Does Dataiku support GPU?

heloo

I wonder if we can use Dataiku's GPU for Deep learning models because my computer does not have.

Alex_Combessie

Hi,

You can use GPUs in Dataiku: https://doc.dataiku.com/dss/latest/machine-learning/deep-learning/runtime-gpu.html

However, you need to have GPUs on your computer or server for Dataiku to use them. If that's not the case, you can rent a managed GPU cloud instance and install DSS on it.

Cheers,

Alex

heloo

Thank a lot for your answer.

tgb417

@heloo
,

Welcome to the Dataiku community. And thank you for this question.

@Alex_Combessie
,

Thank You for your response, I don’t regularly need to the use of a GPU, and my local computer only has built in graphics. I’m running a DSS design node on this local computer. Is there a way to build a container from within DSS that can be setup to run somewhere in the cloud with a GPU? Rather than having an ongoing expense of a cloud node with a GPU?

Alex_Combessie

Hi @tgb417
,

The simplest option is to rent a server in the cloud with GPU, with DSS installed, which you can easily turn on/off for the time you need it.

Else, a more involved setup could be to hook up your local DSS to a Kubernetes cluster with GPU-enabled pods. We support various Kubernetes offerings with GPU images:

- Amazon Elastic Kubernetes Service: https://doc.dataiku.com/dss/latest/containers/eks/unmanaged.html#using-gpus

- Microsoft Azure Kubernetes Service: https://doc.dataiku.com/dss/latest/containers/aks/unmanaged.html#using-gpus

- Google Kubernetes Engine: https://doc.dataiku.com/dss/latest/containers/gke/managed.html#build-a-cuda-enabled-base-image

Yet another option (which is also involved) would be to set up a remote docker daemon on a GPU server: https://doc.dataiku.com/dss/latest/containers/docker.html

Hope it helps,

Alex

-