What i am trying is to measure the CPU global memory bandwidth with SVM. The AMD benchmark "GlobalBandwidth" offered such function for GPU. However my CPU doesn't support SVM, so the data for CPU can't be collected. So i created a context , including both GPU & CPU, and commandQueue for each of them. Then i called clSVMalloc to create a SVM object as argument for the kernel. Finally, i launched kernel with both commandQueue. Though i run into some faults, can i archive my goal?