A new paper from ORNL’s Sparsh Mittal and Jeffrey Vetter seeks to change the mindset of researchers using GPUs. Entitled, “A Survey of CPU-GPU Heterogeneous Computing Techniques,” the paper contends that merely offloading computational tasks to GPUs is not optimal, instead, using both CPU and GPU can lead to potentially higher speedup.
According to Mittal, the document reviews nearly 200 papers and has now been accepted in ACM Computing Surveys 2015.
Researchers are now moving away from the “CPU versus GPU” debate towards a “CPU-GPU collaborative computing” paradigm where the unique features of both can be intelligently combined to achieve optimal performance. A ACM Computing Surveys 2015 paper by Sparsh Mittal and Jeffrey Vetter provides a survey of CPU-GPU heterogeneous computing techniques. This paper discusses motivation behind such a shift and then reviews several papers which provide heterogeneous computing techniques for both fused CPU-GPU chips and systems with discrete GPUs.