The FSR Thread (Announcements & Discussion)

I bet that blurriness would be nicely cleaned up with Radeon Image Sharpening. nVidia has a similar driver side feature doesn't it?

There are three ways: one can use image sharpening in the control panel -- use the sharpening filter in Freestyle -- tweak the lod bias.
 
Are you sure it isn't the left side that has quality enabled? :lol: Seriously, I admit my eyes are older and tired, but the portion of the image on the left looks better to me. Detail on the walkway stones...actually ....everything.

To make a point, not start an argument, I would say new consoles easily beat that in games I've seen.


Well left side should be native (no super resolution applied), the right side would be with super resolution quality mode enabled.

So your eyes aren't deceiving you, it just means AMD has a bit of catching up to do compared to DLSS 2.0.
 
Here's a full resolution image of "quality mode" on a 1060.

Unfortunately, this doesn't look too promising. The right side of the screen where quality is enabled is noticeably blurrier especially in terms of texture detail, like by the columns. Was DLSS 1.0 this bad? Can't remember, but it was rightfully slammed for its blurriness especially in BF5.

Don't know what "ultra quality" looks like yet, we'll find out soon enough. But it does look like AMD has go through their "DLSS 1.0" moment before they get their "DLSS 2.0" act together eventually down the road.

Using the columns for reference, that looks horrid. Hard to tell in the other parts of the image because there's so much post effects going on, perhaps on purpose.. :sherlock:

Very smart of AMD showing it supported on 1060, that's a blow for NV.

Great option to have but as expected, it doesn't look like a real DLSS competitor at this stage.
 
For anyone who's been paying attention, UE 5 already relies on temporal super resolution for its rendering.

It seems like AMD has taken a page out of Epic's playbook (who in turn seem to have been keeping up with research).

You can literally see the effect in motion right now by downloading the UE5 dev kit.

I don't know much about the newer papers on them, but there do seem to be a lot. And I suspect a bunch of them take object recognition into account, so there will almost be nothing in it between DLSS and this.


[yt]SeY-ousiCPg[/yt]
 
yup first thing that struck me as well. Really interested to see this in action for myself, in a game.
 
For anyone who's been paying attention, UE 5 already relies on temporal super resolution for its rendering.

It seems like AMD has taken a page out of Epic's playbook (who in turn seem to have been keeping up with research).

You can literally see the effect in motion right now by downloading the UE5 dev kit.

I don't know much about the newer papers on them, but there do seem to be a lot. And I suspect a bunch of them take object recognition into account, so there will almost be nothing in it between DLSS and this.


[yt]SeY-ousiCPg[/yt]

There is a big difference that we now know though. FSR didn't end up using temporal data, where are EPIC's explicitly does. So there is a divergence in tech there.

But what EPIC has done... it kinda the logical fallout as upscaling tech gets more common.
 
There is a big difference that we now know though. FSR didn't end up using temporal data, where are EPIC's explicitly does. So there is a divergence in tech there.

But what EPIC has done... it kinda the logical fallout as upscaling tech gets more common.

I was going to say that they name dropped temporal super resolution in the AMD release video, but it's actually 'spatial upscaling' they mentioned. https://youtu.be/eHPmkJzwOFc?t=57

That's DLSS 1. :(

DLSS 2 uses motion information.



https://www.anandtech.com/show/1672...x-super-resolution-open-source-game-upscaling
 
Nvidia added motion info to spatial upscaling but seems spatial is still the root of the tech of dlss.

For GPU junkies, many of you will recognize this as a similar strategy to how NVIDIA designed DLSS 1.0, which was all about spatial upscaling by using pre-trained, game-specific neural network models. DLSS 1.0 was ultimately a failure – it couldn’t consistently produce acceptable results and temporal artifacting was all too common. It wasn’t until NVIDIA introduced DLSS 2.0, a significantly expanded version of the technology that integrated motion vector data (essentially creating Temporal AA on steroids), that they finally got DLSS as we know it in working order.

Given NVIDIA’s experience with spatial-only upscaling, I’m concerned that AMD is going to repeat NVIDIA’s early mistakes. Spatial is a lot easier to do on the backend – and requires a lot less work from developers – but the lack of motion vector data presents some challenges. In particular, motion vectors are the traditional solution to countering temporal artifacting in TAA/DLSS, which is what ensures that there are no frame-by-frame oddities or other rendering errors from moving objects. Which is not to say that spatial-only upscaling can’t work, only that, if it’s competitive in image quality with DLSS, that would be a big first for AMD.

For all we know the added year or 2 of work AMD took to work on the tech might have overcome some of the issues. And maybe enabling other features like CAS while using FSR also helps image quality.
 
FSR will not match DLSS in image quality. The best hope is that it will be a "good enough" solution, with massive adoption rates, and kindof brand DLSS as the snooty master-race option, while FSR is the every-day working man solution.

btw, I used to go to church with the guy who led the R&D for DLSS. He's got quite an amazing team over there, and Nvidia has invested in them for several years. This is an area where AMD is simply out-gunned... they're not going to spend a couple years throwing together a solution that matches the work Nvidia has done. FSR will improve over the next couple years, but so will DLSS, and DLSS will ALWAYS have better quality.

FSR will be fine. DLSS will be fine. There's room for both in the market.
 
While I cant see myself using anything other than the top quality option in any upscaler I can see that the top amd upscaler might be one notch below the top dlss.

And thats with adding CAS to FSR but the thing is I cant see myself using any upscaler unless absolutely necessary. If the game runs ok at native I wont use any upscaler.

If Id have to run a lower level quality setting Id rather just upgrade.
 
Finally some games that support both dlss and fsr to compare.

AMD FSR has just been added to Necromunda Hired Gun, a fast-paced first-person shooter. Although this game is relatively new, it already offers support for NVIDIA DLSS as well as AMD FSR, being one of the first games to do so.

By offering DLSS and FSR support there is finally a way to compare both technologies side by side to determine which technology is better. This was partially done by Back4BuckPC Gamer who compared both technologies at 4K resolution when using their best quality profiles. This part is certainly important as DLSS has always been known to do wonders at lower resolutions, it is where machine learning upscaling algorithms show their true potential, reconstructing images from much fewer pixels.

On the other hand, there is AMD FidelityFX Super Resolution (FSR) which is now confirmed to be an implementation of Lanczos resampling that preserves edges. AMD themselves confirmed this by going open source with FSR. The source code, samples, and documentation are now all available to the public.

What is important to note is that NVIDIA DLSS Quality preset upscales the output image from 1440p resolution, while AMD FSR Ultra Quality uses 1662p image. Both upscaling techniques were compared side by side by Back4BuckPC Gamer in Necromunda Hired Gun. He used 4K resolution and the best quality presets. The higher the quality the fewer pixels super-resolution algorithms need to reconstruct, thus, they are expected to perform similarly.
Necromunda-Hired-Gun-AMD-FSR-VS-NVIDIA-2-850x478.jpg


https://videocardz.com/newz/amd-fsr...ance-at-4k-resolution-in-necromunda-hired-gun
 
Necromunda looks insane. FSR looks phenomenal as well .. much better than any other game has shown.

Here's FSR in Necromunda on a 6900XT, comparing 4K Native vs FSR 1.0:

[yt]ystHQ5Ath2w[/yt]

Make sure you watch in 4K.
 
Nice, was wondering if the chainlink fence would show it but cant tell the diff...
 
looks considerably better than ''just an upscaler'' that I read from the other fanboys. Because if it is, than A.I is looking more and more like a marketing flex.
 
To be fair, this is really the only game that I think FSR looks good in. It's new tech, and in classic AMD fashion, they did not showcase FSR in a good way when they revealed it.

If Necromunda will be the standard for FSR implementation, then woo! The other thing, is that DLSS has an Ultra Quality feature also on the way, which should put it back on top. IIRC, DLSS Quality is using a lower resolution image for the upscaling than FSR Ultra Quality. Once Ultra Quality DLSS comes out, I'd expect it to take back the throne. I think the A.I. implementation of DLSS is shown just how good it is, in this case, because it's able to take a lower resolution image and make it look as good or slightly better than the higher resolution image that FSR is starting with.

Either way, I'm glad this isn't turning out to be a **** implementation. I was not impressed whatsoever with FSR in any other examples thus far. This was a great sight to see :up:
 
To be fair, this is really the only game that I think FSR looks good in. It's new tech, and in classic AMD fashion, they did not showcase FSR in a good way when they revealed it.

If Necromunda will be the standard for FSR implementation, then woo! The other thing, is that DLSS has an Ultra Quality feature also on the way, which should put it back on top. IIRC, DLSS Quality is using a lower resolution image for the upscaling than FSR Ultra Quality. Once Ultra Quality DLSS comes out, I'd expect it to take back the throne. I think the A.I. implementation of DLSS is shown just how good it is, in this case, because it's able to take a lower resolution image and make it look as good or slightly better than the higher resolution image that FSR is starting with.

Either way, I'm glad this isn't turning out to be a **** implementation. I was not impressed whatsoever with FSR in any other examples thus far. This was a great sight to see :up:

From what I have seen, just keep it on 4k highest setting. Granted, that requires a 4k monitor.
 
Nah. The much smoother frametimes are welcomed at 4K. FSR/DLSS, in the case of Necromunda, actually looks better in some ways than native 4K; the hair is a big improvement with DLSS over FSR/Native.
 
I mean 4k FSR highest setting. If you can, because from all the games I have seen it is either the same or better than native. Pretty impressive start. Not sure if they can improve the lower resolutions that much. A lot less pixel information to work with.
 
Back
Top