"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis

E47: GPUMaxing with Dr. Ronen Dar of Run:ai

Jul 25, 2023
Dr. Ronen Dar, CTO and co-founder of Run:ai, discusses the critical role of GPU optimization for AI in enterprises. He dives into innovations in GPU scheduling and the geopolitical landscape shaping AI chip technology, particularly the competition between major players like NVIDIA and AMD. The conversation also highlights strategies for securing GPU resources amid shortages and the price implications of cloud versus on-premise solutions. Dar's insights reflect the complexities of AI infrastructure and the challenges posed by ongoing geopolitical tensions.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
ANECDOTE

Compute Needs Scaling

  • In 2012, AlexNet's training on two GPUs was a breakthrough.
  • Years later, ResNet required hundreds, showing a 100x compute increase.
ANECDOTE

Simplifying Distributed Training

  • Run:AI aimed to simplify large model training by abstracting away distributed computing challenges.
  • GPT-3's training on thousands of GPUs exemplified this growing complexity.
INSIGHT

Flexible Platform

  • Run:AI's platform is designed for flexibility, working across on-premise, cloud, and hybrid environments.
  • Their open approach integrates with various tools, adapting to the fast-evolving AI landscape.
Get the Snipd Podcast app to discover more snips from this episode
Get the app