Nvidia open source image upscaler - NIS - FSR killer?

[Nexus18]

Interesting stuff from nvidia today, first up video showing how far dlss has come and overview to what exactly DLSS is and how it works:

Essentially not the FSR attack PR piece that we thought it was going to be but a very good educational piece.

The biggest point they announced today was a FSR competitor:

The latest Game Ready Driver releasing on November 16th provides an update to our existing NVIDIA Image Scaling feature that boosts performance on ALL games and GeForce GPUs through a best-in-class spatial scaling and sharpening algorithm. NVIDIA Image Scaling is accessible both from the NVIDIA Control Panel and GeForce Experience, and includes a per-game sharpening setting tunable from NVIDIA’s in-game overlay.

NVIDIA is releasing the NVIDIA Image Scaling algorithm as an open source SDK that delivers best-in-class spatial scaling and sharpening and works cross-platform on all GPUs. The SDK will be publicly available on GitHub on November 16th for all developers to integrate into their games.

https://www.kitguru.net/gaming/dominic-moass/nvidia-image-scaling-analysis-versus-fsr-dlss/

Personally I'll still use DLSS over FSR/NIS where possible.

 

[Nexus18]

I'm going to go out on a limb here and say that Nvidia introduced this because they are going to remove tensor cores from certain GPU models. So high end will have tensor cores, low end will not.

Or this could be a contingency if they are unable to get power under control for these next generation of cards. Chop off the tensor cores and then pretend like DLSS was never a thing.

I don't see why a dev will use this over FSR, since FSR also works on consoles. NIS also goes against the whole Nvidia markerting campaign about DLSS being a good because of vector data or whatever it is called.

Or this could just be the g-sync compatible thing all over again. Regardless I can smell the fear. Competition is back baby.

1st point could be very valid.

Given the amount of work they have done for dlss (both on the technical front and from a PR POV) and getting it adopted into game engines, open source sdk etc. etc., dlss isn't going anywhere imo, the results/comparisons to your generic upscaler i.e. FSR and NIS show why we still need DLSS kind of tech.

FSR/NIS are good for what they were designed for it i.e. vast adoption rate and for people with lower end/older gpus or ones that just don't have the hardware requirements for it. I have yet to test NIS but from my experience with FSR so far and seeing why people think FSR is good is because of the over sharpening, which initially does look better than a blurry/smeary AA mess but after time you can see its shortcomings. It kind of reminds me of people who like things like the "vivid" preset mode on their TV/LCD and sweetfx/redux mods i.e. blown out highlights, shadows/blacks being crushed.

Given NIS is open source and the code is also up on github, there is no reason why developers can't take it and use it, essentially nvidia have done what amd did, here you go, use as you please. The main possible advantage with NIS here though is nvidia have improved upon FSR by letting people enable/use it through their drivers for "any" game, however, if it is anything like the 3rd party mods where FSR is being forcefully injected, chances are it won't be great and will bring its own fair share of problems.

How many consoles games are using FSR? It seems like most console games still rather use adaptive resolution techniques? Or for PS 5 exclusives, checkerboarding.