Accelerate data science and analytics with GPUs
summary
Live demo scans a 340‑million‑row weather table and returns city‑level aggregates in ~80–95 ms on an NVIDIA L4 attached to Cloud Run versus ~8–10 s on a CPU that only loads 113 million rows, a ~100× effective speedup while also using the full dataset instead of a sample. Uses cuDF/cuML and Colab Enterprise with the one‑line load extension cudf.pandas to switch pandas/scikit‑learn code to GPU, plus built‑in profilers to find CPU spillover.
AI-written summary · about
chapters (16)
- 0:00GPU-accelerated data science intro
- 2:23Why GPUs matter for data
- 3:48Climate analytics live demo
- 9:54CUDA and CUDA-X explained
- 15:54Google Cloud GPU setup
- 16:39cuDF in Colab Enterprise demo
- 29:29GPU profiling with line profiling
- 36:23scikit-learn with cuML
- 45:28Learning GPU data science resources
- 46:32GPU data science beginner guide
- 49:23Multinode GPU for large datasets
- 50:52cuML limitations
- 52:40GPU memory management tips
- 54:47GPU acceleration heuristics
- 56:33GPU adoption next steps
- 59:06Closing remarks
related talks
35mWorking with data in Python for the Aspiring Data Scientists
19mData Science in a Cloud World: What Every Data Scientist Needs to Know
50mFast Analytics On Big Data
04h 02mHands-on Session on Data Science for Beginners
28mKubernetes Meets Climate Science: Building Large-scale Feature Detection From Climate Data Records
35mFast Pandas Performance Tricks You Wish You Knew Earlier
44mHigh Perfomance Pandas
31mBeyond Pandas: lightning fast in-memory dataframes with Polars