The tool offers accelerations of more than 200 times using a single GPU compared to a single CPU core: 1 GPU ≈ 200 CPU cores.
Execution times in logarithmic scale of a single CPU core and a single GPU (NVIDIA K80) fitting ball & sticks model in a whole dataset. The application fits the model in a low resolution dataset with 106,910 voxels and in a high resolution dataset with 684,203 voxels. Different number of diffusion measurements (36/72/108/145/181/218/254/291) and number of fibres (1/2/3 with 6/9/12 parameters) are showed.
Speedup
Measurements | 36 | 72 | 108 | 145 | 181 | 218 | 254 | 291 |
---|---|---|---|---|---|---|---|---|
Low-res 1 fibre | 96× | 145× | 167× | 175× | 183× | 195× | 202× | 196× |
Low-res 2 fibres | 98× | 149× | 174× | 180× | 193× | 206× | 217× | 207× |
Low-res 3 fibres | 92× | 144× | 172× | 195× | 210× | 225× | 235× | 220× |
High-res 1 fibre | 99× | 148× | 171× | 180× | 214× | 225× | 233× | 227× |
High-res 2 fibres | 109× | 168× | 196× | 203× | 215× | 232× | 238× | 230× |
High-res 3 fibres | 96× | 163× | 194× | 201× | 209× | 223× | 236× | 221× |
Hernández M, Guerrero GD, Cecilia JM, García JM, Inuggi A, Jbabdi S, Behrens TE, Sotiropoulos SN. Accelerating fibre orientation estimation from diffusion weighted magnetic resonance imaging using GPUs. PLoS One. 2013 Apr 29;8(4):e61892.