Table 1.

Performance in PA image reconstruction (image size: $1024\times 512$).

Reconstruction method	CPU time	GPU time	Total time	Memory usage	Cores usage
CPU (without parallel)	29.5 s	—	29.5 s	8 MB from host memory	1 CPU core
CPU (with parallel)	3.3 s	—	3.3 s	96 MB from host memory	12 CPU cores
GPU (traditional BP) by rows	13 ms	352 ms	365 ms	256 MB from GPU	512 GPU cores
GPU (traditional BP) whole image	Cannot be realized due to the insufficiency in memory			Unavailable 256 GB from GPU	524,288 GPU cores
GPU (optimized BP) whole image	8 ms	50 ms	58 ms	2 GB from GPU	524,288 GPU cores