Earlier this month, Intel presented an impressive new deep learning system that transforms the 3D in-game graphics of Grand Theft Auto 5 to photorealistic images. It was doing it at a higher precision than other similar neural networks, and it was much faster than photorealistic render engines that take minutes to hours to render a single frame.
But is the gaming industry ready for AI-based photorealistic render engines. In my latest column on TechTalks, I discuss the key technical and logistical challenges that suggest we will have to wait a few more years before the technology is ready for prime time.
Memory requirements, non-linear operations, and the costs of developing and training models are still too prohibitive to make deep learning–based photorealistic rendering available to game developers and gamers.
Read the full story here.