Possible structure for a talk:
- Base AI/GPU chips (Technology and market share)
- Nvidia
- AMD
- Cloud proprietary
- Other
- "Packaging"
- "Unified Memory" or on package CPU/GPU
- NVidia Grace
- AMD Instinct
- GPU Clusters
- NVidia DGX
- Supporting Hardware
- NVLink/Infinity Fabric/PCI-e GenX/CLX
- Capability
- Available Switch hardware
- Memory/Storage
- LPDDR, HBM
- GPU Direct/Flash storage systems
- Ultra Ethernet/Infiniband
- Software tools
- Status of Cuda "monopoly"
- Support and penetration of higher level, hardware independent, development environments (e.g. PyTorch)
- Impact on carbon footprint and data center infrastructure