Background

NVIDIA has requested to showcase the confidential GPU vLLM demo at the upcoming GTC event. The existing demo recording needs to be re-recorded to address visibility issues and use appropriate hardware.

Current Status

• Original demo was recorded on a client machine
• Andre Beausoleil reviewed the demo and provided feedback
• IBM Research (Mengmei) and team are willing to re-record
• Hardware availability confirmed: Intel TDX machines with H100s available at IBM lab
• AMD SNP machines are currently tight on availability

Feedback from Andre Beausoleil

• Text in the four quadrants/windows is difficult to read on 32-inch display
• Need to enlarge text/fonts for better visibility during presentation
• Suggested creating a GMeet recording with narration as train-the-trainer asset
• Script content is acceptable, presentation format needs improvement

Hardware Considerations

• IBM has Intel TDX machines with H100s readily available
• Need to verify if Red Hat's H100 is in SEV-SNP system (Jens checking)
• Alternative option: Use Phoenix system with proper attribution
• Demo needs to run on Red Hat or IBM machine (not client machine)

Requirements

• Re-record demo with larger, readable text/fonts
• Use IBM TDX machine with H100 or Red Hat SEV-SNP setup
• Include clear narration for booth presentation training
• Ensure proper credit to IBM Research and all contributors
• Create train-the-trainer asset for NVIDIA GTC booth staff

Next Steps

• Confirm hardware setup (TDX vs SNP availability)
• Coordinate recording session with Mengmei and Claudio
• Set up demo environment with improved visibility
• Schedule recording session once hardware is confirmed