Hello,
I was wondering whether there is sort of a "CPU constraint" while running LAMMPS on GPU and in particular, does the number of CPU need be equal to the number of GPU to be as efficient as possible? I was thinking about optimal information passing whenever data actualization is required. Thanks in advance for insightful comments.
Regards,
T.T. Le.
Hello,
I was wondering whether there is sort of a "CPU constraint" while
running LAMMPS on GPU and in particular, does the number of CPU need be
equal to the number of GPU to be as efficient as possible? I was
thinking about optimal information passing whenever data actualization
is required. Thanks in advance for insightful comments.
there is no simple answer to this, because what is "optimal" depends a
lot on the kind of hardware you have (on both, CPU and GPU side) and
what kind of efficiency expectations you have (good capability or good
throughput). this kind of topic has been discussed in the past
multiple times. please take some time to dig through the mailing list
archives and look up the previous discussions. no need to repeat them
here.
axel.