PIConGPU
This is a relativistic particle-in-cell simulation of laser particle acceleration in an underdense hydrogen plasma.
We use our new tiling memory model to speed up particle in cell simulations on gpus.
Simulation data:
- 4 x 4 GPUs (4xTesla S1070 an 4 nodes)
- 1920x6016 cells, cell size 4.16e-8 m x 6.24e-8 m, 2 electron and 22 ion per cell (random position in cell)
- one iteration simulates 1.6e-17 s
- sliding window (enhanced simulation area)
- every 50th iteration dumped as visualization image for this video
- 100.001 simulated iterations
Summary:
- particles at start = 233 mill. electron + 233 mill. ion = 466.252.800 makro particles
- simulation run 3 h 44 min 28 s = 13468 s (inclusive image duming from 4 nodes to one shared SATA harddisk)
- we use a sliding window to enlarge the simulation area therefore we had 254.186.207 electron at the end of simulation
Legend:
- Green = Laser
- Red = density of electrons
- Blue = kinetic energy of the fastest makro particle in a cell (light blue is more energy)
More Informations:
- http://saahpc.ncsa.illinois.edu/papers/paper_10.pdf
is that realtime?
SSTriplePilipinas 3 months ago
@SSTriplePilipinas
No it is not realtime. There is too much data and therefore we are bandwith limited, theoretically we can get on this size 15fps if we read any value which is needed for the calculation only one time per iteration.
psychocoder 3 months ago
We generate PNG with any GPU-Host process and than we combine tham with convert and use mencoder to generate the video.
psychocoder 1 year ago
At page 10 is the oversimplify flow, link ( saahpc.ncsa.illinois.edu/presentations/day3/session2/presentation_Juckeland.pdf ). At the moment we only simulate particle-field-particle interactions
psychocoder 1 year ago