Skip to main content
Figure 13 | Journal of Cardiovascular Magnetic Resonance

Figure 13

From: High performance MRI simulations of motion on multi-GPU systems

Figure 13

Simulation of kernel embedded and non-kernel embedded motion models. The blue line shows execution time of the spmd statement for simulating a non-kernel embedded motion model with an increasing number of GPUs on a single-node computer system. The red line shows the execution times for simulating the same motion model using a kernel embedded implementation with an increasing number of GPUs. While non-kernel embedded models take longer to execute, they allow for using non-analytically described motion patterns. (Total number of tissue isochromats 2000000, total number of time-steps 640000. Non-kernel embedded execution times: 851.4 min for 1×GPU, 405.5 min for 2×GPUs, 327.5 min for 3×GPUs, 248.3 min for 4×GPUs, Kernel embedded execution times: 16.4 min for 1×GPU, 8.3 min for 2×GPUs, 5.6 min for 3×GPUs, 4.3 min for 4×GPUs). All experiments were run on a 4×C2075 system.

Back to article page