Wirewalk Instituteof AI
inf-240 Intermediate Waitlist

Cloud GPU Platforms Compared

For architects and engineering leads choosing where to rent GPUs, who leave able to compare hyperscaler and specialist GPU clouds on hardware, networking, storage, capacity mechanisms and managed tooling.

The curriculum is being prepared and reviewed. Enrollment is not open yet.

About this course

Renting GPUs looks simple until you need 256 of them in one placement group next week. The providers differ less in which NVIDIA chip they offer than in how their instances are connected, how capacity is reserved, what managed file systems sit next to the GPUs, and how much of the cluster tooling you must build yourself. This course gives you a comparison method and the current facts to apply it.

You first learn to read an instance specification: GPUs, GPU memory, host memory, local NVMe, and scale-out bandwidth. You then compare the main families: AWS P5, P6-B200, P6-B300 and the P6e GB200/GB300 UltraServers with EFA; Azure ND H100/H200 v5, ND MI300X v5 and ND GB200/GB300 v6 with InfiniBand; Google Cloud A3, A4, A4X and A4X Max alongside TPUs; and Oracle's RDMA superclusters, followed by specialist clouds such as CoreWeave, Lambda, Nebius and Crusoe. The second half covers capacity mechanisms (on-demand, reservations, AWS Capacity Blocks for ML, Google's Dynamic Workload Scheduler, spot and committed spend), managed storage (FSx for Lustre, Azure Managed Lustre, Google Cloud Managed Lustre), and managed cluster layers (SageMaker HyperPod, AWS ParallelCluster, Azure CycleCloud, Google Cluster Director), ending with a scored provider selection exercise. Instance names change often, so every fact is dated to October 2026 and the method is built to be reused.

What you will be able to do

  • Read a cloud GPU instance specification and identify the figures that matter for training versus inference
  • Compare current GPU instance families across AWS, Azure, Google Cloud and Oracle for a stated workload
  • Explain how EFA, InfiniBand and RDMA over Ethernet cluster networks differ across providers and what that means for multi-node jobs
  • Choose a capacity mechanism, such as on-demand, reservation, capacity block, flex scheduling or spot, for a given timeline and budget
  • Select managed storage and cluster tooling that fit a provider choice and an existing Slurm or Kubernetes practice
  • Evaluate specialist GPU clouds against hyperscalers on hardware access, support, contracts and lock-in
  • Produce a scored provider comparison and recommendation for a realistic hypothetical workload

Syllabus

6 modules · 17 lessons · about 10 hours

Module 1: How to Compare GPU Clouds

A repeatable method and the specification fields that actually decide the outcome.
3 lessons
  1. Reading an Instance Spec: GPUs, Memory, Local NVMe and Scale-Out Bandwidth
  2. Bare Metal, Virtual Machines and Rack-Scale Instances
  3. A Scoring Framework for GPU Provider Selection

Module 2: The Hyperscalers' GPU Fleets

Current GPU instance families and cluster networks on the four largest clouds.
4 lessons
  1. AWS: P5, P6 and P6e UltraServers with Elastic Fabric Adapter
  2. Azure: ND v5 and ND GB200/GB300 v6 with InfiniBand
  3. Google Cloud: A3, A4, A4X and A4X Max, and When TPUs Fit
  4. Oracle Cloud Infrastructure: Bare-Metal GPU Superclusters

Module 3: Specialist GPU Clouds

What the GPU-first providers offer, and the questions to ask before signing.
2 lessons
  1. CoreWeave, Lambda, Nebius, Crusoe and Other GPU Clouds
  2. Due Diligence on a GPU Cloud: Contracts, SLAs, Support and Data Residency

Module 4: Getting the Capacity

The purchasing mechanisms that decide whether GPUs exist for you when you need them.
3 lessons
  1. On-Demand, Reservations and Committed Spend
  2. AWS Capacity Blocks for ML and Google Dynamic Workload Scheduler
  3. Spot and Preemptible GPUs: Which Workloads Survive Interruption

Module 5: Storage and Managed Cluster Layers

The file systems and orchestration services that turn instances into a usable cluster.
3 lessons
  1. Managed Lustre and Parallel Storage: FSx for Lustre, Azure Managed Lustre and Google Cloud Managed Lustre
  2. Managed Clusters: SageMaker HyperPod, ParallelCluster, CycleCloud and Cluster Director
  3. Data Gravity: Egress, Object Storage and Moving Datasets Between Clouds

Module 6: Making the Call

Applying the framework to a full scenario and planning for multi-cloud reality.
2 lessons
  1. Worked Comparison: A Hypothetical 512-GPU Fine-Tuning Program
  2. Multi-Cloud and Exit Planning: Portability, Images and Lock-In

Exam and certificate

The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.

Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.

The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, CoreWeave, Lambda, Nebius, Crusoe, NVIDIA, or AMD. Product and company names are trademarks of their respective owners.

More in AI Infrastructure and HPC

All 14 courses in this track
inf-101 Basic Waitlist

AI Infrastructure Explained

A plain-language tour of the stack that runs AI: what a GPU server is, why clusters need special networks and storage, how jobs are scheduled, what an AI data center needs in power and cooling, and how cloud and on-premises options differ. No prior infrastructure background required.

6 h · 15 lessons Join the waitlist
inf-201 Intermediate Available

GPU Infrastructure for AI

The operator's view of data-center GPUs: what the current NVIDIA and AMD systems are, how they are packaged and connected, how the driver, CUDA/ROCm and container stack fit together, how to monitor and triage failing GPUs, how to share and size GPU memory, and what power, cooling and sourcing decisions follow.

12 h · 16 lessons See price at checkout
inf-210 Intermediate Waitlist

Linux Administration for AI Clusters

The Linux skills that differ on an AI cluster: choosing and imaging the OS, packaging GPU and RDMA drivers that survive kernel updates, NUMA and PCIe topology tuning, cgroup v2 isolation, shared identity and file systems, scientific software environments with Lmod, Spack and containers, and patching a fleet without breaking jobs.

14 h · 19 lessons Join the waitlist