First of all I am doing this thread as a teorical one a mental exercice if you want, here I dont care if Rev is indeed based on this or not and if so how, take this as if I asked it about PS2 or XB.
Second I am not sure about anyhing I will write althought I think they right, or maybe is just to late now and I am writing a lot of foolish things, so be gentle.
Going to the topic, what I want to know is what can we do with the original GC chips to update them so they can offer much better AI, physics, gfx ... (that dont mean as good as XB) ( this could be interesting for Rev if it hadnt a big increase on the transistores counts, lets say about 2x as probably it could be a very good balance between price and performance).
I will post some things that I think they would make the bigest diference, and I suposse they can be done (however I dont know so I put that as questions), I guess that if they can do a GC2.0 that looks like this people would see a major improvement in games but not in price.
First of all it seems that the 3 bigger reasons why GC HW is so underutilized is:
1- Low market share (dont matter here).
2-No high level language to the TEV (that make it both hard to programe and costly).
3- Is good enought to do a lot of fxs but it is hard (performance/speed wise) to get them on actual games (eg ERP said that normal map is slow, althought possible, but only used in one game IIRC (SC3), not even in exclusives), mainly because of data formats and ALU limitations that make those fxs need multipass ones.
So can they improve the ALUs in order to them be able to do more and more complex ops per cicle (or even put more ALUs), being programable by high level languages so being more like a pixel shader. BTW If the ops could be done at least in 16bits (hp) instead of 8 could they have some form of HDR?
About the TnL engine what can be done with it?
Culd they just double (or more) the pipes and make it have higher efficience (less performance lost per light etc..)? Or any other thing just to get higher raw performance?
Could they improve their compression systems for less than 1/6 of the texture? What about the displacement maping?
Could they hardwire (integrate in the flipper TnL engine) much more complex shadows/lightning systems (eg D3 or SC3 like ones, or even much more complex ones like UE3 or whatever is better) would that be possible while being flexible enought to suport very diferent games from Q4 to FC or Zelda and MP by just making variations on the lights (like in todays GC)?
About advanced image qualitity functions, like AA in a single pass, what can be done?
In the gekko things should be probably harder to coment but there go a few changes.
Could they make it dual core (or +)?
Could they add VMXs units to it? (there as a rumor about a 750GX that goes up to 1,6Ghz + VMX and dual core but I dont know if that as for real but as I side note , if it is real why not put Broadway faster?)
Could they have some kind of specialized HW (for specialized HW I am thinking eg in a SPU, but I am not saying a SPU along side with Gekkko) for some heavy tasks (like path finding, animation, physics, networking...), or can it have even some fixed function (would that be good?) for that (eg like very smal PPUs/AIPUs/...)(or DSPs).
Or a smal VMX array, could that be done? and really usefull for in game (eg that tasks above exemplified).
What is you educated guess? Can we really see (if expanded) a big boost in GC architeture, that can still be cheap to produce, programe and keep full BC.
Personally if the above can be made then meybe we can see a nice boost, yet it would be at the expense of flexibility of dev creativity/control. It may be a nice solution if one is so limited as this. What do you think?
Some info on GC
http://www.segatech.com/gamecube/overview/
http://www.gamasutra.com/features/20021002/sauer_pfv.htm
http://www.anandtech.com/systems/showdoc.aspx?i=1566
http://www.beyond3d.com/forum/showthread.php?t=18282
http://www.beyond3d.com/forum/showthread.php?p=714395#post714395
Second I am not sure about anyhing I will write althought I think they right, or maybe is just to late now and I am writing a lot of foolish things, so be gentle.
Going to the topic, what I want to know is what can we do with the original GC chips to update them so they can offer much better AI, physics, gfx ... (that dont mean as good as XB) ( this could be interesting for Rev if it hadnt a big increase on the transistores counts, lets say about 2x as probably it could be a very good balance between price and performance).
I will post some things that I think they would make the bigest diference, and I suposse they can be done (however I dont know so I put that as questions), I guess that if they can do a GC2.0 that looks like this people would see a major improvement in games but not in price.
First of all it seems that the 3 bigger reasons why GC HW is so underutilized is:
1- Low market share (dont matter here).
2-No high level language to the TEV (that make it both hard to programe and costly).
3- Is good enought to do a lot of fxs but it is hard (performance/speed wise) to get them on actual games (eg ERP said that normal map is slow, althought possible, but only used in one game IIRC (SC3), not even in exclusives), mainly because of data formats and ALU limitations that make those fxs need multipass ones.
So can they improve the ALUs in order to them be able to do more and more complex ops per cicle (or even put more ALUs), being programable by high level languages so being more like a pixel shader. BTW If the ops could be done at least in 16bits (hp) instead of 8 could they have some form of HDR?
About the TnL engine what can be done with it?
Culd they just double (or more) the pipes and make it have higher efficience (less performance lost per light etc..)? Or any other thing just to get higher raw performance?
Could they improve their compression systems for less than 1/6 of the texture? What about the displacement maping?
Could they hardwire (integrate in the flipper TnL engine) much more complex shadows/lightning systems (eg D3 or SC3 like ones, or even much more complex ones like UE3 or whatever is better) would that be possible while being flexible enought to suport very diferent games from Q4 to FC or Zelda and MP by just making variations on the lights (like in todays GC)?
About advanced image qualitity functions, like AA in a single pass, what can be done?
In the gekko things should be probably harder to coment but there go a few changes.
Could they make it dual core (or +)?
Could they add VMXs units to it? (there as a rumor about a 750GX that goes up to 1,6Ghz + VMX and dual core but I dont know if that as for real but as I side note , if it is real why not put Broadway faster?)
Could they have some kind of specialized HW (for specialized HW I am thinking eg in a SPU, but I am not saying a SPU along side with Gekkko) for some heavy tasks (like path finding, animation, physics, networking...), or can it have even some fixed function (would that be good?) for that (eg like very smal PPUs/AIPUs/...)(or DSPs).
Or a smal VMX array, could that be done? and really usefull for in game (eg that tasks above exemplified).
What is you educated guess? Can we really see (if expanded) a big boost in GC architeture, that can still be cheap to produce, programe and keep full BC.
Personally if the above can be made then meybe we can see a nice boost, yet it would be at the expense of flexibility of dev creativity/control. It may be a nice solution if one is so limited as this. What do you think?
Some info on GC
http://www.segatech.com/gamecube/overview/
http://www.gamasutra.com/features/20021002/sauer_pfv.htm
http://www.anandtech.com/systems/showdoc.aspx?i=1566
http://www.beyond3d.com/forum/showthread.php?t=18282
http://www.beyond3d.com/forum/showthread.php?p=714395#post714395