Gadgetory

we've been hearing quite a bit recently regarding DL SS which stands for deep learning super sampling there's a science to do thanks for learning with us okay but seriously why so much commotion what exactly about DL SS makes it so important for RTX graphics cards well it'll be easier to break this one down a component by component basis will open a minute science super sampling the second half of that acronym is just a fancy term for scaling a game to a resolution beyond what your monitor supports so for example in GTA 5 I can enable frame scaling that's just another term for super sampling and to run a resolution that is twice the density of my 1440p gaming panel in this case that's five thousand one hundred and twenty by twenty eight eighty pixels but why would I want to scale a resolution beyond what I'll actually end up seeing on screen right I can't change my screens resolution less like completely swap monitors so why would I do that a few reasons actually the biggest is because frame scaling can smooth out to a degree sharp edges similar to how anti-aliasing works so since pixels are squares around and diagonal edges can look choppy a bit like stairs and super sampling can in some cases smooth them out though I'd almost always advise enabling a a for a similar effect instead since it tends to have less of an impact on the average frame rate so that's what super sampling is I discuss it more in this video here but what about the deep learning part of DL SS why is that important and why is it necessary for ray tracing well to start all deep learning really means it's basically marketing jargon at this point because a lot of people use it that an AI is that something in the background is improving with time it's studying a pattern repetition it's trying to predict it better in the future and when seen in the context of Nvidia graphics cards deep learning describes the platform's ability to optimize and smooth out the experience in game it's done by running the game in question over and over and over again until the neural network learned so much about a games physics layout shadows composition that it actually improves the user experience with time the process actually involves NVIDIA zone data center so the more information that is fed to them the better optimized in theory the graphics card becomes with respect to that particular title and this is done on a game by game basis it takes a lot of time for these servers to study these games and predicting them better and the super sampling aspect of ELSS is a requirement for these data centers to dissect unique high-resolution scenes break them apart study them and optimize the most visually vital aspects of them so most games can load enough into vram to the point where calling a particular instance is near effortless and instantaneous it's why memory bandwidth matters and why you'll probably never hear it labeled as a true bottleneck and to the extent that memory isn't the bottleneck the GPU can for the most part solve the problems in question and push them in a fairly timely manner I mean if you had any delays that would reflect on your frame rate right it's why parallelization is emphasized in the graphics card market and that's typically why we have thousands of CUDA cores and stream processors occupying our Nvidia and AMD graphics cards but what if those GPUs become too bogged down with computational workloads what if the frame rate when enabling something like DXR drastically lowers sound familiar battlefield 5 when something as taxing as rate racing is turned on by a DX r in a game like battlefield 5 the graphics card in question needs every bit of help it can get yes even the RT X cards real time rate racing calculations to the extent that they aren't predicted in advance with AI are just too intense at this point and the lack of our key core optimization really doesn't help I mean how many games right now support the XR one so NVIDIA has been seeking to solve this problem with legit servers whose job it is to dissect these games and predict events far in advance that's easing the load on the GPU in a real-time and improving the user experience the tensor core is vital by the way for this implementation and that's why only our TX cards to date have access to this feature so in a nutshell DL SS frees up additional graphics card resources for additional rendering a byproduct of this is usually an increase in framerate and a bump and visual clarity which is why it's so important with something like DX R because DX R works the opposite by dropping your frame rate for those added reflections and more realistic just verbiage and game although a major downside of this is that to date only a handful of games about 25 or so actually plan to support DL SS so fans have been videos tearing architecture will need to keep their fingers crossed in hopes that developers catch on and implement these abilities of future tiles in fact I think the number of games that support both the X R and D LSS or that even will in the coming months or years is 5 I want to say it's like five games so yeah they better catch on soon with that if you'd like to contribute to the channel become a member you can do so by clicking the blue icon down below I do appreciate that you can buy merch similar to a shirt I wore in the last video you can subscribe a feeling super fancy and I'll catch you in the next one this is science studio thanks for learning with us