PodcastIntel
Sign in Get Started Free
AI Engineer
AI Engineer

From 15% to 90% GPU Utilization: Fix the Data Pipeline, Not the Model

Oct 10, 2026 · 0:17:02
✨ AI Summary
  • Tarun Sunkaraneni from Amazon AGI discussed optimizing GPU utilization for multimodal training.
  • The key issue was a data pipeline bottleneck, not the model itself.
  • Improvements focused on efficient data loading, decoding, and resizing for faster training.

Guests on This Episode

TS
Tarun Sunkaraneni
1 podcast appearance

More from AI Engineer

View all episodes →

Get AI Summaries for Every New Episode

Subscribe to AI Engineer and get AI summaries, guest tracking, and email digests delivered automatically.

Sign Up Free →