PodcastIntel
Sign in Get Started Free
AI Engineer

Infra behind Krea 2: How to train and serve at scale — Gabriel Jorge Menezes, Krea.ai

Aug 18, 2026 · 0:16:55
AI Summary
  • GPU utilization is misleading; tensor core utilization shows true usage.
  • Training at scale requires tracking tensor core utilization and InfiniBand counters.
  • Most failures were cross-node communication, leading to custom solutions.

Guests on This Episode

GM
Gabriel Jorge Menezes
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 →