Figure 6From: High performance MRI simulations of motion on multi-GPU systemsNon-kernel embedded motion models algorithm. The object was divided in X parts and the pulse sequence in Y parts. Every kernel call was accompanied by a transfer of 1.28GB of data to the GPU global memory through the high-speed serial computer expansion bus (PCIe ×16 Gen2, speed rate of 8GBs/sec).Back to article page