Nvidia's RTX Video Super Resolution now works in VLC to upscale offline videos

Shawn Knight

Posts: 15,294   +192
Staff member
In brief: Nvidia's RTX Video Super Resolution AI upscaling feature now comes baked into VLC media player, allowing RTX 30 and 40 Series GPU users to leverage their card's processing power to give offline videos a visual facelift.

Nvidia announced the upscaling tech in late February as a way to remove compression artifacts and other eyesores from most online videos streamed using Google Chrome or Microsoft Edge. It uses the card's Tensor cores along with AI image processing to make videos look better. Now that it works in VLC media player, the same smarts can be used to upscale videos from your own personal library.

To enable Video Super Resolution, simply open the Nvidia control panel and tick the Super Resolution option under the Adjust Video Settings section.

VLC was originally developed with a client / server infrastructure in place, but that functionality was eventually removed when it transitioned to more of a pure media player. It was released with a GNU General Public License in early 2001.

Have you tried Video Super Resolution with VLC media player yet? I have not, and am not totally clear on exactly how it works. Perhaps the biggest question is, can it be used offline or is an active Internet connection required? That is, does 100 percent of the processing take place locally on your hardware or is a connection to an Nvidia server needed? If so, what's the purpose of the connection? Is Nvidia further training its AI on your personal video library?

I suspect it works just fine offline but in this day and age, it never hurts to ask questions about privacy.

For those that have tried VSR, what has your experience been like? Is the improvement in quality as drastic as Nvidia suggests? Are some content types better for upscaling than others? Any feedback would be greatly appreciated.

You can grab the latest version of VLC with support for RTX Video Super Resolution over on our downloads page.

Permalink to story.

 
This special 4xx MB version of VLC crash on any video I try to play.

Same for me. I picked the beta version on vlc website (3.0.19) which supports rtx video but it crashes. Might need some extra drivers from nvidia?
The final version is supposed to be out on the 18th April so still one week.
 
Interesting... this is cool tech and I hope it really becomes useful - and is locally processed. With that said I'd love to see some sort of tech that can analyze old movies and TV shows that are in 4:3 ratio format and magically create a 16:9 or other sized video. I imagine with a TV series or anything where the same scene is seen over and over, AI or something should be able to fill in the sides of each 16:9 frame that was originally 4:3 with actual viewable area. How freaking cool would that be!?
 
Seems worthy of a proper test rather then just a news article admitting we don’t know much about it. (Thank you for not pretending you know).

Sorry only a 2 series here, so I can’t test.
 
As far as I can tell it's not like this is significantly different to simply using a high quality scaler on mpv, like FSRCNNX, Ravu or Anime4K, I would test it, but only Nvidia and Intel have options with VSR and VPE, so my AMD GPU doesn't support it. Also this has been available for some time as a MPC-HC/BE video renderer, searching "emoose VideoRenderer" should lead you to the github page. People that want to test it can try that option if they prefer.
 
I've tried it in browser, and it's good. Only works on Edge and Chrome though.

It's not as good as NGU so no real point in using it for offline video. As others have stated it's also been available for MPC-HC/BE, although it's still in early stages and from what I've read might be a bit buggy.

@KaitouX I'm curious about those other scalers you listed, how do they compare to NGU?
 
I've tried it in browser, and it's good. Only works on Edge and Chrome though.

It's not as good as NGU so no real point in using it for offline video. As others have stated it's also been available for MPC-HC/BE, although it's still in early stages and from what I've read might be a bit buggy.

@KaitouX I'm curious about those other scalers you listed, how do they compare to NGU?
FSRCNNX and Anime4K(without denoise, dering, darken, and all other things they have as additional shaders) are slightly better than NGU Sharp Very High when it comes to PSNR, SSIM, Ravu is slightly worse on those metrics.
In my subjective opinion, FSRCNNX is the most faithful to the source image, but it results often in ringing artifacts and it amplifies other artifacts, Ravu is similar to FSRCNNX but slightly blurrier, making artifacts be a lot less visible, Anime4K is pretty similar to FSRCNNX, but with artifacts that aren't as visible, I only use dering, as the default already looks clean and the other additional shaders look horribly overdone for me. For 720p in my 1440p monitor I use A4K, for lower resolutions I use RAVU, 1080p I use SSimSuperResolution, so I can upscale directly to 1440p while keeping the quality close enough to the other scalers, rather than having to double it to 2160p and downscale from there.
Performance wise, Ravu is the fastest for the most part, A4K fastest variants are similar to Ravu slowest ones, but the slower variants are significantly slower, FSRCNNX is close to the slower A4K variants.
I would say NGU Sharp Very High is like a slightly blurrier and less accurate FSRCCNX.
 
Just curious. Why is there no mention of why Firefox is the odd man out? Isn't Chrome far more "restrictive" in what it lets 3rd parties do with their browser then Firefox is?

I'm also interested in seeing how Firefox does now that that Win10 slowdown bug was patched. The one I didn't know existed. It'll be interesting because FF has lost a lot of people over the last few years due to performance issues.
 
Interesting... this is cool tech and I hope it really becomes useful - and is locally processed. With that said I'd love to see some sort of tech that can analyze old movies and TV shows that are in 4:3 ratio format and magically create a 16:9 or other sized video. I imagine with a TV series or anything where the same scene is seen over and over, AI or something should be able to fill in the sides of each 16:9 frame that was originally 4:3 with actual viewable area. How freaking cool would that be!?
Well, sorry to tell you that until AI is able to create a whole movie you won't see that technology. There's a huge difference between reconstructing data mainly on a temporal interpolation or shape recognition, and just creating something that is related to absolutely no data.
 
Back