4k vs reconstructed 4k/1440p/dynamic res

I think because of VRR support, we’ll see a lot of games at 60fps most of the time, when we see render time larger than 16ms, then use dynamic resolution to make time up in future frame.

My preference starts with frame rate and motion handling. Some games give me SCREAMING headaches on console but PC didn’t. One example with a well known game is the Witcher 3. Hadn’t made it past the prolog but the background on camera movement made me feel I was going cross-eyed. X enhancement helped in performance mode.

After frames and motion, I prefer detail. Quality character animations. Things like flapping curtains or tents. Particle effects… Lighting. Swaying vegetation, or other “immersive” effect.

After that, I like image quality, but resolution and sharpness don’t always capture the quality base on the style and atmosphere.

Wow thats a gamechanger.

1 Like

Narratives change. Goal posts moved.

More people will always sounds the loudest.

1 Like

Goalposts are being launched into the sun.

1 Like

There are better uses of ML than DLSS-style full screen reconstruction imho. Using ML to upres individual textures at runtime is a much bigger deal.

Are DLSS and DirectML similar competing technologies? First I’m hearing much of DirectML.


Power crown lost…Every negative turned as positive and vice versa you know


If a game has a locked framerate => use dynamic resolution to keep the framerate stable.

If a game supports high framerates and VRR => use whatever keeps the framerate above the VRR threshold.

No they are not the same. DirectML is Microsoft Direct X hardware abstraction API for running machine learning models on different GPUs or CPUs, like Direct3D for rendering. You can run DLSS models through it, like Microsoft demonstrated with Forza Horizon 3 at Siggraph 2018.


Not sure what you mean by competing technologies.

AI upscaling on XSX is most of the time termed as DirectML.

So yes, as per resolution upscaling technique. DLSS and DirectML are competing tech.

DLSS is not proprietary tech. That’s just the name Nvidia gave it since it is descriptive. Just like RT, GI, SSAO are all short hand descriptors of tech approaches but not tech implementations. There are many ways to approach doing RT, GI, etc. Same is true for DLSS.

DLSS on XSX would use DirectML and the ML hardware on XSX, but the ML stuff can also do lots of other stuff. Physics sims, texture upres, HDR, help with VRS, systems simulation, animation smoothing, framerate interpolation, cloth physics, fluid physics, enemy/npc AI…tons of stuff. The potential is in its infancy right now.

To answer the OP, 1440p with effects and framerate. And no, DLSS is a hardware technique, while DirectML is basically all in software. AMD has nothing approaching Nvidia in this area right now

No, DirectML is a software API for abstracting hardware specific stuff like NVidia Tensor Cores away from applications running machine learning models. DLSS is a machine learning model which can be run on NVidias own API or DirectML. Microsoft demonstrated this at Siggraph 2018 where they did indeed run a Super Resolution model from NVidia with DirectML.


Thanks for clarification!

1 Like