In the age of neural network graphics, enthusiasts have discovered unexpected and ingenious uses for multi-GPU configurations. The modder successfully achieved a modern emulation of SLI technology by combining two GeForce RTX 5060 Ti graphics cards into one system and completely isolating the DLSS 5 compute load on the second graphics accelerator.
The main problem with NVIDIA’s new neural rendering technology remains a huge hit to basic performance – even on the Blackwell architecture, the calculation of the neural network’s diffusion filter takes up precious frame time of classic rasterizers and ray units. The modder’s unusual setup solved this problem on the hardware.
Unlike classic alternating frame rendering, which was notorious for micro-freezing and uneven frame times, the card here works asynchronously. The main RTX 5060 Ti is responsible for standard geometry rendering, shaders and ray tracing, and the second card specifically handles the inference of the DLSS-NR neural network through a modified Streamline pipeline. The traditional halving of frame rates when turning on DLSS 5 is actually mitigated – the first card produces its native frame rate without slowing down, while the second card dynamically adds detail to the image.
