Analysis of new architecture fusing CPU and GPU
Yongxin Zhu · Journal of Computer Applications · 2011
When CPU and GPU work in coordination as separate chips,drawbacks such as high power consumption,large size,and low transmission speed can be observed.A merged chip would solve these issues.In this paper,the authors analyzed technical features of CPU and GPU to compare their computing capacity,and measure the performance by high-precision benchmarks.Then a new merger of CPU and GPU was described,utilizing low power processing unit for task allocation.According to the type of workload and input data size,the low power processing unit balanced the task partitions of the sequencing computing core and parallel computing cores array.For the different workload,the different cooperating modes were used to ensure efficient execution of data processing tasks on two types of cores.Performance assessment shows that the computation capacity and power consumption of the merged architecture are improved significantly.