Cyvitrix Learning推出8小时20分钟AI数据中心架构课程，涵盖GPU网络、存储、功耗、冷却与主权AI平台，助初学者从业务需求到基础设施设计全面掌握AI工厂构建。
原始标题：AI Infrastructure Architect: Building Scalable AI Factory
这是由 Cyvitrix Learning 于 2026年10月 推出的 AI 数据中心架构课程（AI Data Center Infrastructure Architecture）的详细大纲。该课程时长 8小时20分钟（包含 41 节课），大小为 5.6 GB，主要面向初学者，旨在教授如何设计 AI 数据中心、GPU 网络架构、存储、功耗、冷却及主权 AI 平台。
Published 10/2026
Created by Cyvitrix Learning | PMP, Security+, CISSP, CISA, CISM, GRC, CGRC, SSCP Trainings
MP4 | Video: h264, 1920×1080 | Audio: AAC, 44.1 KHz, 2 Ch
Level: Beginner | Genre: eLearning | Language: English | Duration: 41 Lectures ( 8h 20m ) | Size: 5.6 GB
Design AI data centers, GPU fabrics, storage, power, cooling, orchestration, inference and sovereign AI platforms.
What you’ll learn
⚡ Translate AI business use cases into measurable infrastructure and service requirements.
⚡ Explain training, fine-tuning, LLM inference, RAG, image, video, and agentic workload characteristics.
⚡ Evaluate CPUs, GPUs, AI accelerators, HBM, PCIe, CXL, and complete compute nodes.
⚡ Assess Ethernet, InfiniBand, RDMA, RoCEv2, congestion, optics, and AI network fabrics.
⚡ Design storage and data paths for datasets, checkpoints, model serving, and retrieval.
⚡ Plan rack power, electrical resilience, air and liquid cooling, and facility readiness.
⚡ Compare Kubernetes, Slurm, GPU scheduling, quotas, automation, and multi-tenancy.
⚡ Explain data, tensor, pipeline, and expert parallelism in distributed AI training.
⚡ Design inference services for latency, throughput, batching, KV cache, and scalability.
⚡ Integrate zero trust, supply-chain protection, observability, availability, and recovery.
⚡ Calculate capacity, total cost of ownership, and assumptions for architecture decisions.
⚡ Assess RFPs, vendor claims, benchmarks, proof-of-concept results, and acceptance tests.
⚡ Build and defend a full AI-factory design through the FalconCore capstone.
Requirements
❗ General familiarity with servers, networks, cloud infrastructure, storage, or data-center technology is recommended.
❗ Basic understanding of enterprise architecture or infrastructure operations will help you connect design decisions.
❗ No advanced AI mathematics, model training experience, or prior GPU-cluster deployment is required.
❗ A spreadsheet or calculator is useful for following sizing and total-cost exercises.
❗ Dedicated GPUs, an AI data center, and paid cloud subscriptions are not required to study the architecture material.
Description
This course contains the use of artificial intelligence.
An AI platform is much more than a collection of GPUs. Its performance, reliability, security, and cost depend on how the workload, compute nodes, network fabric, storage, power, cooling, software platform, and operations work together. This course helps infrastructure professionals understand those dependencies and make architecture decisions using evidence rather than marketing claims.
AI Infrastructure Architect: GPU Clusters & Data Centers is a vendor-neutral masterclass in designing, sizing, evaluating, securing, and operating AI infrastructure. It is intended for enterprise, cloud, telecom, network, and data-center professionals who need to support model training, fine-tuning, LLM inference, retrieval-augmented generation, image and video generation, multimodal AI, and agentic workloads.
Starting with workload discovery, you will learn how model size, precision, memory, batch behavior, throughput, latency, utilization, availability, and data movement affect infrastructure requirements. You will then connect those requirements to CPUs, GPUs, accelerators, high-bandwidth memory, host connectivity, scale-up architectures, and rack-scale systems.
The course explores Ethernet and InfiniBand fabrics, RDMA and RoCEv2, topology and congestion, parallel storage, dataset movement, checkpoints, and the physical realities of power and cooling. From there, you will examine orchestration with Kubernetes and Slurm, distributed training, model serving, KV-cache behavior, request scheduling, cloud and sovereign AI deployment options, and platform security and observability.
A major focus is practical architecture judgment. You will learn to size memory, training and inference capacity, network and storage performance, rack power, and facility requirements; compare alternatives; examine total cost of ownership; and define meaningful benchmarks, RFP requirements, and production acceptance criteria.
The fictional FalconCore Sovereign AI Factory capstone connects the complete lifecycle, from business demand and GPU fabric design through storage, facility planning, model services, security, operations, and procurement. You do not need advanced machine-learning mathematics to follow the course. The emphasis is on clear reasoning, assumptions, trade-offs, and defensible design decisions.
Who this course is for
⭐ Enterprise, solution, cloud, infrastructure, and data-center architects entering AI infrastructure design.
⭐ Network and telecom architects working with high-performance AI fabrics and edge platforms.
⭐ Systems, GPU, HPC, storage, platform, and MLOps engineers looking for an end-to-end architecture perspective.
⭐ Data-center power, cooling, facilities, and operations professionals supporting higher-density AI deployments.
⭐ Cybersecurity and resilience architects responsible for securing AI infrastructure and services.
⭐ Technical decision-makers, procurement teams, and consultants evaluating AI infrastructure investments and proposals.
此处内容需要权限查看
会员免费查看
