Home | News | Documentation | Download

Running TMVA with multi core

Hi!

I’m trying to run TMVA in multi-core mode. It seems only DNN provides the multi-core option.

Is there anyway to run in multi-core for other methods?
i.e. PDERS

p.s. I checked the root multicore page : https://root.cern.ch/how/how-express-parallelism-many-cores
I used the command
int nthreads = 7;
ROOT::EnableImplicitMT(nthreads);
But it seems when it run with TMVA it still used one core.

Thanks.

Best,
Justine

@swunsch @kialbert could you please have a look?

Hi,

Indeed only the DNN and and BDT (with gradient boosting) currently support multi-threading.

Cheers,
Kim

Hi,
I am trying to use the DNN example in ROOT 6.22 with multi-threading(https://root.cern.ch/doc/v622/TMVAClassification_8C.html). But it shows only 1 core is used.

-----------Start of deep neural network training on CPU using MT, nthreads = 1

I also try to set “ROOT::EnableImplicitMT(7);”. It does not help.

Best,
Jie

Hi,
I am trying to use the DNN example in ROOT 6.22 with multi-threading(https://root.cern.ch/doc/v622/TMVAClassification_8C.html). But it shows only 1 core is used.

-----------Start of deep neural network training on CPU using MT, nthreads = 1

I also try to set “ROOT::EnableImplicitMT(7);”. It does not help.

Best,
Jie