Figure 12
From: High performance MRI simulations of motion on multi-GPU systems
Speedup observed with increasing number of available GPU cards on a multi-node, multi-GPU computer system. Performance is normalized to the single GPU board system, which is the reference standard for all comparisons. An almost linear dependence is observed. (Kernel execution time: 16 min for 1×GPU, 8 min for 2×GPUs, 5.36 min for 3×GPUs, 4 min for 4×GPUs, 3.22 min for 5×GPUs, 2.76 min for 6×GPUs).