Full Transcript
Loading transcript…
75th ECTC Wednesday Keynote
Watch with subtitles, summary & AI chat
Add the free Subkun extension — works directly on YouTube.
- Watch
- Subtitles
- Summary
- Ask AI
Professionals and researchers interested in the future of AI, semiconductor technology, and its applications in various industries.
TL;DR
This keynote from ECTC 2025 discusses the explosive growth in AI compute demand driven by larger, smarter models. The speaker, Sam Nasfiger from AMD, argues that this trend is durable, citing transformative applications in medicine, robotics, and science, and outlines key AI evolution trends like post-training reasoning.
Key Takeaways
- The demand for AI compute is exploding due to the exponential growth in model size and parameter counts, far exceeding historical Moore's Law rates.
- AI's potential in the medical field is immense, offering advanced diagnostics by analyzing vast patient data and revolutionizing drug discovery.
- Robotics will be transformed by integrating AI for real-time sensor processing and digital twin technology, enabling adaptable and efficient automation.
- AI for science, including agriculture and material science, promises significant advancements by analyzing complex data patterns to optimize yields and discover new solutions.
- The pace of AI evolution is unprecedented, moving faster than any previous technological era, requiring rapid adaptation in technology development.
- Large model training is shifting from consuming existing internet data to new methods like post-training and chain-of-thought reasoning due to data limitations.
- The economic drivers for AI trends appear durable and sustainable, fueled by transformative applications across multiple critical industries.
- Future AI development will focus on efficient compute solutions to meet the escalating demand, driven by the need for increasingly sophisticated models.
In This Video
- 00:00Welcome to ECTC 2025
General chair Florian Herald welcomes attendees to ECTC 2025, a flagship conference.
- 00:35Gratitude to Sponsors and Exhibitors
Thanks are extended to sponsors for their support and exhibitors for showcasing cutting-edge technologies.
- 01:24Appreciation for Speakers and Volunteers
Speakers and panelists are thanked for sharing expertise. Volunteers are recognized for their efforts.
- 02:31Logistical Information and App
Attendees are reminded to install the app for the program and networking features.
- 02:59Introducing Keynote Speaker Sam Nasfiger
Sam Nasfiger from AMD is introduced as the keynote speaker.
- 03:38The Exploding Demand for AI Compute
Sam Nasfiger discusses the challenges and exploding demand for AI compute in the current era.
- 04:16Exponential Growth in AI Model Size
AI model sizes and compute requirements are growing exponentially, far exceeding Moore's Law.
- 06:10AI Applications in Medicine
AI's immense potential in medical diagnostics, patient consultations, and drug discovery is highlighted.
- 07:56Transformative Impact of Robotics
Intelligent robots, powered by AI and digital twins, will transform industries with 24/7 operation.
- 09:14AI for Science and Agriculture
AI is revolutionizing science, including climate pattern analysis and optimizing agricultural yields.
- 11:00Rapid Evolution of the AI Field
The AI field is evolving at an unprecedented pace, faster than ever experienced.
- 11:36Trends in AI Model Training
AI is getting smarter through post-training, test-time compute, and chain-of-thought reasoning.
Questions & Answers
What is ECTC 2025?
ECTC 2025 is a flagship conference of the ITE electronic packaging society, marking the culmination of hard work and collaboration for this event.
Who is the keynote speaker for ECTC 2025?
The keynote speaker for ECTC 2025 is Sam Nasfiger from AMD, who is SVP and corporate fellow responsible for technical strategy and product architectures.
Why is there an exploding demand for AI compute?
The demand for AI compute is driven by larger, smarter models with an explosion in parameter counts and the compute required for training them at an unprecedented rate.
What are some applications of AI mentioned in the keynote?
AI applications mentioned include medical diagnostics and recommendations, robotics for industrial and adaptive tasks, drug discovery, and AI for science in areas like agriculture and climate patterns.
How is AI getting smarter if internet data is consumed?
AI is getting smarter through approaches like post-training, test-time compute, and chain-of-thought reasoning, where models check each other and generate synthetic data.
What logistical information was provided at the conference?
Attendees were advised to install the conference app for the up-to-date program and a messaging platform for networking. A Q&A session after the keynote would use the app.
Key Terms
- ITE electronic packaging society — An organization that hosts ECTC, a flagship conference focused on electronic packaging.
- Exaflop — A unit of computing speed equal to one quintillion (10^18) floating-point operations per second, used to measure the compute power needed for AI models.
- Digital twin — A virtual representation of a physical object or system, used here in robotics for modeling and training robots with data from other environments.
- Multimodal models — AI models that can process and understand information from multiple types of data, such as text, images, and videos.