events
Faster AI: GPU Rooflines, Agent Learning & Homa’s Fast Transport Protocol
Tre interventi sulle prestazioni dell’IA: l’utilizzo delle GPU in presenza di utenti simultanei, la trasformazione delle tracce degli agenti in dati per l’addestramento e il trasporto di piccoli…

In inglese
- Starts
- 2026-10-19 16:00 UTC
- Organizer
- AI Performance Engineering Meetup (New York)
- Location
- Online event
The AI Performance Engineering Meetup runs an online session with three talks plus organiser updates. Rajat Dandekar, co-founder of Vizuara, uses code and visuals to show how concurrent users change GPU utilisation, linking vLLM measurements on an H100 to the roofline model. Rustem Feyzkhanov of Snorkel AI describes turning agent traces into reproducible simulations, verifiable rewards and model improvement through fine-tuning or reinforcement learning. John Ousterhout, professor emeritus at Stanford, explains how Homa reduces small-message latency, with its relevance to AI workloads framed as an open question.
Event website