Do GPU provide faster training times?

Andy3

Hello,

I have heard that Rapidminer have some similarities with H2O, so when I read the article (sefiks.com/2019/11/07/why-you-should-build-xgboost-models-within-h2o/) that the H2O library have some turbo-button when the GPU is enable I hope the same thing was applied in Rapidminer though I haven't been able to test this out. My impression has been that only the CPU has been the most important thing when trying to decrease training times with the GBT-algorithm. Could someone from RM verify that RM has the same turbo-button or not?

Best regards

Andy

tkenez

Hi @Andy3,

We are actually using the open source H2O library as a backend for some of our learners (GBT, GLM, LR and DL - not the one in the Deep Learning Extension, but the built-in one). In our upcoming 9.7 release we actually did some work to bump up the used H2O library to the latest stable version and implemented some improvements on these learners.

I can only confirm what @mschmitz wrote, currently we don't support GPUs for the above learners. However, we are actively looking into how we can leverage the library better, which includes adding additional learners, and enabling GPU support where it makes sense.

I cannot commit to a release date yet, but this is an active topic in our engineering team.

Regards,

Tamas

MartinLiebig

Hi @Andy3,

we have GPU support for neural networks in the Deep Learning extension, but not (yet) for GBTs.

Best,

Martin

Andy3

Hope to have it soon in for the GBTs.

Andy3

Glad to hear that you do some work to update RM to latest stable version. Love to hear more about it