#516: Accelerating Python Data Science at NVIDIA

Episode Link

https://share.snipd.com/episode/e814cc9b-febe-4e83-aa91-4a4c160b5568

Episode publish date

August 19, 2025 8:00 AM (UTC)

Last edit date

Aug 21, 2025 4:22 PM

Last snip date

August 21, 2025 5:17 PM (GMT+1)

Last sync date

August 21, 2025 5:18 PM (GMT+1)

Show

Talk Python To Me

Show notes link

https://talkpython.fm/episodes/show/516/accelerating-python-data-science-at-nvidia

Snips

Warning

⚠️ Any content within the episode information, snip blocks might be updated or overwritten by Snipd in a future sync. Add your edits or additional notes outside these blocks to keep them safe.

‣

Episode show notes

Your snips

‣

[07:16] GPUs Are More Than Graphics

‣

[10:46] High-Level APIs Hide Hardware Complexity

‣

[26:26] Use Zero Code Change First

‣

[28:37] Read CSVs On GPU For Big Files

‣

[31:32] Zero-Copy Interop Is Foundational

‣

[34:46] cuDF Development Is NVIDIA-Led

‣

[37:24] Expect Realistic GPU Speedups

‣

[45:59] Scale GPUs With Dask Integration

‣

[53:07] Handle Larger-Than-GPU Data Safely

‣

[56:19] Start With Colab Or Cloud Images

‣

[58:09] Vector Search Is A Natural ML Fit

‣