This video is part of the HIP workshop playlist. View full playlist here:    • Advanced HIP Workshop  

Managed memory is another form of allocating memory that is accessible from both the host and the compute device. Managed memory leaves it to the runtime to manage or cache pages of memory on either the host or the compute device, and shuffle memory between host and device as required. Since pointers to managed memory are meaningful in both host and device memory spaces, just one pointer can be used for both modes of access. This mode of memory access will be slower than an allocation on the device in almost all cases, until we use a computing architecture where both CPU and GPU use the same pool of memory.