Copying data from GPU to another GPU

In older versions of CUDA, was necessary to use main memory to copy data from one graphic card to another. In 4.0 we can do it without it - just one copying operation.
Is this way possible in OpenCL?

Is this way possible in OpenCL?

Such details are hidden from the application in OpenCL. Memory transfers between devices in the same context will happen automatically for you. You simply need to create a context with two devices.