Xbox 2 hardware overview leaked?
- Thread starter Bowie
- Start date
- Status
Megadrive1988
Veteran
so basicly Xenon graphics are from the R400, R500, R600 family, and not
from the R300, R420, R520 family.... is that what you mean Dave?
from the R300, R420, R520 family.... is that what you mean Dave?
More or less. My understanding at the moment on the PC front is that R520 will be their next generation SM3.0 part, but will use the R300 architectural platform as the basis. They will adopt the unified shader architecture for DirectX next in R600 and this will use what they are developing for Xenon as the basis but obviously with further developments in order to round off the DX Next specifications. Basically whatever features / functionality Xenon has will never completely come to the PC front, however the architectural platform of Xenon will be used as for DX Next (and potentially onwards) parts. I'd guess that if you used Xenon to the full you'd need a DX Next part to port it intact (graphically) across to the PC.
Development of the shader capabilities isn't really tht much of an issue, whats really being focussed on is the shader instruction scheduler - it has to get the best from a unified architecture you want as few bubbles in those ALU's as possible, so the more intelligent the scheduler is the more efficient the architecture will be.
Development of the shader capabilities isn't really tht much of an issue, whats really being focussed on is the shader instruction scheduler - it has to get the best from a unified architecture you want as few bubbles in those ALU's as possible, so the more intelligent the scheduler is the more efficient the architecture will be.
tngregoire
Newcomer
Can a few things be cleared up for a non-tech person like me? Fake or not, does this system utilize the vitual video memory features of DXNext? I found that to be a rather interesting feature based on the article written on this site. And can someone explain or provide real world examples of the features or beneficial affects of having the shaders being able to directly access memory, textures, ect. I mean, all this time, I just assumed GPUs were already capable of that. Also, when the system block was initially leaked, the biggest concern (coming from Democoder, I believe,) was that there wasn't enough E-Dram to support HDTV resolutions. With the statements made in this overview (720p native) is that concern still valid? Finally, there was a statement to the effect that the peak vertex rate of 500m/sec was attainable with non-trivial shaders. So, does that mean that we could see close to that high a number in game with a decent amount of effects?
10MB should be enough to support 1280x720p, you only need about 7.3MB for both color and z buffer at that resolution.tngregoire said:
Giving ALU direct access to memory make it easier to program and advanced programming features become possible: the length of shaders can be no longer limited, you can use as many register as you like, even creating array and object will be possible. I really hope the HLSL in DirectXNect can support pointer...
edit: the calculation of required framebuffer size is based on no AA situation. when using 4xAA, the number will grow up to 30MB??!! :?
Hellbinder
Banned
Right.. However.. The origional way it was put out was that Xbox2 Chip would be based off R600 and development was not seriously underway yet. At least thats the way i understood what was being said.DaveBaumann said:
Which is what i disagreed with. The RX5A Chip (or whatever its nomenclature is now) is a predicessor to R600 not a result of R600. Its been in Development for a quite a while.
Megadrive1988
Veteran
Xbox 2 graphics development has probably been underway since early to mid 2002 and will end sooner than Revolution's development, which means, Revolution could possibly have a better graphics processor unless Nintendo is going the extremely cheap route.
991060 said:
1: CPU builds display lists in L2 for GPU to consume; much faster than writing them out to main RAM for GPU to DMA back from, and also avoids burning main memory bandwidth on trivial transactions. AGP fastwrites was supposed to fill a similar function (allowing fast CPU access to the GPU), but that was only with relatively small transactions, thereby maybe keeping efficency low.
2: CPU does vertex shading, leaving all of the GPU to do pixel shading. GPU reads back finished transformed meshes and then does the actual drawing. This is an evolution of scenario 1.
Polarbear53
Newcomer
GwymWeepa said:
And how would it look compared to unreal 3, could this maybe do 12,000 polys a character or should we expect just unreal 3 level. How do you guys think these stats will change the way A.I. is?
Megadrive1988
Veteran
I'd hope Xenon could run a game like Unreal 3 with higher polygon count (30,000 ~ 50,000 poly models) and at 60fps. should be possible since good Xenon games will be optimized specifically for its hardware.
Megadrive1988
Veteran
somewhat like what PS2 does with EE handing the vertex shading / polygon & lighting computations, sending them to the GS which does the drawing.
Polarbear53
Newcomer
Megadrive1988 said:
Although that does seem rather high, if that did happen, I would take 20,000 in each model. So then it would be like the last 5 min in halo(without the drop in framerates). With all the things on screen at once, that was fun. Can we expect virtual displacemnt to be commonplace in next gen?
Megadrive1988 said:
When would you actually be close enough to anything to appreciate that kind of detail level?
Also, there's fewer times that are more annoying than characters more detailed than the world they lived in.
- Status
