Optimizing GPU Performance in C++23 with NVC++ (Part 2) скачать в хорошем качестве

Optimizing GPU Performance in C++23 with NVC++ (Part 2)

🚀 How far can you push GPU performance using modern C++? In this Part II of his hands-on session at the Gray Scott School 2025, Pierre Aubert dives deep into performance optimization for scientific simulations written in C++17/20/23 using NVC++ and CUDA. This session goes beyond the basics — showing how profiling tools like NVIDIA Nsight, smart thread block configurations, kernel fusion, and memory layout strategies can drastically improve performance. 📌 What you’ll learn: How to profile and compare CUDA vs NVC++ kernel execution Improving GPU occupancy by adjusting thread block sizes and axis ordering Speeding up code by swapping grid layouts and fixing kernel crashes Using streams and concurrency to maximize overlap of compute & transfer Implementing kernel fusion to avoid overhead from micro-kernels Real-world gains: Going from 12s to 4.1s runtime for the same workload! 🎓 This video is for developers and researchers working on high-performance simulations who want to extract the most from their GPU code using modern C++. Useful Link : https://gitlab.in2p3.fr/CTA-LAPP/COUR... _______________________________________________ 🌟 Join us for the Gray Scott School 2026! 🌟 Curious and want a sneak peek of what’s coming? Check out the videos from the 2025 edition on this channel and get inspired for Gray Scott School 2026! 👉 Explore the 2026 program and the agenda of the Gray Scott Thursdays here: https://cc-fr.eu/gray-scott-school-2026/ Don’t miss out — join us this year and dive into the world of high-performance computing with experts, peers, and the HPC community! 💡 Stay up to date with all announcements, webinars, and updates by following us on LinkedIn: https://www.linkedin.com/company/cent...