Join the co-founders of Run:AI at GTC for two dynamic sessions on AI cluster orchestration and maximizing GPU utilization.
Run:AI is participating in two GTC ’21 sessions that can be viewed in the GTC portal on-demand from Monday, April 12 through Friday, April 19. Participation is free and registration can be found here.
Cluster Orchestration for AI Productivity [S32851]
Omri Geller, CEO and co-founder, Run:AI
Readily available compute power, particularly from NVIDIA GPUs, is helping to fuel the enormous growth of AI. But getting AI initiatives to production quickly and efficiently is still challenging. Learn how Run:AI provides smart orchestration of AI clusters running on NVIDIA GPUs to help solve AI productivity challenges.
From 20% to 80% GPU Utilization: Best Practices for Maximizing GPU Utilization [SS33287]
Ronen Dar, CTO and co-founder of Run:AI, and Elan Raja, CEO of Scan Computers
Run:AI and Scan Computers often meet deep learning research teams with relatively low utilization of their compute infrastructure. Ronen Dar, CTO and co-founder of Run:AI, and Elan Raja, CEO of Scan Computers, will present a customer use case and describe concrete steps for increasing utilization. Specifically, we’ll discuss:
- The difference between allocation and utilization of resources, and why this matters.
- The impact of static resource allocations on jobs and experimentation.
- How to differentiate between resource allocation for different workload types, such as build, train, and inference workloads.
- The importance of visibility tools in understanding GPU usage.
- Understanding how concepts from optimization of HPC workloads can be “borrowed” for AI experimentation to increase utilization.
We hope you’ll join us at GTC!!