above? It doesn’t look like that from a mile away - that really would choke the vertex processing ability of the GPU - but it doesn’t matter, because you can’t see that much detail from a mile away anyway. So all this annoying crud that the programmers, artists, and CPU used to be managing (LODs) will be done automatically, and much more nicely, by the GPU’s tessellation unit. Do you know what mipmapping is? Because (broadly speaking), tessellation does for geometry what mipmaps and trilinear filtering did for texture detail.
One of the coolest things about tessellation is that it needn’t change the artists’ workflow at all. They already work by creating a high and a low poly mesh for normal map baking purposes - you’d just need to generate a height map as well as a normal map (sometimes this is done anyway as a middle step to create AO maps and stuff, and either way, it’s something the computer does for you). So games ported from older platforms would benefit a great deal from tessellation hardware even without the developer doing the porting having to create new art assets and stuff.
TL;DR: Watch this crazy shit, paying particular attention to the detail on the roof tiles, brickwork, etc. In my opinion it’s a bit overdone but hey, it’s a tech demo.
Disclaimer: I haven’t worked directly with tessellation. My understanding of how it works in principal is sound, I believe, but without actual development experience I can’t be certain that the above description is how things would play out in reality. Still, broadly speaking, to my knowledge it is a correct portrayal of why tessellation is worthwhile.
That’s a pretty good description of tesselation. I just wanted to back it up by saying… you see the effects of LOD all the time, and it’s one of the immersion-breaking things about video games. You know when you’re walking along a road, and you can see clear gradients in the quality of the dirt and rocks or cobblestones or what have you that make up that road? That’s the level of detail thing - everything within 15 feet of you might look really smooth and detailed and clear, and then you’ll have another tier of 15-30 feet that looks good bit but more blurry, and then a tier for 30+ feet that looks blurred and crappy. Or the immediate area near you will have fully modelled individual grass, but after so many feet it’s abstracted blurry grass, and then after that no grass at all. It’s very immersion breaking because you can see clear breaks in the gradients based on distance no matter which way you turn.
Same applies to everything - models of cars, buildings, any objects. At some point you’ll get X distance away from it, and suddenly more details pop into view. Maybe the car suddenly has side mirrors and little finishing details that you didn’t see before. It’s jarring. All of that happens when the game decides you’re close enough to tell the difference between a low detail and high detail model and the high detail model pops in.
But because it swaps in something all at once, that might be 10 times more detailed, the effect can be very obvious, and really jumps out at you.
With tesselation, that doesn’t happen. All the graphical detail is there at every stage, but because of distance, it won’t render the full detail because there’s simply not enough angular resolution to see the differences - but this is the same way vision works in real life where you can’t resolve as much detail on a distant object as a close one - but the model gradually gains detail as it becomes closer to the viewer. But just like approaching an object in real life, it does it in such a natural and smooth way that you never see a higher model object popping in - you just gradually, naturally see more detail.
Not only does this mean less work for the artists and programmers (instead of making 3 or 4 versions of an object, the modeller only has to make one high quality one and the computer does the rest, nor does the programmer have to decide when to swap out the level of detail models in a way that’s the least distracting), but it actually makes the whole thing run faster. The computer doesn’t have to store multiple models in memory, the program never has to figure out when to swap a model’s full detail for 3/4ths detail, etc. There’s a dedicated piece of hardware handling it.
So not only is it less work for the artist/programmer which makes it easier/cheaper to produce, but it’s actually a higher quality, more immersive experience for the player, and at the same time it’s actually faster and easier for the computer to do, meaning you can make more detailed worlds. So… cheaper, faster, AND better - great, right? Except since we’re fucking locked into 2004 hardware for a fucking decade, we can’t use it. There are actually a lot of different advances similar to this, where things were actually easier/cheaper/better/faster, but they came a generation after the current consoles and we can’t use them. We’re stuck with harder to make/more expensive/lower quality experience/slower for 10+ fucking years.
The current generation of consoles actually came at the worst possible time, right before we switched to dx10 and had a clean start and had so many things that made developer’s lives easier while getting better results for the user. We miss out on an entire decade of a golden age revolution in graphics and physics and world size and detail because of their shitty timing.
nevermind
Heh yeah, I was going to put a little caveat about that but I forgot - my remark about LODding happening without the player knowing is uh… a little bit optimistic. 
One other thing I didn’t mention, but follows on from SenorBeef’s comments about tessellation simplifying art pipelines and whatnot, is that it potentially (not 100% sure if this is how things will go) eliminates the need for lots of crazy shenanigans in the pixel shader. I don’t know what the performance payoff is using a tessellated and displaced surface versus some crazy bullshit* like relaxed cone step relief mapping, but I’m pretty sure that if all the pixel shader techniques that are used to hack together the appearance of different types of geometric detail suddenly were replaced by one thing that the graphics API pretty much does for you, code bases will become a lot simpler and that’s a pretty big plus in itself.
Yeah, I can’t really disagree there. sigh It’s another thread, but, Microsoft not bringing DirectX 10 to Windows XP combined with Vista’s lack of popularity sure didn’t help the situation. The consoles plus that pretty much killed us for a good while there.
Still, the current situation where the general public doesn’t really know what modern GPUs are capable of puts Nintendo (or whoever) in the position of being able to release a console with a midrange card in it and blow everybody’s mind with its next-gen awesomeness.
- No disrespect intended by authors/practitioners of said crazy bullshit - I take my hat off to you. But seriously, that bullshit be crazy.
Nintendo’s finally confirmed Project Care, or whatever the fuck it’s going to be called, will be at this year’s E3, and will be playable!
And it’s launching in 2012. Clearly this system marks not just the end of the Wii, but all mankind.
Nintendo announces Wii successor for 2012
Nintendo has confirmed that it will be released in 2012. So, it is real.
Now, this isn’t an update to the Wii, is it? It’s a whole new console, I think. At least, I assume they are making a brand new console, not just an upgrade, right?
I wonder what they’ll call it. Super Wii?
They’ll call it the Dolphin Revolution, because some names are too good to waste.
(Everyone who answered my tessellation question, thanks. I was getting caught up in the literal 7th grade math meaning of tessellation which I think hurt my understanding a bit.)
The Super Dolphin Revolution!
I would totally buy that system just for the name.
The Super Virtual Dolphin Revolution 64!!!
The super virtual dolphin revolution happy fun time 64 3D!!!
I was just thinking “Wii2” 
Super Wii isn’t bad I guess. It’s hard to match the name “Wii”, it’s memorable and easy to say.
Will casuals buy the system? Will the hard-core?