The "massive leap" will come more from speed than the shaders IMO. There are alot of things that would be possible with shaders 2.0 (real-time radiosity, for example), except for the fact that the hardware just isn't fast enough. Once you get to the point where you can run basically any arbitrary code in the shaders, you're only limited by the speed of the hardware.
XBox 2 Dev Kits
- Thread starter Dave Baumann
- Start date
Yes, and Shader 4.0 is pretty much at the point of any arbitary code, PS2.0 is quite far away .
DaveBaumann said:
Not true, Shader 4.0 is likely to still have limited write capabilities (fixed outputs), thats a major difference (and limitation) from a CPU which can equal read/write abilites.
Shaders have as a purpose to produce a limited set of output attributes, while the input streams and memory access (textures) will become largely unlimited and the computation ability limited purely by speed, the output is still restricted to those attributes.
Clever tricks will increase the output but this will never be (until shaders are just CPU's) there forte (because its this restriction that gains the speed). Some restrictions to input will still there for a similar reason (its unlikely for you to be able to get the pixel to you left due to synchronisation issues).
So theres something to talk about for a few more years yet
DeanoC said:
Yes, but I'm talking comparatively to PS2.
So theres something to talk about for a few more years yet
OT from here, but I'm beginning to think Shader 4 will probably represent the peak of interest in the development of 3D chips. From here I doubt we'll see radical changes in architecture - I believe that we'll only see incremental difference in the capabilities of the chips and a lot of development on the instruction schedulers.
DaveBaumann said:
Err PS2 is too overloaded a term, do you mean PS_2_0 (pixel shader) or Playstation2 vertex units? (or even some really old IBM computer ;-) )
Obviously much more flexible then PS_2_0 but not as flexible as Playstation2 Vertex Units, as they have a general output model (because they are just weird CPU's). Of course Playstation2 VU's weren't designed for pixel shading and suck pretty badly at this job (though do a pretty good job on square textures...)
DaveBaumann said:
Yep unless they decide to drop the stream architecture (seems unlike but may happen), in which they become CPU's and even thats a pretty incremental change...
Mind you rendering will eventually have to undergo a revolution from local to global illumanation, that likely to involve a massive change in shaders... We mai have to move to a physically correct rendering method one day, thats likely to cause a flurry of interesting hardware.
But that may be for the grand kids to worry about running on there personal quantum computers
Megadrive1988
Veteran
that sounds facinating. cannot wait for this sort of rendering to arrive.
I suppose this upcoming generation of consoles, as well as DX10 / R500 / NV50 won't see that shift to global illumanation & massive change & a physically correct rendering method.
perhaps by 2010-2011 with the next-next generation consoles such as PS4-N6-XB3 and DX12 ?
From my faint memory on PPC, there was a part that has x86 compatibility but I am not sure if that part was released at all. I had read about that part but it was already so many years. It was something like 645/615 or something like that.
Wonder if it has anything to do with XB2 IF it should employ the PPC architecture.
Wonder if it has anything to do with XB2 IF it should employ the PPC architecture.
zidane1strife
Banned
I'd like to see specs for the console on the order of 50+ times the performance of the current top of d-line gpu... What think?
DeathKnight
Regular
M$ bought the rights to Virtual PC which is an x86 wrapper for PPC.
DeathKnight said:
And judging by VPCs performance, emulating the XBox CPU shouldn't be a problem, generally. Nor is achieving industrial strength stability an issue. Apple emulated 680X0 processors on the PPCs after they made the shift in processor family and IBM does this pretty much all the time in their business iron. VirtualPC is remarkably stable particularly considering that it emulates the whole crufty platform and not only the CPU.
Efficient emulation always use a technique where it saves recently translated code snippets for re-use in order not to have to decode the instruction stream to be emulated on the fly all the time. There are variations on this differing in how much you save, where you put it, if you optimize the translated code, and so on but this is the basic premise. The problem is with non-iterative code, but it is generally assumed that non-iterative code parts execute so fast anyway that they contribute negligeably to the overall execution time for intensive use. This is typically a reasonable assumption, but you could definitely construct cases where a given emulator runs into efficiency problems.
Even so, a 2005 period PowerPC designed for the task should have little to no problems emulating the xBox x86 CPU. Particularly considering that the xBox2 will have several times more memory available than required for xBox code (making efficient emulation a whole lot easier as you can basically save a full translation of the game code in memory).
zidane1strife said:I'd like to see specs for the console on the order of 50+ times the performance of the current top of d-line gpu... What think?
I think that would be a terribly unrealistic expectation, considering it would go into a consumer product which can't sell for 50+ x the price of today's top of d-line GPUs.
I think it would be a terribly unrealistic expectation to hope for a GPU even 5x faster. It wouldn't NEED to be 5x faster anyway since it's going to run at TV resolutions anyway. All it needs in the way of fillrate is 1Gpix or so sustainable, nobody will ever need more in a videogame. You're not going to fill your screen 1000+ times per second anyway.
Still, I don't think next-gen console GPUs will have "only" 1Gpix of fillrate, they'll have loads more, that's for sure. Not 50x more than today's top of the line parts though, that's if not physically, so at least economically impossible at the moment and in the foreseeable future.
Panajev2001a
Veteran
Fill-rate is not what worries me, shaders' execution speed is.
A 9700 PRO could run full instruction length PS2.0 shaders on every pixel to about the same rate as a Voodoo 2 can just rendering textured pixels. I'd estimate that they will be about 4-5X the 9700 PRO shader rates by 2005/6.
zidane1strife
Banned
Well, at least it should see that increase in vertex performance. Consoles have usually increased by 100-300x in that area for some time, the current top of d-line gpu hovers around 3x in that area... thus 50x increase seems reasonable, No?
Megadrive1988
Veteran
as everyone and his dog knows, Sony has been saying that PS3 will be a 1000x more powerful than PS2. they've said that since 1999, before the PS2 came out. Sony also said PS2 is 200~300x as powerful as PS1. I think 215x was an actual figure. obviously Sony is not going by one specific spec (i.e. FP performance) but overall rendering performance.
I have never seen Microsoft say Xbox2 will be 1000x more powerful than Xbox. or any other figure.
I think a reasonable increase would be 10~20x Xbox's graphics performance, for Xbox2. In comparing NV2A to R300, we have a 2-3x leap. if R420 is 2x faster than R300, and R500 is 2x faster than R420, and if R500X (for Xbox2) is 2x faster than the normal R500 for PC, plus some extra efficiency, we might get a 10~20x performance leap from Xbox to Xbox2. a 50x leap is probably unlikely. bandwidth will go from 6.4 GB/sec to probably 50~60 GB/sec.
now, putting speculations aside, all I want out of Xbox 2 is to have IN-GAME graphics that are just as good and just as smooth as the pre-rendered CGI that MS showed for Xbox1's announcement at GDC 2000. (the Robot demo, and the Afro Thunder demo)
I have never seen Microsoft say Xbox2 will be 1000x more powerful than Xbox. or any other figure.
I think a reasonable increase would be 10~20x Xbox's graphics performance, for Xbox2. In comparing NV2A to R300, we have a 2-3x leap. if R420 is 2x faster than R300, and R500 is 2x faster than R420, and if R500X (for Xbox2) is 2x faster than the normal R500 for PC, plus some extra efficiency, we might get a 10~20x performance leap from Xbox to Xbox2. a 50x leap is probably unlikely. bandwidth will go from 6.4 GB/sec to probably 50~60 GB/sec.
now, putting speculations aside, all I want out of Xbox 2 is to have IN-GAME graphics that are just as good and just as smooth as the pre-rendered CGI that MS showed for Xbox1's announcement at GDC 2000. (the Robot demo, and the Afro Thunder demo)
As Dave Baumann sort of implied, all these discussions about "X times higher performance" really make no sense unless the devices are asked to do the same job. They aren't however, so comparing this generation consoles to the next that way is pretty meaningless.
The proof of the various puddings will be in the eating. Initially, I'd assume the extra horsepower will be used to take that extra step that developers wanted, but couldn't, on this generation hardware. That will most probably be the case for the projects that cross generations. It will presumably take some time before there are games out that have been conceptualized with the new hardware in mind and been programmed from the ground up with intimate knowledge about how to wring the most out of the next generation architectures.
Not that the voice of reason ever had much effect on the speculatively inclined. It's a fun way to pass time, but even speculation can be more or less meaningful.
The proof of the various puddings will be in the eating. Initially, I'd assume the extra horsepower will be used to take that extra step that developers wanted, but couldn't, on this generation hardware. That will most probably be the case for the projects that cross generations. It will presumably take some time before there are games out that have been conceptualized with the new hardware in mind and been programmed from the ground up with intimate knowledge about how to wring the most out of the next generation architectures.
Not that the voice of reason ever had much effect on the speculatively inclined.
It's a fun way to pass time, but even speculation can be more or less meaningful.Similar threads
