Skip to main content
. Author manuscript; available in PMC: 2011 Jul 25.
Published in final edited form as: J Parallel Distrib Comput. 2008 Oct;68(10):1307–1318. doi: 10.1016/j.jpdc.2008.05.013

Fig. 8. Performance of advanced MRI reconstruction.

Fig. 8

(a) The reconstruction time includes the time to compute FHd and the time to run 60 iterations of the linear solver. The number at the bottom of each bar is the reconstruction time in minutes. (b) Performance of FHd computation. The first six configurations (CPU.DP-GPU.ConstMem) compute the trigonometric functions in software, using approximately 13 and 12 FLOPS for the sin and cos operations, respectively. The remaining configurations compute the trigonometric operations in hardware; therefore, each sin or cos accounts for a single FLOP.