Snowcap: GPU Profiling Using Machine Learning | Al and Games #83

Support our work at http://www.aiandgames.com Snowcap is a plugin for the Snowdrop engine built by @UbisoftMassiveEntertainment, used in games like The Division, Star Wars Outlaws, and Avatar: Frontiers of Pandora, that estimates GPU performance on consoles - but without running the game on the target hardware. Whether it's Xbox, PlayStation (or Switch 2?), Snowcap uses a trained machine learning model to estimate whether the game will run within acceptable performance ranges. Tommy sits down for a conversation with three developers involved in Snowcap's development - engine programming lead Franck Maestre, associate production director Elia Wrzesniewski, and expert data scientist David Renaudie - to find out how it works, how the project started out, and how it has evolved into an essential part of the process for artists and level designers working in Snowdrop. Note that this episode is not sponsored by @Ubisoft. We did not receive any assets (be it games, hardware, cash, or otherwise) from the company in making this video. [00:00] Intro [01:39] What is Snowcap? [04:26] A Change in Philosophy [09:04] How it Works [13:26] Visualising Performance [20:20] "Simple is Beautiful" [22:37] Credits -- Writer and Voice: Tommy Thompson Editing and Production Support: Shraddha Gupta AI and Games Logos: Andy Carolan Theme Music: Ben Ridge -- 'AI and Games' is a YouTube series on research and applications of Artificial Intelligence in video games. Check out all of our content: - Website and Newsletter: http://www.aiandgames.com - AI and Games: http://www.youtube.com/@AIandGames - AI and Games Plus: http://www.youtube.com/@AIandGamesPlus - AI and Games Conference: http://www.youtube.com/@AIandGamesConference - Goal State: http://www.youtube.com/@GoalStateAI - Livestreams on Twitch: http://www.twitch.tv/aiandgames - Join our Discord Community: https://discord.aiandgames.com -- Follow 'AI and Games' on social media. BlueSky: https://bsky.app/profile/tommy.aiandgames