Uploaded September 2026 | Updated November 2026, 13 hours ago
At Stripe, training DNN models on a huge amount of transactional data, $1.9T in 2025, is often limited by slow, I/O-bound CPU preprocessing.
At Ray Summit 2026, Lee Yang, Sr. Principal Software Engineer at NVIDIA, and Pratik Gujjar, Machine Learning Engineer at Stripe, explore shifting data tasks to the GPU using Ray map_batches with cuDF: why GPUs win on shuffle hash partitioning, the tradeoffs of heterogeneous pipelining, and how to run Ray under heavy memory pressure.
You'll leave able to build a stable, GPU-accelerated pipeline that keeps your models fed.
Liked this video? Check out other Ray Summit breakout session recordings: youtube.com/playlist?list=PLZNaJyYZRPdo
Subscribe to our YouTube channel to stay up-to-date on the future of AI: youtube.com/c/anyscale
🔗 Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
At Stripe, training DNN models on a huge amount of transactional data, $1.9T in 2025, is often limited by slow, I/O-bound CPU preprocessing.
At Ray Summit 2026, Lee Yang, Sr. Principal Software Engineer at NVIDIA, and Pratik Gujjar, Machine Learning Engineer at Stripe, explore shifting data tasks to the GPU using Ray map_batches with cuDF: why GPUs win on shuffle hash partitioning, the tradeoffs of heterogeneous pipelining, and how to run Ray under heavy memory pressure.
You'll leave able to build a stable, GPU-accelerated pipeline that keeps your models fed.
Liked this video? Check out other Ray Summit breakout session recordings: youtube.com/playlist?list=PLZNaJyYZRPdo
Subscribe to our YouTube channel to stay up-to-date on the future of AI: youtube.com/c/anyscale
🔗 Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com










