This can be great for parallelism, as tasks specific to certain areas of memory may be relegated to the specific CPUs that can access that block of memory. This will allow for increased optimization and less competition for memory, because there may be less need for message passing.
However, as soon as CPUs need memory from blocks of memory owned by other CPUs, there will be increased message passing, higher latency and synchronization will be slow, so this could be worse for parallelism.
This can be great for parallelism, as tasks specific to certain areas of memory may be relegated to the specific CPUs that can access that block of memory. This will allow for increased optimization and less competition for memory, because there may be less need for message passing.
However, as soon as CPUs need memory from blocks of memory owned by other CPUs, there will be increased message passing, higher latency and synchronization will be slow, so this could be worse for parallelism.