Modern SLI? Modder runs DLSS 5 across two RTX 5060 Ti GPUs, with one dedicated to Neural Rendering

Dual-GPU DLSS 5 mod separates game rendering and Neural Rendering © Marcelo Guibout This is probably not the best mod given how expensive modern GPUs are, but if you just so happen to own two relatively high-end NVIDIA cards, then perhaps this mod is for you. A new experimental mod can move DLSS 5 Neural Rendering processing away from the GPU rendering the game. Developer Marcelo Guibout demonstrated the setup in Cyberpunk 2077 using two GeForce RTX 5060 Ti 16GB graphics cards. One RTX 5060 Ti renders Cyberpunk 2077 while the second card handles the neural post-processing stage. Guibout says this is not an NVIDIA feature and it does not work like SLI. It doesn't actually separate GPU work across two GPUs, on the contrary, one works simply works on the output generated by the first card. The project is implemented as a public ReShade add-on. One GPU renders, the other handles DLSS 5 The approach works because the Neural Rendering pass operates on an already completed frame. The finished frame is copied to the second GPU, which performs the neural workload and displays the processed result. This avoids sending the finished image back to the primary GPU. Marcelo Guibout: Cyberpunk 2077 with DLSS 5 Neural Rendering Offloaded to a Second GPU — Dual RTX 5060 Ti 2,917 views Guibout says moving the neural workload away from the rendering GPU frees resources that DLSS 5 would otherwise consume on the same card. The demonstrated system used both RTX 5060 Ti cards through CPU-connected PCIe 5.0 x8 links, with one display connected to each GPU. A similar method is already used by Lossless Scaling users, who dedicate one GPU to actual rendering and a secondary GPU to handling multi-frame generation. So is splitting workloads across multiple GPUs our new future? Marcelo Guibout: Multi-GPU DLSS Neural Rendering — Dual RTX 5060 Ti (Dawnwalker, Cinematic) 3,731 views We all know what happened to SLI, CrossFire and even DirectX 12 Multi-Adapter. None of these technologies are widely used anymore, and companies have stopped releasing dual-GPU graphics cards. On the other hand, multi-GPU setups remain common in workstations for 3D rendering and, of course, AI workloads. Source: Marcelo Guibout