diff --git a/content/posts/2022/2022-01-27-running-kubeflow-inside-kind-with-gpu-support/index.md b/content/posts/2022/2022-01-27-running-kubeflow-inside-kind-with-gpu-support/index.md
index 314c2199..ccb09dc1 100644
--- a/content/posts/2022/2022-01-27-running-kubeflow-inside-kind-with-gpu-support/index.md
+++ b/content/posts/2022/2022-01-27-running-kubeflow-inside-kind-with-gpu-support/index.md
@@ -30,6 +30,15 @@ _This blog post is intended more as personal notes than instructions, so take ev
 - You'll need an up-to-date version of the [Docker runtime](https://docs.docker.com/engine/install/ubuntu/), mine is `20.10.12`.
 - You'll want the [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/overview.html).
 - You should also install [kubectl](https://kubernetes.io/docs/tasks/tools/), [kustomize](https://kustomize.io/) and [helm](https://helm.sh/) for interacting with our Kubernetes cluster.
+- You should make sure you have nvidia fabric manager installed (ec2 instances by default do not have fabric manager installed), below are a set of commands you can run to install fabricmanager:
+```
+driver_version=$(nvidia-smi | grep -oP "(?<=Driver Version: )[0-9.]+")
+driver_major=$(echo ${driver_version} | cut -d. -f1)
+
+apt-get install nvidia-fabricmanager-${driver_major} -y
+apt-mark hold nvidia-fabricmanager-${driver_major}
+systemctl enable nvidia-fabricmanager.service
+```
 
 If you can run the following example you're all set.