I have a GTX 1650 Mobile and was wondering how OptiScaler handles DLSS on hardware without RTX Tensor cores. The tool lets me select DLSS as both the input and output upscaler, and the result appears sharper than some of the other options. Is it actually recreating DLSS with the GPU's regular shader or CUDA-capable cores, or is it redirecting the game's upscaling request to another technique under the hood? I'd appreciate a detailed explanation of what those input and output settings mean.
2 Answers
OptiScaler can intercept the upscaler call that a game makes and redirect it to a different backend. On a GTX 1650, selecting DLSS in the interface does not magically add Tensor cores or run native DLSS. It is more likely using a compatible shader-based implementation or translating the request to another supported upscaling method while presenting the result through the DLSS slot. In other words, it may resemble DLSS from the game’s point of view, but the actual work is being performed by regular shader cores rather than dedicated AI hardware.
A GTX 1650 cannot run genuine NVIDIA DLSS because DLSS relies on Tensor cores, which are only found in RTX GPUs. It can use alternatives such as FSR or XeSS, depending on the game and the tool’s support. Those run as compute workloads on the normal shader cores, so they can provide a similar general idea—reconstructing a higher-resolution image—but they are not the actual DLSS algorithm running on Tensor hardware.
That makes sense for FSR and XeSS, and I can use both of those. What confused me is that OptiScaler also lets me choose DLSS for both input and output, and it seems sharper than the other options.
So the DLSS option is effectively a compatibility or replacement path in this case—something formed with compute shaders rather than genuine Tensor-core DLSS. That explains why it can still produce a sharper image without the card being RTX.