Search

AI Fundamentals and Cisco AI-Ready Infrastructure – 2 days (AI-PW2)

  • Overview

    Description, Pre requisites

  • Content

    Lessons, Course Structure

Artificial intelligence (AI) is a major focus in all the sectors of industry and government.  It is a rapidly evolving space with many advanced features that provide greater insight, knowledge and operational efficiencies in many areas of operation.

Many businesses have indicated that AI is a strategic objective, but few have advanced to implementation and use.

The aim of this session is not to explore the depths of AI but to understand the landscape and how Cisco is evolving as an industry player in both offering products that use AI as well as providing AI ready infrastructures.

The session is broken down into two parts: Presales and Presales Technical

Presales Objectives

  • This is aimed at providing an overview of the AI space and how Cisco is positioning itself in the evolving market
  • This will include products as well as partnerships with other key players to create a framework for AI based around Cisco infrastructure

Presales Technical Objectives

  • This is a more detailed look at the technologies that form the AI landscape and how Cisco infrastructure and ecosystem interact

Target audience:

Presales

  • Account managers
  • Internal Sales
  • Presales SEs
  • IT managers

Presales Technical

  • IT Architects
  • Presales SEs
  • Network Engineers
  • Server Administrators

Pre-requisite skills:

  • Network Admin skills
  • Understanding of programming logic
  • Conceptual understanding of VM and containers
  • Basic knowledge of Cisco UCS server environment
  • Basic Linux overview

Duration:

Presales: Half day

Presales Technical: 2 days

Part 1 – Presales and Presales Technical

Module 1: AI Overview

Overview of AI

  • History and evolution
  • Hype vs reality
    • AI vs Natural intelligence
  • The AI landscape: Narrow AI, General AI, Super AI.
    • Future long way off?
  • AI type overview
    • AI umbrella
    • Machine Learning (ML)
    • Deep learning
    • Supervised and unsupervised
  • AI models
    • What are models
    • Language (LLM)
    • Speech
    • Image/video
    • Mixed modal
  • What is generative AI

Use Cases

  • Industry uses – who is using
    • Real-world examples: healthcare, finance, transportation, entertainment, etc.
  • Detailed Use cases
    • AI assistant
    • AI applications
    • Research and education
    • Content Generation
    • Network analysis
      • Predictive
      • Root cause
      • Log analysis
  • Security
    • Cisco Hypersecure
  • Automation and programming
  • Fraud detection
  • Self-driving Cars
  • Robotics
  • Chatbots

Where are we at now

  • Why recent boom
    • Hardware
    • Code / software availability
    • (Why big now)
  • Is Market ready?
    • % of business say they are ready
    • What do they want
    • How do they deploy
    • Market Statistics
  • Business opportunities
    •  Cisco Focus

Making AI Systems

  • Deployment Options overview
    • Cloud vs on prem
      • hybrid
    • Turnkey/bundle vs DIY
    • How AI is planned and implemented by existing organizations
  • Key AI cloud platforms
    • OpenAI – ChatGPT
    • Google – Gemini and Vertex
    • Microsoft – Copilot and Azure
    • IBM – Watson
    • AWS – Q, Bedrock and Sagemaker
    • NVIDIA – DGX cloud
    • Claude
  • Custom Models
    • Inhouse trained
    • Finetuned Pretrained models
    • AI augmentation
      • Customization with specific information
  • Placement decisions
    • What to look for
    • On prem vs cloud
  • Neural Network Overview
    • What is it and what does it do
    • Deep learning
  • AI Model Lifecycle
    • Preprocess and validate data
    • Training
    • Post work
      • Validating and adjusting
      • Reinforcing
      • other
    • Inference

Impact of AI

  • Challenges and concerns
    • Security and guardrails
    • Negative aspects of an AI system
    • Learning and knowledge
      • Statistical probability
      • Garbage in Garbage out
      • Bias
      • Hallucinations
      • (A little deeper as to how it works)
    • Socialistically changes
    • Human impact jobs
    • Work output and efficiency

Demos

  • LLM demo via Chat GTP
  • NVIDIA image generation
  • Custom AI environment that simplifies network automation

Module 2: Cisco AI

AI System Overview

  • Host
  • GPU
  • Network
  • Storage
  • Code

Cisco market position

  • Roadmap and vision
  • Cisco AI ready infrastructure
  • CVD –
    • options and overview

Cisco Equipment

  • Overall architecture and how component provide complete offering,
  •  Cisco Products used in AI architecture
    • UCS
      • C200 series (customer config)
      • C885 and C845 (pre spec’d)
        • Architecture
    • UCS X
      • Inference focus
      • Intersight
      • Other servers
  • Hyperfabric AI pod Networking
    • Hyperfabric
    • Nexus/NDFC
    • ACI

Partner Overview

  • AI stack – Cisco and partners
  • Nvidia
    • GPU
    • NVAIE
  • Redhat
    • Openshift
    • RHEL
  • Storage
    • Pure Flashstack
    • Netapp FlexPod
    • (others)
  • Databases and data retrieval
    • ( add context for why relevant)
  • Opensource community
    • (important as it is an industry focus)

Part 2 – Presales Technical only

Module 3: AI Infra Components

Host Overview

  • NVIDIA MGX architecture
  • GPU at the heart of AI
    • GPU vs CPU
      • Parallelisation
  • vRAM
    • Sizing importance
  • Cores
    • CUDA
    • Tensor
  • GPU cards and models
    • Nvidia
    • AMD
    • Intel
  • GPU scaling
    • RDMA

Networking

  • Why need fast network
  • Speeds
    • 400-800G
  • Hosts and NICs
    • RoCE2
    • Off load and DPU
      • Nvidia Connect X and Bluefield
  • Switches and fabrics
    • Hyperfabric
    • Latency, QOS and loss/delay
    • InfiniBand positioning
    • (focus will be on Converged Ethernet)

Lab and Demos

  • Connect Housley labs
    • Hyperfabic demo
  • Intersight

Module 4: AI Software Operations

Neural networks

  • Basic overview
    • Nodes
    • Deep and wide
    • Patterns and predictions
    • Weights and Bias
    • Back propagation
  • Model Numbers and naming
    • Parameters
    • Model Size

AI Coding Landscape

  • Python
  • Jupyter Notebook intro
    • NVAIE workbench
  • Key Libraries
    • PyTorch
    • Tensor Flow
    • Langchain
  • Extra packages
    • Numpy
    • Pandas
    • SciKit-Learn
  • AI provider Libraries, repos and APIs
    • Huggingface.co
    • Openai
    • Ollama
    • Github
    • NVIDIA NGC

Containerization

  • Docker
  • Ready-made containers
    • GPU drivers
    • Python Libraries
    • NVIDIA NIMs
  • Kubernetes Positioning
    • Orchestration
    • Scalability
    • High Availability

Labs

  • Explore build environments in a docker env
    • External and internal to container
  • Look at some code for inference in Jupyter lab via NVAIE
    • Load prebuilt env
    • See libraries
    • Run code
  • Look GPU usage
    • Small model (llama 3.1-8B) and then using Nvidia-SMI to observe memory consumption of the GPU

Module 5 – AI Enhancements and Tuning

Overview

  • Pretrained model selection
  • Augmentation mechanisms
  • Role of LangChain

RAG

  • What it is and how it works
    • Tokens and embeddings
    • Vector DBs
  • Prompt engineering

Retraining – AI Finetuning

  • How works
  • Pros and cons

AI Customization

  • Pipelines/workflows
  • Filtering and privacy
  • Search Watch and update
    • Scan WWW
  • Agents
  • Tools

Labs

  • Using Housley labs
    • Load Ollama and OpenAI GUI
    • Add docs to base model to create RAG enhanced env
    • Explore Prompt engineering and filtering
    • Look at Workflows and enhanced AI tools
      • Possibly on www

Module 6 – Partner Offerings

NVIDIA AI

  • Hardware and drivers
    • GPU
    • NICs/DPU
    • (some product examples)
  • NVAIE overview
    • (need to set some limits around this as it is a topic in own right)
    • NVIDIA NIM API endpoints
    • NIM Agent Blueprints and additional AI workflows for common industry use case
    • AI Workbench for development and prototypi
    • NVIDIA NeMo™, RAPIDS™, Riva, and other SDKs and libraries
    • NVIDIA tools for Kubernetes and virtualization support
    • Downloadable NIM containers to self-host on preferred infrastructure
    • Enterprise security, including proactive security patching and remediation guidance
    • Flexible software lifecycle support with API stability
    • Enterprise support, including guaranteed response times and access to NVIDIA experts
    • Base Command™ Manager Essentials for infrastructure management
  • Community knowledge transfer

Redhat Openshift and AI

  • Overview
  • RHEL and AI
  • Openshift
  • AI platform
    • extensions
    • Operators

Labs

  • Using Housley labs
    • Explore
      • NVAIE workbench
      • Openshift
  • Connect to NVIDIA site
    • look over the different type of AI
    • View different NIMs
    • Explore pipelines available and how built on
  • Look at Various Repos and what models they have
    • Hugging face
    • Ollama