You make some good points, but when you consider that many GAMECUBE games absolutely destroy most of the Wii games we've seen then you should realize there might be something wrong with the developers, and not the hardwdare.
WiiGeePeeYou (Hollywood) what IS it ?
- Thread starter Megadrive1988
- Start date
- Status
You make some good points, but when you consider that many GAMECUBE games absolutely destroy most of the Wii games we've seen then you should realize there might be something wrong with the developers, and not the hardwdare.
So your taking a PR rep's opinion 6 months before launch?
I guess by your logic we can believe all the PR spin Ken Kutaragi has proclaimed.
I don't think so but from what I saw it, it's a technique very good that change all the ambiance! Look like vertex shader illumination
The ps1.1 havent'got any jump instruction.As I remember the diference between the gf256 and the gf3 shader model was minimal,and mainly came from the additinal units in the gf3 (other cards was able to do the same thing without ps)
But though the GF2 they already include per pixel shading capabilities, don't they?
Rainbow Man
Both opinions are speculation based to varying degree's on facts. I've already explained to you why I think the 'facts' you've mentioned don't neccesarily follow to the conclusion you've come to (devs getting Wii hardware extremely late into development is a reasonable explanation). But I'm yet to see you even acknowledge, never mind address, a certain fact I've mentioned a couple of times now:
Flipper is 112mm^2 on a 180nm process, at 90nm the same chip should be 28mm^2 (roughly, no smaller and not much bigger). Fact Hollywood is 72mm^2 on a 90nm process. Now explain to me how your speculation fits those facts? What is taking up all that extra die space?, wishful thinking transistors?
Both opinions are speculation based to varying degree's on facts. I've already explained to you why I think the 'facts' you've mentioned don't neccesarily follow to the conclusion you've come to (devs getting Wii hardware extremely late into development is a reasonable explanation). But I'm yet to see you even acknowledge, never mind address, a certain fact I've mentioned a couple of times now:
Flipper is 112mm^2 on a 180nm process, at 90nm the same chip should be 28mm^2 (roughly, no smaller and not much bigger). Fact Hollywood is 72mm^2 on a 90nm process. Now explain to me how your speculation fits those facts? What is taking up all that extra die space?, wishful thinking transistors?
Last edited by a moderator:
Rainbow Man
Veteran
I don't see how "sceptic until proven otherwise" can be turned into "speculation", but I guess there's always room for interpretation..
Thatf inal wiis may or may not have been delivered late to devs does not increasew probability of there existing extra features in said wiis that have yet gone untapped. You know this, or you should anyway.
Why should I? I've no special insight into how the chip works so anything I say will automaticllay be useless speculation. All I'm saying is I'm not gonna go "aah HAA! Since the chip is X big, it must mean it has lots of hidden stuff in it!" like you do because there's no evidence to support that conclusion.
Maybe it wasn't shrunk as muhc as you think it was. Maybe there's an entirely different explanation.. Who can say?
What IS definitive fact is, nobody has said anything about any extra features of any kind so I'm not going to assume there are any until - like I've said many times now - I see them in action.. Anything else would be foolhardy in my opinion.
Peaxe.
You don't need special insight. You only need a modicum of technical understanding of chip hardware, gleaned from this forum one would hope, and a bit of logic.
Logical point : You don't put transistors into a chip unless they're going to do something.
Technical point : A process of 90nm producing Flipper would produce a die of ~ 28mm^2 (taking Teasy's numbers)
Fact : Hollywood is 72mm^2
Logical point : Hollywood has more transistors than Flipper
Logical point : There's transistors are there to do something!
The evidence is in the chip size and the knowledge about processes. What those transistors do is unknown, but their presence is a sure-fire guarentee that Hollywood is more than just a shrunk Flipper.
If we start to speculate about the internals of the hollywood, we can se what can not be in the gpu:
-It haven't got additional memory for the frame buffer/texture buffer (480p)
-The ati can not modify the pixel pieline.It is an integrated part of the memory system,and the memory size is tha same as was in the flipper(if they modifythe pixel pipeline,they can re-design the mem subsystem too)
The remaining area is the tnl unit,and the memory interface.Possibly area is the on chip memory for the vertex data , and a programable vertex shader.Maybe a tesselator?I don't think.The tesselators are not too populars.
For me only a re-designed tnl unit make any sense.(easyer programm development,more control above the caracter pipeline and so on)
-It haven't got additional memory for the frame buffer/texture buffer (480p)
-The ati can not modify the pixel pieline.It is an integrated part of the memory system,and the memory size is tha same as was in the flipper(if they modifythe pixel pipeline,they can re-design the mem subsystem too)
The remaining area is the tnl unit,and the memory interface.Possibly area is the on chip memory for the vertex data , and a programable vertex shader.Maybe a tesselator?I don't think.The tesselators are not too populars.
For me only a re-designed tnl unit make any sense.(easyer programm development,more control above the caracter pipeline and so on)
Technical point: Transistors consume power adding heat to the overall heat output.
Logical point: You don't add heat producing transistors if all they did was produce heat and nothing elese!
Technical point: Larger dies cost more money to fab.
Logical point: You don't waste die area with garbage unless you like to throw money away.
We do not know this. It could have a larger framebuffer eDRAM that's fully BC with the old 3MB of eDRAM.
Rainbow man, I thought it was pretty clear that the last paragraph in my post was the fact I based my opinion on. Nothing to do with my opinion on developers not getting ample time with final Wii hardware. Which was simply a counter argument to your point about the quality of visuals in Wii launch games.
I'd have thought it would be obviously why you should try to explain away a fact that completely contradicts your opinion. As Shifty says you don't need to be technically gifted to see the basic logic at work in my argument. I've given you my reasons for not believing your 'facts' as any sort of conclusive proof that Hollywood is simply the same as Flipper. The facts I've given you are at least as solid if not more so, but all you seem to do is dismiss them. Basically what you seem to be saying is that because you aren't capable of discussing the facts I've brought up those facts don't count and are in fact just wishful thinking or delusion...
I'd have thought it would be obviously why you should try to explain away a fact that completely contradicts your opinion. As Shifty says you don't need to be technically gifted to see the basic logic at work in my argument. I've given you my reasons for not believing your 'facts' as any sort of conclusive proof that Hollywood is simply the same as Flipper. The facts I've given you are at least as solid if not more so, but all you seem to do is dismiss them. Basically what you seem to be saying is that because you aren't capable of discussing the facts I've brought up those facts don't count and are in fact just wishful thinking or delusion...
Last edited by a moderator:
Ok.Issue:
-resolution limited to 480p
-strict restrictions in the flipper due the hardware (indexed textures position and so on) . That mean you have to spen irrealsiticly big amount of tie to be backward compatible.
-Nobady used till now all of the possibility of the filpper,due the memory and skill limitations
-The texture buffer and the frame buffer is a part of the pixel pipeline.This is not connectid with a memory controller,thses are effectivly high sped caches.
-If they don't touch the size of the memory any change on the pixel pipeline is simply not make sense.
The botleneck in the case of the wii can be the floating processor power.So,with a good VS unit you can re-ballance it.
Oh,and an other point:this solution is the cost optimal.Possibly in the case of GC nobady used the character pipelin that was made by the nintendo.(maybe except the nintendo) .
So,the target can be to improve it.
(the design philosophy of the WII have to be the "improve the weakest point")
Actually a few said that it is upgraded (eg ATI and Retro, said it is a new architeture), lke others member of this forum, IMO I think that even one or two devs here gave such a hint.
Would it need more, ie it is a botleneck in GC?, because if not then you have two choises if you want HD you will need a lot more power just to get equal level gfx (not what Nintendo wants), if not it could only be usefull to do AA but it will also use a lot of filrate or needs a new AA method. IMO they would made better if they spend transistors in others things.
Why cant they have made a major redesign, it isnt because lack of time as they worked for a long time in it
?And the dsp, anyway they could add a lot of things that could make sense, from HW to physics to just offload the CPU, to brand new or the most desirable (hardwired or not) fxs, to free AA, to floating point processor that could be used to whatever devs want... Any of those should be able to some significant improvements on the games that use them.
The wii need probably an 8 bit stencil buffer,but that mean 64 bit word with,and that need 3 megs of memory.
The FSAA need at least 3 times more memory (3 sampl/pixel).So , 9 megs.
The new VS mean physics .The AA isn ot so possible without huge fillrate and big memory,as you can see it on the pc and on the xb2 an ps3.
The biggest disadvantage of the GC is the raw calculating power,the mem size and the memory bandwith.
Which is a long winded way of saying...
Logical point : You don't put transistors into a chip unless they're going to do something.
Edit : With the exception of redundancy for yields.
*plays off-topic warning jingle*
The GF3 pipeline has two "register combiners" chained together in each pixel pipeline. To the programmer though it looks as if there were eight, because he can configure a chain of eight (abstracted with "insructions" in Direct3D, or more directly in OpenGL) and the result will be as expected. This is accomplished by looping data from the end of the pipeline back to the beginning of the pipeline and putting in extra work cycles with the remaining "instructions". I was talking about that kind of loop, nothing to do with a jump instruction
The GF2 can't do that. It has the same ALU hardware but because there is no loop-back, you can only "program" for two register combiners, not eight.
Of course each succeeding generation was extended significantly, so just giving the word "shading" a nod isn't meant to imply that the thing could somehow compete with a Geforce 8.
It doesn't take a jump instruction to have a loop mechanism in the hardware.
The GF3 pipeline has two "register combiners" chained together in each pixel pipeline. To the programmer though it looks as if there were eight, because he can configure a chain of eight (abstracted with "insructions" in Direct3D, or more directly in OpenGL) and the result will be as expected. This is accomplished by looping data from the end of the pipeline back to the beginning of the pipeline and putting in extra work cycles with the remaining "instructions". I was talking about that kind of loop, nothing to do with a jump instruction
The GF2 can't do that. It has the same ALU hardware but because there is no loop-back, you can only "program" for two register combiners, not eight.
In a way, yes. It's limited to very short "programs" though (3~4 "instructions") and doesn't support dependent reads (i.e. computing new texture coordinates per pixel, which is required for most advanced bump mapping and lighting techniques). It won't blow your socks of if you're used to current PC graphics cards, but I think it was davepermen who squeezed per-pixel specular highlights out of it.
Of course each succeeding generation was extended significantly, so just giving the word "shading" a nod isn't meant to imply that the thing could somehow compete with a Geforce 8.
Not sure what you mean but you're a little off, so let's calculate the stuff out for once.
Isomorphic 16:9 480p is, err ... let's say at most 854 pixels wide. 409920 pixels in a frame. Tops.
Code:
Bytes per pixel | memory needed
3 1.23MB
4 1.64MB
6 2.46MB
7 ??? 2.87MB
8 3.28MBEight is what we would expect from a PC card, it would allow the Wii to get rid of the RGBA6666 mode, and the color banding with it, and bump up stencil precision without compromising Z precision.
pc999,
Multisampling AA is not a fillrate problem, it's for all intents and purposes free as long as you have the bandwidth and the space.
IMO, if the improve yields, they are doing somethingWhich is a long winded way of saying...
Logical point : You don't put transistors into a chip unless they're going to do something.
Edit : With the exception of redundancy for yields.
- Status
