Setting GPU when training on Zorro

Posted By: marr

Setting GPU when training on Zorro - 08/13/19 15:38

Hi guys i have a deep learning script in R which GPU is enabled when training, it took only few minutes to complete the whole dataset to train. When i run it with Zorro it seems to use the CPU since it takes hours to finish up training. Is there any way to switch to GPU when training on Zorro?
Posted By: AndrewAMD

Re: Setting GPU when training on Zorro - 08/13/19 20:40

Zorro supports GPU training via R.

Your setup needs to be correct. Either you failed to get your GPU to work, or you created a bottleneck. But it is impossible to identify the culprit with what little context you have provided.
Posted By: marr

Re: Setting GPU when training on Zorro - 08/14/19 05:24

i am using conda env on my R to load the tensorflow gpu, and my the script is similar to zorro example. my gpu is 1051ti 4gb, cpu is i3-4170 3.7GHz
cpu bottle neck is possible culprit, my goal is to make zorro training run as fast as i run on R console using gpu

the code that i use to load conda env

Code
use_condaenv("r-dl", required=T)
Posted By: AndrewAMD

Re: Setting GPU when training on Zorro - 08/14/19 11:40

I haven't used Tensorflow directly (I've used Keras->Tensorflow), but I believe there is a way to verify which processing unit you are using. Confirm your processing unit, both via R and then via Zorro->R.
Posted By: marr

Re: Setting GPU when training on Zorro - 08/14/19 12:11

I mean I am using Keras with Tensorflow GPU and able to confirm that on R terminal, I don't know how to verify from Zorro->R
Posted By: AndrewAMD

Re: Setting GPU when training on Zorro - 08/14/19 12:52

How about the same way you did in the R terminal?
Posted By: marr

Re: Setting GPU when training on Zorro - 08/15/19 06:18

the way i did in R terminal is when I run the neura.test() it has a info message the it detects the gpu and using it and fast when training. but the process is slow when i click Train on zorro panel and theres no message that it use gpu, only shows cpu the number of cpu cores in use
© 2024 lite-C Forums