build-in fidelityFX cas VS FSR（loss scaling） @ monsterhunter word

[haloduring]

fellas I got a 6800xt running whw @ 4k around 50 to 60.fps It is not stable enough. so i tried these two upscalers: cas and fsr (3200x1600 --> 3820x2160) but I wonder if CAS had higer native resolution. higher native resolution means better result right?

[ogioto]

Try 1449p->4k using LS1 with 0 or 1 sharpnes (you can put more but that should be good initially)

[haloduring]

Diposting pertama kali oleh ogioto:

Try 1449p->4k using LS1 with 0 or 1 sharpnes (you can put more but that should be good initially)

so which one offers better results？ FSR/LS1 seems to be implemented with a better algorithm, but Cas is built-in

[UnreasonablySaucyConcrete]

CAS is just an image sharpener apparently. FSR has CAS within it so it both upscales and sharpens.

LS1 would be better than both FSR and CAS because it's a tailor made spacial upscaler attempting to be better than FSR. So basically just use LS1 if you can, it'll give you more performance and in most cases better image quality too.

[kripcision]

Diposting pertama kali oleh haloduring:

Diposting pertama kali oleh ogioto:

Try 1449p->4k using LS1 with 0 or 1 sharpnes (you can put more but that should be good initially)

so which one offers better results？ FSR/LS1 seems to be implemented with a better algorithm, but Cas is built-in

thats something you can decide for yourself.

[ogioto]

Diposting pertama kali oleh haloduring:

Diposting pertama kali oleh ogioto:

Try 1449p->4k using LS1 with 0 or 1 sharpnes (you can put more but that should be good initially)

so which one offers better results？ FSR/LS1 seems to be implemented with a better algorithm, but Cas is built-in

CAS stands for Contrast Adaptive Sharpening, I would recommend LS1 as it is machine learning based spatial upscaler

[haloduring]

Diposting pertama kali oleh UnreasonablySaucyConcrete:

CAS is just an image sharpener apparently. FSR has CAS within it so it both upscales and sharpens.

LS1 would be better than both FSR and CAS because it's a tailor made spacial upscaler attempting to be better than FSR. So basically just use LS1 if you can, it'll give you more performance and in most cases better image quality too.

Diposting pertama kali oleh kripcision:

Diposting pertama kali oleh haloduring:

so which one offers better results？ FSR/LS1 seems to be implemented with a better algorithm, but Cas is built-in

thats something you can decide for yourself.

thank you guys. but to be frankly, sometimes I can't tell which works better. Is there anyways to check internal resolution whlie the game is running ? like a debuger or something So I can choose the higher one?

[haloduring]

Diposting pertama kali oleh ogioto:

Diposting pertama kali oleh haloduring:

so which one offers better results？ FSR/LS1 seems to be implemented with a better algorithm, but Cas is built-in

CAS stands for Contrast Adaptive Sharpening, I would recommend LS1 as it is machine learning based spatial upscaler

Yes and is based on lanzos maybe according to AMD document .
Nvidia uses DGX supercomputer to train DLSS models Maybe LS1 doesn't get enough training data（I'm disgusted by Nvidia's monopolistic behavior, and here is just an example）:D

[ogioto]

Diposting pertama kali oleh haloduring:

Diposting pertama kali oleh ogioto:

CAS stands for Contrast Adaptive Sharpening, I would recommend LS1 as it is machine learning based spatial upscaler

Yes and is based on lanzos maybe according to AMD document .
Nvidia uses DGX supercomputer to train DLSS models Maybe LS1 doesn't get enough training data（I'm disgusted by Nvidia's monopolistic behavior, and here is just an example）:D

LS1 had a perfect amount of training data, it is just spatial solution unlike DLSS which is temporal. LS1 has almost maxed the limit of spatial upscaler with machine learning. It is the best solution compared to the other spatial upscalers.

[haloduring]

Diposting pertama kali oleh ogioto:

Diposting pertama kali oleh haloduring:

Yes and is based on lanzos maybe according to AMD document .
Nvidia uses DGX supercomputer to train DLSS models Maybe LS1 doesn't get enough training data（I'm disgusted by Nvidia's monopolistic behavior, and here is just an example）:D

LS1 had a perfect amount of training data, it is just spatial solution unlike DLSS which is temporal. LS1 has almost maxed the limit of spatial upscaler with machine learning. It is the best solution compared to the other spatial upscalers.

ok i'll try it

[haloduring]

Diposting pertama kali oleh ogioto:

Diposting pertama kali oleh haloduring:

Yes and is based on lanzos maybe according to AMD document .
Nvidia uses DGX supercomputer to train DLSS models Maybe LS1 doesn't get enough training data（I'm disgusted by Nvidia's monopolistic behavior, and here is just an example）:D

LS1 had a perfect amount of training data, it is just spatial solution unlike DLSS which is temporal. LS1 has almost maxed the limit of spatial upscaler with machine learning. It is the best solution compared to the other spatial upscalers.

sometimes the upscaled one is way better than native! :D it really surprises me

[ogioto]

Diposting pertama kali oleh haloduring:

Diposting pertama kali oleh ogioto:

LS1 had a perfect amount of training data, it is just spatial solution unlike DLSS which is temporal. LS1 has almost maxed the limit of spatial upscaler with machine learning. It is the best solution compared to the other spatial upscalers.

sometimes the upscaled one is way better than native! :D it really surprises me

I'm sure the dev will be happy to see that :)
For more help/question, you can join the discord server of the app for easier/faster communication.