NVIDIA AI Enterprise Software PlatformAI Proving GroundHigh-Performance Architecture (HPA)BlogArubaAI & DataHPENVIDIA
Blog • • 5 minute read

HPE PCAI - Behind the Curtain

In this blog we will take a peek behind the curtain of HPE PCAI. This will give the infrastructure engineer an idea of what this solution and how they may end up supporting it.

In this blog

  1. The setup
  2. Conclusion
  3. References

Welcome to this blog on HPE PCAI (Private Cloud AI). In this blog, we will show you what it takes to set up this environment and give you, as an infrastructure engineer, tips on how to support it. 

Part of the NVIDIA AI Computing by HPE portfolio, HPE Private Cloud AI is a co-developed scalable, pre-configured, AI-ready private cloud that gives AI and IT teams powerful tools to innovate while simplifying ops and keeping your data under your control.

HPE Private Cloud AI comes in various sizes based on your use case, ranging from small—inference (rack servers, switches and storage) to XL—RAG/fine model tuning (multiple rack solution with at least 11 nodes and 670 TB of storage). Our solution at WWT's Advanced Technology Center (ATC) is a small T-shirt size, which can be shown off in a demo for both the account team and customers via this schedulable lab in the AI Proving Ground: HPE Private Cloud AI.

The demo that we would walk through for you would consist of the following:

  1. Deployment of solution from the ground up
    1. Rack/Cable
    2. Day 0 operations to get the system online
  2. Solution overview from the eyes of an infrastructure engineer
    1. Green Lake for Files Array
    2. vCenter and K8 cluster components
    3. Switching - NVIDIA and Aruba
  3. Day 2 operations: Connecting to Green Lake Portal
    1. Providing access to data scientists and dev engineers to deploy apps
    2. Deploying a chatbot or virtual assistant example

What makes this any different from any other solution out there? Why would I choose to use this? Let's examine more to see if we can answer those questions.

It's a fully integrated solution designed to allow enterprise users to start building their AI applications without having to think about sizing up their infrastructure.

In this case, the big problem is having the hardware together like either of those solutions did and combining HPE AI Essentials and NVIDIA AI Enterprise software to make it useable for your AI use case. This is what sets this solution apart from all others. The best thing about this solution is that all these tools are included from the get-go. You don't have to go out and download Spark, feast, kubeflow, ezpresto; they have workflows you can utilize directly from the solution.

Included tools and Frameworks in HPE PCAI

 The setup

Our solution consisted of four servers (three servers used in the VMware cluster to run the K8 virtual nodes and 1 DL380 as the inference node worker node), two NVIDIA switches, four Aruba switches and a couple of nodes to make up a Green Lake for Files array. 

HLD and Network Design of HPE PCAI SMALL

Once we completed all the networking online and firmware upgrades, we used the steps below to deploy the solution:  

  1. Fill out the deployment sheet for both Green Lake for Files and HPE Private Cloud AI.
  2. Setup two different /24 networks - One for internal networks and One for an External Network to be used in deployment
  3. Deploy the DSC VM that will be used to deploy the infrastructure and connect to Green Lake
  4. Set the DSC VM management network and activate it in the Private Cloud AI portal
  5. Roll the rack into the Data Center
  6. Setup port channel from our core to the NVIDIA Switches
  7. HPE GreenLake cloud setup and onboarding
  8. Deploy HPE GreenLake for Files
  9. Initiate Private Cloud AI setup wizard – which will configure the Infrastructure pieces:
    1. Infrastructure - Management, iLO and data network info control plane node discovery, vCenter, ESXi, iLO credentials.
    2. Control plane VM - Control plane VM networking, credentials.
    3. Worker node servers - Discovery Management, iLO and data network info.
    4. Platform ingress IP
    5. Integration with HPE GreenLake for Files.
  10. Verify Setup is complete and access the Green Lake AI portal to start deploying AI solutions and providing access to the data scientists and developers

Once the above setup was completed, we connected to the AI Private Cloud instance via the GreenLake portal below.

Resource home screen for HPE PCAI

From there, you can start configuring roles and allowing users access to add a virtual chatbot app or even a Jupyter Notebook.

Home Page of AI PG - HPE PCAI
Ex. Jupyter Notebook

Example of the virtual chatbot.

Virtual Chatbot Example

 Conclusion

Hopefully, this blog has piqued your interest in what HPE PCAI (Private Cloud AI) can do to help speed up the time to deliver AI solutions to your customers and give the engineer who is on the hook to get their AI initiatives up and running a good idea of what they would need to support and set up. 

 References

https://www.hpe.com/us/en/solutions/artificial-intelligence/nvidia-collaboration.html

https://www.hpe.com/us/en/private-cloud-ai.html?jumpid=ps_t2647zgj4_aid-521080176&ef_id=EAIaIQobChMIgeHrhvToigMVow2tBh1rmjGfEAAYASAAEgLrD_D_BwE:G:s&s_kwcid=AL!13472!3!702443328466!e!!g!!hpe%20private%20cloud%20ai!21384663562!168777105292&gad_source=1&gclid=EAIaIQobChMIgeHrhvToigMVow2tBh1rmjGfEAAYASAAEgLrD_D_BwE

Technologies