This Pedro Valadas guy is bringing the heat and also a huge pile of rocks inside his skull. I'm essentially at the part where he asks if DLSS performance is going to be the main focus going forward or if we're going to see gpu improvements that could help native rendering. Like ... compute power is basically general purpose now. It's going to keep improving. DLSS only takes a few milliseconds to run, so there aren't huge performance wins if you just increase tensor cores. FG will obviously keep improving, but you need frames to feed into it. This is the most obvious question with the most obvious answer. After having asked about whether upscaling was letting developers "off the hook" so to speak, I hope he's not this dumb and he's just asking the questions because a bunch of smooth brains on his reddit were demanding he ask them.
Edit:
Bryan Catanzaro, Nvidia: In the far future we'll have purely neural-network rendering
Jakub Knapik, CD Projekt RED: What he just said scares me
Edit:
If I were to ask a feet to the flames question it would have been about Ray Reconstruction. They mention integration is more complicated, and more tightly coupled with the renderer. A good question would have been about proprietary technology in renderers, how easy it should be to swap in/out competing solutions and if things that are more tightly coupled should be open. DLSS Super Resolution is very easy to swap in out with competing solutions. So where is the line?
It sounds like Nvidia's investment in time and money for something like ray reconstruction is enormous, so keeping it proprietary does make sense. But there is also general industry health, and consumer choice to worry about. Would have been way more interesting than, "Does DLSS make developers lazy" type questions.
Edit:
Also Bryan Catanzaro's responses to the fake frames etc was pretty funny. Kind of trolling and also being insightful about his perspective.