45 free lesson minutes · Start now, no credit card needed
Shiur Prati

Private AI infrastructure: GPUs, inference and costs lessons with AI

AI infrastructure: GPUs, inference and costs with a live private lesson: the teacher talks with you out loud and explains on a shared board, moving at your pace through a clear plan, lesson after lesson.

Live 1:1 lesson with an AI teacher

Teacher: Noam

  • Course updated 11/2026
  • Hebrew, and 5 more languages
  • 3 levels
  • 36 lessons · about 20 hours of lessons

AI infrastructure: GPUs, inference and costs

AI infrastructure: GPUs, inference and costs with a live private lesson: the teacher talks with you out loud and explains on a shared board, moving at your pace through a clear plan, lesson after lesson.

12 lessons · 4 modules · Live private lesson with the teacher on the board

What you'll learn

  • What Happens Inside a Model
  • Why AI Needs GPUs
  • Training vs Inference
  • Memory and Large Models
  • Latency and Throughput: Where the Time Goes
  • Smaller and Faster: Quantisation and Serving Engines
  • Where to Run: Cloud vs On-Prem and Capacity Planning
  • Inference Economics: Cost per Request
  • Caching and Fast Response
  • Monitoring and Operating in Production

What's included

  • Live 1:1 lessons with an AI tutor on a shared board
  • 20 hours of live one-on-one lessons
  • 36 lessons on a shared board, each with a clear goal
  • Practice and a understanding check in every lesson
  • A summary after every lesson, saved in the history
  • A completion certificate for each level
  • Computer, tablet or phone
  • Hebrew, plus 5 more languages

Course plan

3 levels · 10 modules · 36 lessons · 19 hr 35 min total

BeginnersBeginner: GPUs, Training vs Inference and Memory4 modules • 12 lessons • 6 hr 30 min
  1. What Happens Inside a Model2 lessons • 1 hr
    1. 1A Model Is Numbers: Parameters and MatricesTrial lessonabout 30 min
    2. 2Tokens: How a Language Model Generates Textabout 30 min
  2. Why AI Needs GPUs3 lessons • 1 hr 40 min
    1. 3CPU vs GPU: Few Smart Cores vs Thousands of Simple Onesabout 30 min
    2. 4FLOPs: Counting How Much Compute a Model Needsabout 35 min
    3. 5Inside the GPU: Cores, HBM and Memory Bandwidthabout 35 min
  3. Training vs Inference3 lessons • 1 hr 40 min
    1. 6Training: How a Model Learns Its Weightsabout 35 min
    2. 7Training vs Inference: One-Time vs Every Requestabout 35 min
    3. 8A Request's Journey: The Serving Stack Mapabout 30 min
  4. Memory and Large Models4 lessons • 2 hr 10 min
    1. 9How Much Memory Does a Model Needabout 30 min
    2. 10Inference Memory: KV Cache and Context Lengthabout 35 min
    3. 11Training Memory: Why Training Needs Several Times Moreabout 30 min
    4. 12When a Model Doesn't Fit: Splitting Across GPUsabout 35 min
IntermediateIntermediate: Efficient Serving — Batching, Quantisation, Cloud vs On-Prem3 modules • 12 lessons • 6 hr 30 min

Mapped to exam: NVIDIA-Certified Associate: AI Infrastructure and Operations (NCA-AIIO), ברמת ההבנה; אין בחינה ישראלית רשמית

  1. Latency and Throughput: Where the Time Goes5 lessons • 2 hr 40 min
    1. 1Prefill vs Decode: The Two Phases of a Requestabout 30 min
    2. 2Why Decode Is Slow: Memory-Bound Work and Arithmetic Intensityabout 35 min
    3. 3Batching: Sharing One Weight Read Across Requestsabout 30 min
    4. 4Continuous Batching and the Request Queueabout 35 min
    5. 5Managing KV Cache Memory: PagedAttentionabout 30 min
  2. Smaller and Faster: Quantisation and Serving Engines4 lessons • 2 hr 10 min
    1. 6Quantisation: From FP16 to INT8 and INT4about 35 min
    2. 7Quantisation in Practice: Methods, Formats and Quality Checksabout 35 min
    3. 8Speculative Decoding and Other Decode Speedupsabout 30 min
    4. 9Serving Engines: vLLM, TensorRT-LLM and SGLangabout 30 min
  3. Where to Run: Cloud vs On-Prem and Capacity Planning3 lessons • 1 hr 40 min
    1. 10Deployment Options: Managed API, Cloud GPU, On-Premabout 30 min
    2. 11Capacity Planning: How Many GPUs for Given Trafficabout 35 min
    3. 12Cloud or On-Prem: The Utilisation Break-Evenabout 35 min
AdvancedAI Infrastructure Advanced: Cost, Caching and Production Monitoring3 modules • 12 lessons • 6 hr 35 min
  1. Inference Economics: Cost per Request4 lessons • 2 hr 10 min
    1. 1From GPU-Hour to Cost per Million Tokensabout 30 min
    2. 2Cost per Request: Input, Output and Long Conversationsabout 30 min
    3. 3A Product Cost Model: Users, Peaks and Headroomabout 35 min
    4. 4Cutting Cost: Model Routing, Trimming and Batch Jobsabout 35 min
  2. Caching and Fast Response4 lessons • 2 hr 10 min
    1. 5Latency Budget: TTFT, TPOT and Percentilesabout 30 min
    2. 6Prefix Caching: Never Compute the Same Prefix Twiceabout 35 min
    3. 7Response Caching: Exact-Match and Semantic Cacheabout 30 min
    4. 8Routing, Autoscaling and Cold Startsabout 35 min
  3. Monitoring and Operating in Production4 lessons • 2 hr 15 min
    1. 9Serving Metrics: What to Actually Measureabout 30 min
    2. 10Cost and Quality Monitoring: Attribution and Evalsabout 30 min
    3. 11SLOs, Alerts and Common Incidentsabout 35 min
    4. 12Capstone: Design, Price and Monitor an AI Serviceabout 40 min

What you need

  • No prior knowledge needed for beginner level: the teacher checks what you already know and starts from there
  • For higher levels: what you learned at the level before (you can skip it if you already know it)
  • A computer, tablet or phone with a microphone, and a browser

Who it's for

  • Adults: for work, study, an exam, or just curiosity

Exams and certifications

  • Intermediate: NVIDIA-Certified Associate: AI Infrastructure and Operations (NCA-AIIO), ברמת ההבנה; אין בחינה ישראלית רשמית
  • AI & Machine Learning

    with Noam

    Live 1:1 lesson with an AI teacherabout 19 hours35 lessons3 levels
  • Using AI Wisely

    with Noam

    Live 1:1 lesson with an AI teacherabout 12 hours24 lessons2 levels
  • Machine Learning

    with Noam

    Live 1:1 lesson with an AI teacherabout 20 hours36 lessons3 levels
  • Deep learning

    with Noa

    Live 1:1 lesson with an AI teacherabout 18 hours34 lessons3 levels
  • Computer Vision

    with Noam

    Live 1:1 lesson with an AI teacherabout 20 hours36 lessons3 levels
  • Generative AI & Prompt Engineering

    with Noa

    Live 1:1 lesson with an AI teacherabout 19 hours36 lessons3 levels
See all courses in Artificial Intelligence

Your teacher

NoamNoamAI private tutor, in a live lesson on a shared board

A private AI teacher who teaches in a live conversation: asks, listens, draws on the board, and lets you solve it yourself with hints. Remembers what you've learned and starts each lesson where you left off.

What's a lesson like?

  • Your child talks with the teacher out loud, like a video call, and asks about anything unclear.

  • The teacher explains, writes and draws on a shared board, and your child solves along on the same board.

  • Snap a photo of a worksheet or homework and work through it, one problem at a time.

  • Parents get a report: study time, what went well and what needs more practice.

Lesson rules

  • Just Learning

FAQ

How does a private AI infrastructure: GPUs, inference and costs lesson work?
Pick AI infrastructure: GPUs, inference and costs and a level, then talk with the teacher out loud. The teacher asks what you already know, explains, writes and draws on the board, and lets you solve on your own with hints until it's clear.
How long does the AI infrastructure: GPUs, inference and costs course take?
12 lessons of about 33 minutes at the first level, at your own pace. You can pause, go back, or skip what you already know.
Do I need prior knowledge?
Not for beginner level. At the start of the first lesson the teacher checks what you already know and adjusts the level, and you can jump straight to an advanced level.
How do I start for free?
Sign in with Google or a code sent to your email, enter your child's name and grade, and the lesson starts right away. Every family gets 45 free lesson minutes in any subject, counted by actual lesson time, with no credit card. Then choose a paid plan and cancel anytime.
What device do I need?
Any computer, tablet or phone with a browser and a microphone. You can also install the site as an app on your home screen.

Want to try it?

Sign in, enter your child's name and grade, and the lesson begins. 45 free lesson minutes, no credit card.

Start for free

Shiur Prati with AI: what it is and how it works

AI infrastructure: GPUs, inference and costsThe first 45 minutes are free

Private lessonStart free