Fig. 14From: Parallel acceleration of CPU and GPU range queries over large data setsAverage GPU speedup a) relative to the iterative CPU method and B) relative to the reduction CPU method vs. data set compression ratio (compressed size / uncompressed size, shown below each data set on the x-axis). The data for hybrid and ideal hybrid in (b) are indistinguishable at the scale of this figure. The Zipf data set skews are appended (e.g., Zipf0 has a skew of 0). The horizontal axis is logarithmically scaled. Both plots share the same legend and x-axisBack to article page