Working more on the Checkerboad rendering vs. DLSS 2.0 video today for Death Stranding and I think it is time to write an open letter to about DLSS 2.0 as a quality of life feature on PC that Devs should consider if they use TAA already. Devs please read! 🙂 (1/11 Thread) 
TAA has been a great compromise for quality and speed since graphics engines are full of specular aliasing, aliasing in motion, and MSAA and SSAA are too expensive or complex to integrate. Single frame post-process AA (FXAA/SMAA) is just too unstable over time. (2/11)
With DLSS 2.0 we are essentially looking at a more perfect TAA where it is not showing the same characteristic full-frame faults that usual accumulation TAA has, or even 2 frame TAA has like SMAA 2Tx. It is doing temporal reprojection and rejection so much better that (3/11)
it has a number of elements that look visually more stable over time or more detailed than native resolution with the standard TAA shipping with games. It does it better for many cases while also being rendered at an internal resolution that is much lower (1440p, or 1080p) (4/11)
The nature of the AI driving it is a black box and not running through Direct ML. It also only runs on HW that has Tensor Cores (RTX atm). So I understand the trepedation of not want to utilise solutions you have less control over, also it is a matter of pride for some (5/11)
I think PC developers should look at DLSS like one would look at new hardware accelleration features - like MSAA of old. MSAA also was once new, required HW support, and lacked programability: but it had a massive benefit for performance and visual quality for many users. (6/11)
In this case the visual/performance benefits have reached a tipping point where not utilising DLSS 2.0 is setting a game's visual performance ratio at a disadvantage. An RTX 2060, the cheapest RTX GPU, can run Death Stranding at 4K with DLSS in performance mode at 60 fps. (7/11)
It is enabling a visual and performance experience for mid to low end users which is frankly incredible. It is doing more with less which is rare enough as is. Yet the userbase for DLSS is not exactly small - something like 15 million RTX devices out there that is only... (8/11)
Going to increase over time as Ampere enables DLSS for even lower end devices presumably and increases the market base. As I see it, if you are spending the time to integrate perf intensive "Ultra" settings in games which cater to a select few with more minimal visual gains(9/11)
then I think looking into utilising DLSS 2.0 is on a similar wavelength, except it has dramatic visual and performance considerations for a larger base. Yes, AMD does not support it and it will most likely never be the case until they have a competing GPUopen equivalent (10/11)
But in the mean time, DLSS 2.0 looks like low hanging fruit to enable a graphical and performance experience for an ever-growing large group of users that IMO is too good to be ignored. (11/11)
