Originally posted by: racolvin
I think its a great idea but until the PhysX processor is integrated onto the same board as the GPU, I won't be buying one. Given the amout of data that will need to be crunched for the physics to work as anticipated with this card, the PCI bus is just too slow. It belongs on the video card.
No. It doesn't belong on the videocard. Mechanics doesn't have anything locally to do with rendering. It would be silly to load the videocard's bus with stuff going in and out, that have nothing to do with the videocard. I'd assume it would mess up interfaces and drivers too.
It would also be silly to tie hardware physics processor and graphics processors together like that. As silly as putting audio on the videocard or the videocard inside the monitor.
We would like to choose and upgrade our videocards independantly from physics card.
You definitely want to keep video and physics separate. With separate interfaces and separate devicedrivers.
The Simulation/Game-world lives as a data model in ram. Software running on the CPU manages this model. This is where the CPU could request computed/recomputed data from the PPU. This results in changes to the data model. The data model is then viewed by the 3D-render engine. This too, lives in ram and is run on the CPU. The 3D-render engine generates, then sends D3D or OGL render commands to the videocard. Only then does the videocard enter the picture. The vidocard only draws the picture. Collisions, movements, whatever, is all inside the world data model in ram. And runs on the CPU.
As I hope you see, the PPU has no business being on the videocard, at all.
While the amount of processing that the PPU will have to do, may be large, this doesn't mean that the amount of communication to/from the PPU will necessarily be great. So we can probably do with PCI and PCI-eX1 to begin with.
Later, MB will maybe feature a bigger PCI-e for this.