LBL-EESA / TECA

TECA, theToolkit for Extreme Climate Analysis, contains a collection of climate anlysis algorithms targetted at extreme event detection and analysis.
Other
57 stars 21 forks source link

run deeplab ar detector on cori GPU nodes #348

Closed burlen closed 1 year ago

burlen commented 4 years ago
  1. verify the deeplab detector works on the cori gpu nodes.
  2. verify that it is faster than when run on the cpu
  3. document options for using gpu in rtd docs(see issue #349)
burlen commented 4 years ago

@taobrienlbl who should take this on? I think it would be either you or Abdel - probably Abdel?

taobrienlbl commented 4 years ago

@ndkeen - do you have an account on the Cori GPU cluster, or do you think you could get one easily? And would this be of interest? We're looking to examine the performance of the new TECA Machine Learning AR detector on GPU nodes vs CPU nodes.

@elbashandy or I can provide a test problem.

elbashandy commented 4 years ago

Yes thanks for bringing this up @burlen. Can we use TECA's current profiling component in this evaulation?

burlen commented 4 years ago

Yes thanks for bringing this up @burlen. Can we use TECA's current profiling component in this evaulation?

@elbashandy One could do that. There are some reasons we might hold of on such low level profiling at first (better discussed over zoom). To start we could get a quick and dirty validation that things are working as expected. The bash time command would probably do the trick. Or if you run in a batch script slurm will time it for you. Batch scripts also document the test and make it easy to reproduce.

burlen commented 3 years ago

534 gets the deeplab ar detector app working with CUDA.