My 12GB RTX 4070 Ti runs multiple DLSS 5 layers thanks to one OptiScaler setting - XDA

Nvidia's DLSS 5 neural rendering is officially reserved for the company's new Blackwell 50-series GPUs, but a software tweak is allowing enthusiasts to bypass this restriction on older hardware.

Samarveer Singh, an author at XDA, has successfully run multiple layers of DLSS 5 on his RTX 4070 Ti by using the open-source tool OptiScaler. This method effectively turns the 12GB card into a more powerful machine for AI-driven upscaling.

The core issue with DLSS 5 is its high memory consumption. Neural rendering passes require significant VRAM, and running multiple layers can quickly exhaust the 12GB buffer on the RTX 4070 Ti, often leading to crashes.

Spoofing Blackwell to unlock features

OptiScaler works by reporting the RTX 4070 Ti as an RTX 50-series GPU to the game client. This spoofing unlocks multi-frame generation and the ability to inject DLSS 5 neural rendering files into titles that would otherwise block them.

Once the software is injected and the correct DLSS files are placed in the game directory, the user can launch a title and experiment with different configurations. The goal is to maximize the number of neural rendering passes without triggering a VRAM crash.

The tool provides a flexible pipeline for the rendering process. Users can choose to apply the neural rendering pass before or after Super Resolution upscaling, a setting that drastically changes the performance and memory footprint.

Reordering the pipeline for better performance

The most critical discovery is the 'NR before SR' option, which reorders the rendering steps. Instead of processing the neural pass on a full 1440p or 4K image, the tool allows the neural rendering to operate on a lower internal resolution before upscaling.

This optimization reduces the workload for each neural pass, allowing the GPU to handle more layers than it otherwise could. The result is a significant increase in frame rates, particularly when using multiple neural rendering passes.

Samarveer tested this setup in Elden Ring, comparing configurations with neural rendering applied after Super Resolution versus before it. The difference was stark, with the 'before SR' method delivering nearly double the frame rate in some scenarios.

Experimental results

The data from Samarveer's tests highlights the efficiency of the reordering technique. Running a single DLSS 5 neural pass after Super Resolution resulted in 71 FPS and 10.1 GB of VRAM usage.

When the neural pass was moved to before Super Resolution, the frame rate jumped to 98 FPS with only 9.8 GB of VRAM used. This configuration proved to be the most stable for a single layer.

For multiple layers, the benefits were even more pronounced. Running two neural passes after Super Resolution dropped the frame rate to 52 FPS, while the 'before SR' configuration recovered to 72 FPS with 10.5 GB of VRAM.

The results show that while the 12GB VRAM limit is still a hard constraint, the OptiScaler pipeline optimization allows gamers to push the RTX 4070 Ti much further than the official specifications suggest.