IMAGE: 1000 x 1000 grid and 1,000,000,000 rays

COMPILE AND RUN

• SERIAL
  • make serial
  • ./serial <grid_size> <num_of_rays>
• CUDA
  • make cuda
  • ./cuda <grid_size> <num_of_rays> <num_threads_per_block>

OPTIMAL SETTINGS ON THE GPU

• GRID_D: 1,000 N_RAYS: 1,000,000,000
• BLOCKS: 65535 THREADS PER BLOCK: 32
• time elapsed: 1.235815(s)
• BLOCKS: 65535 THREADS PER BLOCK: 64
• time elapsed: 1.228932(s)
• BLOCKS: 65535 THREADS PER BLOCK: 128
• time elapsed: 1.202047(s)
• BLOCKS: 65535 THREADS PER BLOCK: 256
• time elapsed: 1.201033(s)

The optimal setting was with 256 threads per block

TIME COMPARISON as a function of NUMBER OF RAYS

#Rays	TIME (s) serial	TIME (s) CUDA
10,000	0.004454	0.000092
100,000	0.043513	0.0001999
1,000,000	0.419113	0.001409
10,000,000	4.183725	0.013441
100,000,000	41.865421	0.134746
1,000,000,000	420.276577	1.200679