AMD Releases Radeon Software Adrenalin 21.6.1 Driver with FidelityFX Super Resolution Support

Tsing

The FPS Review
Staff member
Joined
May 6, 2019
Messages
12,268
Points
113
amd-radeon-software-adrenalin-2020-edition-1024x576.jpg
Image: AMD



AMD has released Radeon Software Adrenalin 21.6.1. The latest driver adds support for Radeon RX 6800M mobile graphics for laptops, Tuque Games’ new Dungeons & Dragons: Dark Alliance title, as well as FidelityFX Super Resolution (FSR) support for select titles. According to a leaked list, the first titles to support FSR at launch are 22 Racing Series, Anno 1800, Evil Genius 2, Godfall, Kingshunt, Terminator: Resistance, and The Riftbreaker.









Fixed Issues



Radeon FreeSync may intermittently become locked while on desktop after performing task switching between extended and primary displays upon closing a game, causing poor performance or stuttering.Anno 1800 may crash upon launching this game when running DirectX 12.AMD cleanup utility may clean up chipset/RAID installer related folders/registries from the system.Some Radeon Graphics...
Click to expand...

Continue reading...


 
I really do not understand AMD with this FSR thing.

NVIDIA was very proud to launch DLSS, the press was briefed prior, it was a big "launch."

With AMD FSR? Nothing. As far as I know, the press hasn't been contacted? I have seen anything official from AMD, no briefings prior to launch, no information sent promoting the feature, no games being offered to press to test it out, no "launch" reviews lined up to go live when the feature was enabled.

It's like beyond Computex, there's no promotion of it? No official launch. Nothing for the Press has been set up to test and review for a launch date.

I'm boggled by this.
 
Seems we didn't get the memo.

Seriously though, there wasn't an official "launch" time or embargo for a review, as far as I know. I did not have access to an early driver.

I'm not going to let that stop me from testing it, however. Might not be immediately, but I will get to it and cover it when I can.
 
Brian_B said:
Good read



Hasn’t the trade off between performance and IQ always been the crux of the matter?
Click to expand...

I would argue, never in this way.

Remember the whole Quake/Quak issue? Remember lowering Trilinear, heck even Bilinear Filtering and Mip Mapping quality to improve performance? Those things were counted as CHEATS back in the day, and everyone got mad and frowned upon it. And now we are accepting a lowering of IQ in this manner. A decade ago, it would be sacrilidge. So yeah, I get it. It's a topic I'd bring up in an article as well when I do my article. It's really weird how we are all ok with it now.
 
To expand upon that. With DLSS there is the potential the rendered image can look AS good, or even BETTER.

But with FSR, that will never be the case. It will always be, under.
 
Brian_B said:
I'm still highly skeptical of that.
Click to expand...

DLSS uses a very high resolution reference image for the machine learning/AI to learn how to reproduce. It can keep learning, to improve it.

AMD FSR does not have a reference image, it will never know what the end result is supposed to look like and always be running multiples of resolution beneath the native resolution.
 
Reading around the web - consensus seems to be that it does help a good bit on performance (although the "high performance" options destroy IQ pretty badly). It isn't as good as DLSS is, but it it works, and it works on a much wider range of hardware.

Sounds to me like it doesn't have to be better than DLSS to be useful - it will have console support, and cross-vendor support. It really looks like it's setting up to be about like FreeSync vs G-Sync... a lesser open standard (that no other brand picked up for a really long time) versus a better, more expensive proprietary vendor lock. One doesn't necessarily need to replace the other. but they don't exactly compliment each other either.
 
This is what I would write in my conclusion if/when I do an article.

1.) FSR is easier to implement for developers than DLSS. Takes less time, and less support.
2.) FSR can be used on a wider variety of devices: PC/Mobile/Console. It really is hardware agnostic, but it is up to the GPU vendor to optimize that performance.
3.) Both FSR and DLSS require Developers to integrate it into the game and provide toggles in the game to enable it. How well it is implemented and tweaked for performance and IQ will rely a lot on the developer for FSR.
4.) AMD's biggest problem in the past, and today, is developer relations and getting developers to implement their technology. NVIDIA pays a lot of money to have their stuff, in games. They have a huge Dev Rel team they send out to developers to implement things like DLSS, they actually do the work for the developer, to make sure it is implemented. This is AMD's weak area. NVIDIA will send a team to you to optimize DLSS performance and IQ, whereas for FSR, it is more up to the developer to do that.
5.) DLSS will have the IQ advantage because it actually uses a reference higher resolution image to learn with machine learning how to upscale to it. DLSS does more work than FSR. This means the potential for DLSS to improve will continue as machine learning and AI improve the upscaling. The potential for it is to only get better over time. It can provide sharper and better looking images than native, the potential is there.
 
Finally available in its 4 quality presets

-BAD
-WORSE
-TERRIBLE
-ARE YOU KIDDING ME!!!!???

Ok I'll cut them some slack as its ther 1st try and we all know how DLSS wasn't any better initially, but I mean it really doesn't look different than normal upscaling+sharpening filter even at ultra quality.

At from what I've seen it even looks worse at 4K. At least DLSS actually looks pretty good in performance mode at 4k.
 
Brent_Justice said:
To expand upon that. With DLSS there is the potential the rendered image can look AS good, or even BETTER.

But with FSR, that will never be the case. It will always be, under.
Click to expand...
True true. I do wonder it it might be possible to come up with set predictive algorithms that could correct many of the image aberration AMDs solution will suffer from.
Then again more processing, diminished returns.
I don't think this is any " response" to DLSS . Its apple and anvils.
 
Brent_Justice said:
This is what I would write in my conclusion if/when I do an article.

1.) FSR is easier to implement for developers than DLSS. Takes less time, and less support.
2.) FSR can be used on a wider variety of devices: PC/Mobile/Console. It really is hardware agnostic, but it is up to the GPU vendor to optimize that performance.
3.) Both FSR and DLSS require Developers to integrate it into the game and provide toggles in the game to enable it. How well it is implemented and tweaked for performance and IQ will rely a lot on the developer for FSR.
4.) AMD's biggest problem in the past, and today, is developer relations and getting developers to implement their technology. NVIDIA pays a lot of money to have their stuff, in games. They have a huge Dev Rel team they send out to developers to implement things like DLSS, they actually do the work for the developer, to make sure it is implemented. This is AMD's weak area. NVIDIA will send a team to you to optimize DLSS performance and IQ, whereas for FSR, it is more up to the developer to do that.
5.) DLSS will have the IQ advantage because it actually uses a reference higher resolution image to learn with machine learning how to upscale to it. DLSS does more work than FSR. This means the potential for DLSS to improve will continue as machine learning and AI improve the upscaling. The potential for it is to only get better over time. It can provide sharper and better looking images than native, the potential is there.
Click to expand...
Lil amd is 2 years away from no longer having excuses of being lil or broke.
They really need to get their ducks in a row in all the other aspects of expanding the company.
 
Brent_Justice said:
This is what I would write in my conclusion if/when I do an article.

1.) FSR is easier to implement for developers than DLSS. Takes less time, and less support.
2.) FSR can be used on a wider variety of devices: PC/Mobile/Console. It really is hardware agnostic, but it is up to the GPU vendor to optimize that performance.
3.) Both FSR and DLSS require Developers to integrate it into the game and provide toggles in the game to enable it. How well it is implemented and tweaked for performance and IQ will rely a lot on the developer for FSR.
4.) AMD's biggest problem in the past, and today, is developer relations and getting developers to implement their technology. NVIDIA pays a lot of money to have their stuff, in games. They have a huge Dev Rel team they send out to developers to implement things like DLSS, they actually do the work for the developer, to make sure it is implemented. This is AMD's weak area. NVIDIA will send a team to you to optimize DLSS performance and IQ, whereas for FSR, it is more up to the developer to do that.
5.) DLSS will have the IQ advantage because it actually uses a reference higher resolution image to learn with machine learning how to upscale to it. DLSS does more work than FSR. This means the potential for DLSS to improve will continue as machine learning and AI improve the upscaling. The potential for it is to only get better over time. It can provide sharper and better looking images than native, the potential is there.
Click to expand...

1) I'm not really sure that's the case, considering DLSS is already implemented in UE4 and 5 and upcoming in Unity and CryEngine.
2) Surely, it will probably be a boon for consoles and mobile gaming.
3) doesn't that contradict point 1?
4) pretty much spot on. Say anything you want about nvidia mistreating developers and partners, its DR is second to none. AMD really needs to invest into this for more developers to embrace it and help them make it work well
5) I actually expected AMD to use some sort of AI implementation as it is indeed capable of handling AI, not anywhere near Tensor core level, but still. AFAIK DLSS is only the starting point on AI technologies for games. There's been rumors of texture AI upscaling and a deeper integration with raytracing/pathtracing.
 
Stoly said:
1) I'm not really sure that's the case, considering DLSS is already implemented in UE4 and 5 and upcoming in Unity and CryEngine.
2) Surely, it will probably be a boon for consoles and mobile gaming.
3) doesn't that contradict point 1?
4) pretty much spot on. Say anything you want about nvidia mistreating developers and partners, its DR is second to none. AMD really needs to invest into this for more developers to embrace it and help them make it work well
5) I actually expected AMD to use some sort of AI implementation as it is indeed capable of handling AI, not anywhere near Tensor core level, but still. AFAIK DLSS is only the starting point on AI technologies for games. There's been rumors of texture AI upscaling and a deeper integration with raytracing/pathtracing.
Click to expand...

1.) It is, FSR is much quicker to integrate.
3.) No. There's a difference between just making it work, and making it work well. It takes very little time to integrate, but if you want to get the best image quality and performance balance, you need to further tweak FSR.

Now, there is a downside to DLSS also that a lot of people don't talk about. DLSS is heavily tied, well, directly tied, to a game's implementation of TAA. How good TAA is implemented in a game determines how good DLSS is going to look. Cyberpunk has a poor implementation of TAA, and hence is not the best at DLSS IQ as some other games.

However, this is one advantage DLSS also has, over FSR. DLSS operates temporally, so in-motion AA looks better. Whereas FSR is spatial and doesn't operate on the Temporal AA level. Maybe the next version will, hopefully.

So on one hand being reliant on a game's implementation of TAA is bad, but on the other hand, you do want to do Temporal AA with your upscaling, rather than not.

Improvements for DLSS will be to de-couple it from a game's implementation of TAA, and still retain the ability to provide Temporal AA.

FSR's improvement will be to actually gain the ability to Temporal AA.

So both have room for improvement.
 
Everyone talks about machine learning. And how that will give dlss the edge.

I say why can't we let games teach amds solution? You install the game and the high resolution texture pack. It offers to 'teach' fsr for your native resolution. Then it runs through and let's it programmatically look at the best solution on average in the game by running through the various assets. Once it is done you turn on fsr or whatever it is called and you get quality and performance customized to your hardware. Some games like forza horizon 4 do something similar.
 
Honestly, I'd rather just have faster hardware than rely on either. Isn't that what we would do in the old days? Just buy a faster video card. Or in the case of Crysis, wait for that video card to get created.
 
The problem is Ray Tracing now. It's never going to be fast enough, especially when a game is fully ray traced or path traced, to work without some sort of method like upscaling to improve performance. The cart is running ahead of the horse on this particular feature.

I had a nasty thought, imagine not even rendering locally anymore, just streaming from a cloud AI your rendered game. Maybe the future isn't even local rendering AT ALL. A "video card" that is nothing but AI rendering cores to stream in data and AI it to your screen. Now that's a scary thought.
 
You must log in or register to reply here.
Become a Patron!
Back
Top