• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

[DF] Inside FSR 2.0 – The Making of FSR 2.0, AMD's Future Upgrades + the Xbox Connection

DenchDeckard

Gold Member
I honestly think we won't see many FSR2.0 games, most devs will continue to use either Unreal Engine's TSR or TAAU... or their own engine's up-sampling which are usually either on par with FSR2.0 or better.

interestingly Ghostwire Tokyo didn't use TSR on PS5 even tho it has TSR implementation on PC...
so render time could also be a factor. it could be that it simply is too taxing to use and devs will opt to just use a simple upscaler like Ghostwire Tokyo did on PS5

that would be a shame if the tech improves and is genuinely a good option. You don't think we will see it now it's a apart of Xbox XDK? or GDK or whatever the latest one is called lol
 

01011001

Gold Member
that would be a shame if the tech improves and is genuinely a good option. You don't think we will see it now it's a apart of Xbox XDK? or GDK or whatever the latest one is called lol

I mean maybe, but not often. Call of Duty and other games/engines with established up-sampling tech will continue using their own upscaler, UE4 titles might just use TAAU to save resources and some will not up-sample at all outside of a simple upscaler.

I wonder tho if Ghostwire didn't use TSR for performance or aesthetic reasons. that could be another thing that developers will look at I think. some might not want the disocclusion artifacts that these have.

I think it will be mostly used by smaller devs and games that do not have other good up-sampling methods in the engine they use
 

Black_Stride

do not tempt fate do not contrain Wonder Woman's thighs do not do not
that would be a shame if the tech improves and is genuinely a good option. You don't think we will see it now it's a apart of Xbox XDK? or GDK or whatever the latest one is called lol
It absolutely will be used.
It has a ~2ms impact on render times with an XSX, thats phenomenal.
Will big developers be using it, probably not because they have their own implementations.
But those that dont or have an implementation thats more costly on the render budget will jump on a middleware thats proven and pretty widely tested that gets the job done.

Unreal Engine 5 devs probably wont cuz TSR works so well already, its likely to only get better.
But FSR could still get usage because of how open it is for devs to make it bespoke to their game.

On PC I expect it to be ubiquitous now with any competent release.
Pascal gets to live on!

I mean maybe, but not often. Call of Duty and other games/engines with established up-sampling tech will continue using their own upscaler, UE4 titles might just use TAAU to save resources and some will not up-sample at all outside of a simple upscaler.

I wonder tho if Ghostwire didn't use TSR for performance or aesthetic reasons. that could be another thing that developers will look at I think. some might not want the disocclusion artifacts that these have.

I think it will be mostly used by smaller devs and games that do not have other good up-sampling methods in the engine they use
GhostWire Tokyo was likely already in a gold or near gold state prior to the TSR stable release build.
Yes, they could have cvar activated it on PS5 but then they would have to reprofile the game for stability.
On PC you reveal the option if client computer cant handle it, they can turn it off.
If the PS5 versions rendertimes are suddenly higher or it causes issue in this one area of the game, it could take them weeks to fix.
Better ship whats already "stable" and call it a day.
 

01011001

Gold Member
GhostWire Tokyo was likely already in a gold or near gold state prior to the TSR stable release build.
Yes, they could have cvar activated it on PS5 but then they would have to reprofile the game for stability.
On PC you reveal the option if client computer cant handle it, they can turn it off.
If the PS5 versions rendertimes are suddenly higher or it causes issue in this one area of the game, it could take them weeks to fix.
Better ship whats already "stable" and call it a day.

true, but it's also curious that Fortnite doesn't get an update on console to use TSR... that game is Epic's posterboy for Unreal Engine and the first ever UE5 game on the market. it has a 120fps mode on console, and that one runs at a pretty low resolution... so TSR could do wonders here, or FSR2.0 of course
 

Black_Stride

do not tempt fate do not contrain Wonder Woman's thighs do not do not
true, but it's also curious that Fortnite doesn't get an update on console to use TSR... that game is Epic's posterboy for Unreal Engine and the first ever UE5 game on the market. it has a 120fps mode on console, and that one runs at a pretty low resolution... so TSR could do wonders here, or FSR2.0 of course
TSR might not be as "cost free" on console as it is on PC.
 

01011001

Gold Member
TSR might not be as "cost free" on console as it is on PC.

yeah and I wonder if that will be true for FSR 2.0 as well, and if it is then I think we won't see it much.
we already know it will not work at all on the Switch given how much of a performance impact it has when used on the Steam Deck. so I guess it's wait and see
 
Last edited:

Black_Stride

do not tempt fate do not contrain Wonder Woman's thighs do not do not
yeah and I wonder if that will be true for FSR 2.0 as well, and if it is then I think we won't see it much.
we already know it will not work at all on the Switch given how much of a performance impact it has when used on the Steam Deck. so I guess it's wait and see
He said they have it at ~2ms on XSX.
I imagine it has a similar rendertime cost on PS5.
Thats a very low render cost for what you get out of it.
 

01011001

Gold Member
He said they have it at ~2ms on XSX.
I imagine it has a similar rendertime cost on PS5.
Thats a very low render cost for what you get out of it.

sure, but then it becomes a thing where devs will have to decide if they rather push for slightly higher settings and/or if they are fine with the artifacts.
some may chose a lower output res with better motion clarity, some might use the 2ms of render time for other things.

I also assume that render time goes up the lower the internal resolution is compared to the targeted up-sampling resolution, which will be another factor
 

Black_Stride

do not tempt fate do not contrain Wonder Woman's thighs do not do not
sure, but then it becomes a thing where devs will have to decide if they rather push for slightly higher settings and/or if they are fine with the artifacts.
some may chose a lower output res with better motion clarity, some might use the 2ms of render time for other things.

I also assume that render time goes up the lower the internal resolution is compared to the targeted up-sampling resolution, which will be another factor
FSR gets faster the lower your base resolution is.
Im guessing they use Balanced (1.7x res bump) as the "baseline" target for when they say its at ~2ms.
Which makes the current impact similar to what a 5700XT does when targeting a 4K resolution output.
They can target Performance which is a 2 x res bump which cuts 0.2ms from the impact.
The good thing about GPUOpen is we have all the documentation and all the code so we can make some pretty accurate albeit speculative assumptions.

For the Series S if its targeting 1440p or 1080p the impact is practically imperceptible we are talking in the 1ms range even on things like the Vega56.
So unless TSR can get its act together I wouldnt be shocked if pretty much every dev struggling to hit 1440p/1080p just uses FSR to get up there without stressing the GPU too much....those memory constraints well....theyll figure it out.
 

01011001

Gold Member
FSR gets faster the lower your base resolution is.

wouldn't it take longer the bigger the gap between the input res and the output res is? FSR would need to fill in more gaps in such a scenario no?
I haven't looked much into FSR but that intuitively makes sense to me at least...
 

Black_Stride

do not tempt fate do not contrain Wonder Woman's thighs do not do not
wouldn't it take longer the bigger the gap between the input res and the output res is? FSR would need to fill in more gaps in such a scenario no?
I haven't looked much into FSR but that intuitively makes sense to me at least...
Nope the opposite is true, the lower the base res the faster it gets.
i havent scoured the whole code, but im guess its due to the algorithm filling in more information itself since 1080 to whatever it only needs to read 2073600 pixels then can get to work, starting from a base resolution of 1440p it needs to read 3686400 pixels before it can do its magic.
 
Top Bottom