I am running the memory access speed test on APU. I find that GPU access the memory(no matter host buffer or device buffer) at a faster speed than CPU. GPU can access memory at 15-20GB/s, but CPU can only reach about 5GB/s.
If I use the GPU and CPU access the memory at the same time, the total speed will be reduced, which is between the GPU and CPU access speed. Is it normal?
If I use the HSA feature, can CPU and GPU access the memory at the same fast speed? and if I use CPU and GPU access the memory together, can I get a faster hybrid speed than GPU only does?