Skip to main content

Table 1 Kernel performance was assessed for different combinations of blocks and threads/block

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

Threads/Block 64 128
Blocks 14 28 56 112 224 448 14 28 56 112 224 448
Kernel time (sec) 154.6 82.3 47.0 33.5 33.4 37.2 87.1 51.6 32.9 32.8 36.6 43.8
Threads/Block 256 512
Blocks 14 28 56 112 224 448 14 28 56 112 224 448
Kernel time (sec) 64.0 33.4 33.5 37.4 44.7 59.1 34.8 34.5 38.9 46.3 61.5 60.1
  1. Kernel execution times were compared among different combinations of blocks and threads/block for the application of a pulse sequence of 296000 time steps on a 3D object of 500000 isochromats. The shortest kernel execution time has been achieved for 112 blocks and 128 threads per block (shown in bold letters).