NVIDIA DLSS (Deep Learning Super Sampling)
To put it simply, both technologies (DLSS & FSR) basically do the same thing. Depending on the selected mode, the game will be rendered in a lower resolution and at the end it will be rendered up to the desired monitor resolution or the resolution fixed in the game. Upscaling is the keyword here! So much in advance and in general. However, since DLSS and FSR differ in detail, I would like to explain the details very briefly and as simply as possible at this point. Especially for those readers who do not have the detailed knowledge regarding DLSS and FSR, this short digression should serve as an introduction to the topic. At the end of the day, this also makes it easier to classify the performance in terms of graphic details!
https://www.nvidia.com/en-us/geforce/news/june-2021-rtx-dlss-game-update/
NVIDIA goes all in on AI support! In particular, with respect to DLSS, NVIDIA makes use of a network to teach the AI. The graphical details and motion vectors must be taught to the DLSS algorithm in the first place. The learning of the AI ends, if one describes it simplistically, in a large data pool, where the learned images with up to 16K resolution are located. The AI network is fed or trained from this pool. What ends up back on your monitor later.
https://www.nvidia.com/en-us/geforce/news/june-2021-rtx-dlss-game-update/
By looking at motion vectors and the previous high-resolution output, DLSS can track objects from frame to frame, resulting in stable motion. It can even reduce flicker and pop artifacts. Well, looky here! This process is called “temporal feedback” because it uses the past to obtain information for the future. Give me the lottery numbers, that’s all I’m saying. Hehe! With access to previous frames and motion vectors, DLSS can track each pixel and take multiple samples of the same pixel across frames (known as temporal supersampling), resulting in better detail and edge quality than traditional upscaling solutions.
https://www.nvidia.com/en-us/geforce/news/june-2021-rtx-dlss-game-update/
The so-called tensor cores, which are additionally installed on every RTX graphics card, in a way take over the tasks to display the image in the desired resolution “upscaled” or to bring the details for the higher resolution as true to the original as possible on the screen. The tensor cores are just a part of the learning network. NVIDIA promises to relieve the actual shaders for image calculation and to generate even more performance. At this point it should be enough on the topic of DLSS and we’ll switch to the Red Team.
AMD FidelityFX Super Resolution (FSR)
Now, unfortunately, it will not be as detailed as on the subject of DLSS from NVIDIA. AMD is officially still keeping a very low profile on the extent to which FSR actually works. The fact is that an upscaling algorithm is also used with regard to FSR, which is supposed to be able to spatially reconstruct graphic details. That’s all I can tell you about this as of today, but as soon as I hear more details from AMD, I’ll keep you posted! So today we’re just looking at the usual PR slides.
AMD reports that FSR is very close to the native resolution. In addition, FSR is said to be significantly better than point or bilinear upscaling. Let’s keep looking.
A beautiful slide that, thankfully, says everything – but nothing at all. Let’s keep looking.
What about hardware support? Compared to NVIDIA DLSS much better!
There is not much more officially on the part of AMD on the subject of FSR. Therefore I don’t want to keep you in suspense with more PR slides. At the end of the day, it’s all about the performance that comes across to us users on the monitors at home. So let’s take a closer look at the measurements and graph details. Next page please!
48 Antworten
Kommentar
Lade neue Kommentare
Urgestein
Urgestein
Urgestein
Veteran
Veteran
Urgestein
Mitglied
1
Veteran
Neuling
Mitglied
Urgestein
Moderator
Moderator
Moderator
Moderator
1
Urgestein
Neuling
Alle Kommentare lesen unter igor´sLAB Community →