Dask Dashboard with Local Clusters

Accessing the Dask dashboard for local clusters

Dask clusters on Saturn Cloud enable you to scale your workloads across a cluster of machines. There are scenarios when it is advantageous to run Dask on a single node, utilizing the CPUs as workers for the “cluster”. This is called a LocalCluster. If you have run Dask on your laptop and initialized a Client(), then you have run a LocalCluster!

Saturn Cloud has Jupyter instances with up to 4TB of RAM, so if that is enough for your workload, you can consider choosing one of those machines and running a LocalCluster. Another common workload is using multiple GPUs on the same instance with a LocalCUDACluster. In these cases, the Dask section of the resource page will not display your link to the Dask dashboard. This article explains how to access the Dask dashboard for local clusters.

Local Clusters

You can create a local cluster on a Jupyter server by leaving the arguments to the Dask Client blank:

from dask.distributed import Client
client = Client()

Dask is able to take advantage of multi-GPU acceleration using packages like RAPIDS and XGBoost. When using multiple GPUs across multiple instances, you would initialize a SaturnCluster. If you are using a single Jupyter server with multiple GPU cards on the same instance (such as a V100-8XLarge), you would initialize a LocalCUDACluster from dask-cuda:

from dask_cuda import LocalCUDACluster
from dask.distributed import Client

cluster = LocalCUDACluster()
client = Client(cluster)

Accessing Dask Dashboard

Previewing the client object displays information about the Dask client along with a link to the Dask Dashboard:



  • Scheduler: tcp://
  • Dashboard:


  • Workers: 2
  • Cores: 2
  • Memory: 15.50 GB

The “Dashboard” URL links to the localhost, but since this is running inside of Saturn Cloud, that link will not work. Saturn utilizes a Jupyter proxy to be able to access interfaces hosted on the server. You can copy the URL of the Jupyter window from your browser and replace /lab/* with /proxy/8787/status. For example, your Jupyter URL might be:


Then your dashboard URL would be:


If you find yourself doing this often, or if you want multiple local clusters in one Jupyter Server, you can utilize the following function to display a clickable link to the dashboard:

import os
from IPython.display import display, HTML

def local_dashboard_link():
    link = client.dashboard_link.replace(

        f'<a href="{link}" target="_blank" rel="noopener">{link}</a>'


If you find that you need to scale your workloads up more, then you will want to consider a full Dask cluster!

Need help, or have more questions? Contact us at:

We'll be happy to help you and answer your questions!