More DX Next and Longhorn release details
- Thread starter Richard
- Start date
Jawed
Legend
Well it's all speculation of course, but how are the virtual memory and unified shaders components of R600's architecture going to work if they're missing WGF2 support for DX7/8/9? That seems to me the kind of point that Geo is making (I'm guessing those aspects of the architecture are part of the issue).
If a GPU architecture is tuned to a new software architecture, but that software architecture is missing, I can imagine it causing problems. If the (WGF2) GPU architecture is dependent upon the WGF2 software architecture to implement DX9 or lower functionality then that sounds like a bit of a car crash if WGF2 isn't ready.
I'm not suggesting it is definitely a problem, I'm just joining in with Geo's speculation.
Jawed
If a GPU architecture is tuned to a new software architecture, but that software architecture is missing, I can imagine it causing problems. If the (WGF2) GPU architecture is dependent upon the WGF2 software architecture to implement DX9 or lower functionality then that sounds like a bit of a car crash if WGF2 isn't ready.
I'm not suggesting it is definitely a problem, I'm just joining in with Geo's speculation.
Jawed
This is just speculation, but I think the two API models will be completely split. AFAIK, the compatibility of Longhorn with the older Windows versions is more or less a virtual Windows XP machine. All the older stuff runs there, all the newer stuff can choose between that and the new API. And I think they will integrate the stuff that runs in the VM inside the desktop and start menu only.
_xxx_ said:
WDM was a smaller step then the step from the XP model to the LDDM.
LDDM moves many part of the driver from kernel to user space. But one of the main problems is support. If MS want to add LDDM to XP they need a service pack for this. After this each IHV need to have 2 driver for XP one for the old model and one for the new one. The next problem ist the DX runtime. If you have 2 driver models you will need 2 sets of the DX Runtime in the redist.
Jawed said:
There is allways a driver between the GPU and the API. If a Chip supports unified shaders the driver still can use them as SM3 pixel and vertexshaders. If it is make right you will get the loadbalencing benefit even with non WGF 2.0 software.
This is nothing new. A current GPU don't have fixed function cores for vertex and pixelprocessing any more but if you use a API that use this functions the driver will convert everthing that it can run on a shaderunit.
The only problem you will see with a WGF 2.0 chip without WGF 2.0 is that you are not able to use all features of this new chip. But it will work well with current software. You need to remeber that the longhorn windowmanager don't use WGF 2.0 but it have to work will with WGF 2.0 chips.
Demirug said:
But that was pretty straightforward since there are high-performance shader replacements for dedicated T&L hardware. What guarantee do we have that an architecture tailored towards WGF2.0 will be flexible enough to provide high performance using the older driver model?
trinibwoy said:
If you look at the know details about WGF 2.0 you will see that it contains the features of DX9. A good WGF2.0 chip will be a good DX9 at the same time.
Anyway the first WGF 2.0 chip need to be fast even if the run DX9 software because if they are not faster than current chips with this games nobody will buy them.
