Efficient Intra-SM Slicing for GPU Multiprogramming
October 19, 2016
Track: ACM SRC
Thanks to massive execution resources, GPUs have become one of the key execution engines in many high performance platforms. However, not all applications can consume all available resources. A profiling-based dynamic intra-SM slicing strategy is developed to better utilize the resources, which improves performance by 23% over the baseline multiprogramming approach.