This topic describes how to configure CPU, GPU, and memory quotas
for users of an ML workspace.
Required Role: MLAdmin
Make sure you are assigned the MLAdmin role in CDP. Only users
with the MLAdmin role will be logged into ML workspaces with Site
Administrator privileges.
There are two types of quota: Default and
Custom. Default quotas apply to all users of the
workspace. Custom quotas apply to individual users in the workspace, and take
precedence over the default quota.
Log in to the CDP web interface.
Click ML Workspaces, then open the Workspace for which
you want to set quotas.
Click
Admin > Quotas.
Switch the Default Quotas toggle to ON.
This applies a default quota of 2 vCPU and 8 GB memory to each user in the
workspace.
If your workspace was provisioned with GPUs, a default quota of 0 GPU
per user applies. If you want users to have access to GPUs, you must modify
the default quotas as described in the next step.
If you want to change the default quotas, click on Default (per
user) .
CML displays the Edit default quota dialog box.
Enter the CPU, Memory, and GPU quota values that should apply
to all users of the workspace.
Click Update.
To add a custom quota for a specific user, click Add
User.
Enter the user name, and enter the quotas for CPU, Memory, and GPU.
Click Add.
Enabling and modifying quotas will only affect new workloads. If users have already
scheduled workloads that exceed the new quota limits, those will continue to run
uninterrupted. If a user is over their limit, they will not be able to schedule any more
workloads.To specify the maximum number of replicas in a model
deployment, go to Site Administration > Settings > Model Deployment Settings. The default is 9 replicas, and up to 199 can be set.