How to Write a CUDA Program - Parallel Programming #gtc25 #CUDA

Click to watch the full session from GTC25: "How to Write a CUDA Program - The Parallel Programming Edition"

Join one of CUDA's architects on a journey through the concepts of parallel programming: how it works, why it works, why it's not working when you think it should be working, and in particular why it's different on a GPU from a CPU.

We'll look at different approaches to parallel programming in CUDA and how to take advantage of the hardware it runs on. This is the next episode in what has become a series looking at the way that CUDA and the GPU work, and why they work this way. This session will focus on how to think about parallel programming on a massively-parallel GPU and why that might be different to what you're used to. If you've never even written a parallel algorithm then you'll learn all sorts of new things, but even if you're a ninja you'll walk away with some new tricks up your sleeve.

Speaker: Stephen Jones, CUDA Architect, NVIDIA

Key Takeaways:
▫️Introduction to parallel algorithms
▫️Hardware and even software design is running up against the laws of physics, which will change computing in a fundamental way
▫️How parallel programming on a GPU is different to what you might be used to
▫️Techniques for getting the most out of CUDA

Topic: CUDA Development and Optimization - Programming Languages / Compilers
Level: Technical - Beginner
Replay of NVIDIA GTC 2025 Session ID: S72897

#CUDA #NVIDIA #gpucomputing #parallelcomputing #parallelprogramming

[GameDevNerd]

I'm a ninja, looking for a battle and a banner to follow. :-)

[CJDcow]

He’s upset people use PyTorch more

Also, why didn’t he just use Python directly, it’s the main CUDA language now, I know more than him, HAHAHAHA