NVIDIA DLSS 3 Delivers Huge Gains to Marvel’s Midnight Suns, HITMAN World of Assassination, and Other New DLSS 3 Titles on GeForce RTX 40 Series

My point is that if DLSS3 evolves it could be another viable tool. I jumped on the DLSS/tensor core train early on and remember how 1.0 barely got the job done and now, even with a 4090 that can do 60+ FPS in a number of games w/o it, I've tested those games with it on/off in 4K and can only just barely notice the difference in quality when off. Ironically when off I've still ended up cranking up AA settings to remove jagged lines and artifacts that became more prominent w/o DLSS. If frame generation reaches a point where its current issues are improved upon similarly then it could be good. I'm not saying it is now, just that it has potential.
It's really not the same, DLSS3 uses a wholly different method to DLSS2. The latter is an AI upscaler, proven technology, very simple. DLSS3 is a frame interpolator except it is trying to do that without the next frame, so basically it is guessing the next frame based on the last few. Which is literally impossible to do perfectly. And also due to this nature DLSS3 will never be able to make the gameplay smoother or the lag less, because it is not reducing the render load on the GPU, but adding a predictor load, hence why DLSS3 increases lag over just using DLSS2 alone.

And this is why I believe it is a marketing tool with no real world benefits.
 
DLSS3 is a frame interpolator except
I've been saying that all along and never said it was the same as 2 and frame generation definitely is not the same as super sampling. Them calling it DLSS3 was a big mistake IMO. I'm just adding that if it were to evolve it could get a lot better at doing it. Who's to say what technology NVIDIA might add down the road to improve upon with it and I'm not expecting perfection but I do expect improvement. TVs have been doing their own version for decades now and, to my knowledge, this is NVIDIA's first round at it.
 
TVs do it by examining Frame -1 and Frame +1 then interpolating an intermediate frame between those. Easy, don't even need ai for this. And Nobody cares about the added give or take 100ms lag that this causes for watching TV.

Nvidia must do it by examining frame -1 -2, then trying to figure out the next frame, it is pure guesswork, it is literally not figuratively impossible to be 100% accurate and it is also impossible to improve responsiveness of the game with it. DLSS3 would be a dud even in the theoretically possible best case scenario. And if not for the marketing value of being able to claim double or triple FPS it would probably never have seen the day.
 
I don't own a console and don't intend to own one any time soon. Just observing how someone entering the market will be making a hard choice.
I agree. You can still build a PC for under $1K but trying to beat console performance is going to be tough, if not impossible, and would probably be closer to around $1500.
 
I agree. You can still build a PC for under $1K but trying to beat console performance is going to be tough, if not impossible, and would probably be closer to around $1500.
If you're JUST thinking about the tower sure and that's still a maybe.
 
Become a Patron!
Back
Top