Deployment & Scaling · Advanced
DGX, Grace Hopper, Blackwell and Running NVIDIA AI on the Clouds
A survey of the platform layer: what NVLink and NVSwitch actually change, what DGX and rack-scale GB200 systems are for, and how the same NVIDIA software stack is consumed on AWS, Azure, Google Cloud and neoclouds.
- 1.3 hours lesson
- intermediate level
- NVIDIA AI Track
- Narrated presentation
What you’ll learn
- NVLink and NVSwitch as the difference between eight cards and one coherent machine
- DGX systems, SuperPOD, and rack-scale GB200 NVL72 domains
- Grace Hopper / Grace Blackwell CPU-GPU coherence and when it matters
- NVIDIA AI Enterprise and NIM available across AWS, Azure, Google Cloud and neoclouds
- Choosing instances by interconnect and memory, not headline GPU name
- Power and liquid cooling as the real on-premises constraints
Inside this lesson
A narrated, slide-by-slide presentation with 1 prediction exercise, 1 knowledge check, 3 videos. You answer as you go, and the platform checks your work.