Originally posted by piotrj3
View Post
Announcement
Collapse
No announcement yet.
HITMAN 3 Runs Well On Linux With Steam Play - Open-Source Radeon Performance Especially Good
Collapse
X
-
Originally posted by shmerl View Post
DLSS is the current version of such trick. Or upscaling in general.
Also about DLSS - DLSS shouldn't be here compared to classic upscaling as DLSS trades a bit lower internal resolution, but gets additional information from previous frame, motion vectors of objects and generally trained neural network so DLSS gets more access to useful information then native (in terms of raw volume of data) just not always that data is useful and in best cases you have better image then native that removes artifacts/shimmering with perfect super samping AA and even reproduced details, while in worst case scenario you have AI artifacts with ghosting and only reason why it runs faster is because tensor cores can be extremly efficient as extremly specialized piece of silicone.Last edited by piotrj3; 24 January 2022, 09:24 PM.
- Likes 6
Comment
-
Originally posted by piotrj3 View Post
DLSS is not compared vs non-DLSS in performance so it is not trick to falsify benchmark scores. It is just feature (and very good feature) that i wish existed on AMD so people like you could expierience it and stop spreading ignorance. At least Intel understands it.
You seem to be one of those who think it's always useful. Not uncommon, given how much money Nvidia put into marketing for it. You don't need their marketing though to understand when upscaling can be useful and how it never can be the best option in general.
DLSS can be trained, but trained neural network is never perfect. That's why you have to mention "best cases". So to put it simply, any kind of such technology in general case means reducing image quality, there is no way around it. So you simply don't want to upscale ever, in the perfect set up you want your hardware to be able to cope with the resolution you choose. When you hardware is too weak, then yeah. Such kind of tricks can come handy.Last edited by shmerl; 24 January 2022, 09:32 PM.
- Likes 5
Comment
-
piotrj3 DLSS is POS in term of image quality. It's very visible in Assetto Corsa Competizione and maybe other UE4 games.
Now it seems AMD washes floor with nvcrap on Linux. The gap will be even bigger thanks to Valve and Open Source developers. Nvidia with their legacy development model can go to hell. Only nobrainer will buy nvidia for Linux.
- Likes 1
Comment
-
Originally posted by shmerl View Post
It's a trick in a sense that it reduces quality while making it appear "good enough". It is presented as a feature, so better than some hidden thing happening without user's consent. Whether it's useful - may be. Upscaling can be useful for lower end hardware cases.
You seem to be one of those who think it's always useful. Not uncommon, given how much money Nvidia put into marketing for it. You don't need their marketing though to understand when upscaling can be useful and how it never can be the best option in general.
DLSS can be trained, but trained neural network is never perfect. That's why you have to mention "best cases". So to put it simply, any kind of such technology in general case means reducing image quality, there is no way around it. So you simply don't want to upscale ever, in the perfect set up you want your hardware to be able to cope with the resolution you choose. When you hardware is too weak, then yeah. Such kind of tricks can come handy.
- Likes 2
Comment
-
Originally posted by partcyborg View Post
I would put money down on a wager that you would fail to correctly select 4k upscaled with dlss on quality from native 4k in a blind test more than 50% of the time (aka choosing randomly). I know i couldnt.
- Likes 1
Comment
-
Good image quality and good perceived UX of DLSS do not change the fact this tech was created to workaround the issue where current GPUs are too weak for RTRT.
Even though modern real time 3D rendering uses a lot of "cheats" to optimize performance, all of them used to be deployed locally on the PC. DLSS took it to the next level, where training is done on the farm, which effectively means that a lot of computations, required to produce final result of the rendered view are not even done locally on PC. That's the largest technical cheat ever made really, and it was done by NVIDIA.
Also, there is no doubt that NVIDIA is worse than AMD from a anticonsumerish economical point of view, simply because it hadls a near monopolistic major market share and all companies in such a situation try to exploit their market position. Intel does the same. AMD would do the same, but it, yet again, does not change the fact NVIDIA and Intel are at this position currently. So yeah.
Comment
-
Originally posted by shmerl View Post
It's a trick in a sense that it reduces quality while making it appear "good enough". It is presented as a feature, so better than some hidden thing happening without user's consent. Whether it's useful - may be. Upscaling can be useful for lower end hardware cases.
You seem to be one of those who think it's always useful. Not uncommon, given how much money Nvidia put into marketing for it. You don't need their marketing though to understand when upscaling can be useful and how it never can be the best option in general.
DLSS can be trained, but trained neural network is never perfect. That's why you have to mention "best cases". So to put it simply, any kind of such technology in general case means reducing image quality, there is no way around it. So you simply don't want to upscale ever, in the perfect set up you want your hardware to be able to cope with the resolution you choose. When you hardware is too weak, then yeah. Such kind of tricks can come handy.- DLSS does not reduce quality
- No one tests AMD's native res against NVIDIA's DLSS res, there's no cheating, everyone knows what it is
- Likes 2
Comment
Comment