Kashish Mittal
Staff Software Engineer at Uber focusing on hyperscale ML infrastructure, data I/O bottlenecks, and maximizing GPU efficiency for large-scale training; previously built scalable ML systems at Google for YouTube Ads and Core Search Ranking.
Best podcasts with Kashish Mittal
Ranked by the Snipd community
31 snips
Apr 3, 2026 • 53min
Fixing GPU Starvation in Large-Scale Distributed Training
Kashish Mittal, Staff Software Engineer at Uber who builds hyperscale ML infrastructure, talks about solving GPU starvation in large-scale training. He recounts full-stack profiling and tracing to find hidden CPU bottlenecks. He explains reshaping data reads, packing tensors to cut transfers, caching transformed NumPy tensors, and trade-offs between latency and utilization in serving.
