The textures of two separate GPUs are physically in different memory chips. One GPU cannot access memory allocated on an another (except by copying the data via system memory).
Some Crossfire/SLI implementations could implement resource synchronization but the connection does not have nearly the same bandwidth as a PCI-X bus.
OpenGL drivers do software synchronization to make this scenario possible. In D3D, you can manually copy the data to multiple physical textures. Note that readbacks from GPU to system memory are usually relatively slow.
Show differencesHistory of post edits
#2Nik02
Posted 09 April 2012 - 03:08 AM
The textures of two separate GPUs are physically in different memory chips. One GPU cannot access memory allocated on an another (except by copying the data via system memory).
Some Crossfire/SLI implementations could implement resource synchronization but the connection does not have nearly the same bandwidth as a PCI-X bus.
OpenGL drivers do software synchronization to make this scenario possible.
Some Crossfire/SLI implementations could implement resource synchronization but the connection does not have nearly the same bandwidth as a PCI-X bus.
OpenGL drivers do software synchronization to make this scenario possible.
#1Nik02
Posted 09 April 2012 - 03:07 AM
The textures of two separate GPUs are physically in different memory chips. One GPU cannot access memory allocated on an another (except by copying the data).
Some Crossfire/SLI implementations could implement resource synchronization but the connection does not have nearly the same bandwidth as a PCI-X bus.
OpenGL drivers do software synchronization to make this scenario possible.
Some Crossfire/SLI implementations could implement resource synchronization but the connection does not have nearly the same bandwidth as a PCI-X bus.
OpenGL drivers do software synchronization to make this scenario possible.