DLSS 5 mod lets you have a dedicated GPU for neural rendering, so you don’t have to tank your performance A modder named Marcelo Guibout has created a mod that offloads Nvidia's DLSS 5 neural rendering to a dedicated second GPU, using two RTX 5060 Ti 16GB cards and a ReShade add-on called MGPU Bridge, allowing gamers to avoid the performance hit of running neural rendering on the main render card. Benchmarks in The Blood of Dawnwalker show that with DLSS 5 on the second card, frame rates match or closely approach the baseline (e.g., 67-70 FPS at DLAA, 91 FPS at Quality, 106-107 FPS at Performance, 157 FPS at Ultra Performance), compared to significantly lower FPS when DLSS 5 runs on the render card (44, 54-55, 59, 69-71 respectively). DLSS 5 mod lets you have a dedicated GPU for neural rendering, so you don’t have to tank your performance Read More https://www.pcguide.com/earnings-disclaimer/ Table of Contents With DLSS 5 officially available https://www.pcguide.com/news/dlss-5-is-officially-being-released-this-week-starting-with-nba-2k27-days-after-a-leaked-build-surfaced-online/ and unofficially via fan-made mod tools https://www.pcguide.com/news/dlss-5-manager-adds-support-for-rdna-4-graphics-cards-with-new-amd-mode-in-latest-update/ , gamers have been testing out its new neural rendering feature on all kinds of setups. Here’s the latest we’ve come across: a modder has neural rendering running on a dedicated GPU. Nvidia currently only supports current-gen RTX 50 Series GPUs, though it says proper RTX 40 support is coming in the future. In any case, the modder, Marcelo Guibout, has managed to offload the performance-taxing https://www.pcguide.com/news/dlss-5-benchmarks-reveal-up-to-60-performance-drop-but-its-already-5x-better-than-nvidias-original-reveal/ DLSS 5 neural rendering feature to a second GPU. He’s using two RTX 5060 Ti 16GB graphics cards to get the job done, handling game rendering on the main card and neural rendering on the second, while using a ReShade add-on known as MGPU Bridge https://github.com/maohgad-web/Neural-coprocessor to transfer over the DLSS 5 visuals. DLSS 5 neural rendering offloaded to a second GPU Back when Nvidia first showcased DLSS 5 https://www.pcguide.com/news/dlss-5-release-date-everything-you-need-to-know/ earlier this year, it was using two flagship RTX 5090 cards to do so, with one of them running the game as normal, and the other dedicated to neural rendering. Between then and launch, the company has optimized the feature https://www.nvidia.com/en-gb/geforce/news/dlss-5-3d-guided-neural-rendering/ so it runs around 5x faster and is suitable for a single GPU setup . But what if you want to go the double GPU route? Latest PC & Tech deals ASUS ROG Strix 34” QD-OLED XG34WCDG - was $7499 now $629 Hisense 75" U8 Series ULED Mini-LED 4K TV - was $1,499 now $1,354 ASUS ROG XG27AQNGV- was $999 now $649 TCL 98 Inch Class QM8L - was $2,499 now $2,309 CyberPowerPC Gaming PC RTX 5060 - was $599 now $459 Prices correct as of August 20th, 2026. Well, with Guibout’s mod, you can enjoy much better performance using this method. It’s important to point out that dual-GPU setups are a thing of the past after Nvidia fully cut support for physical SLI and NVLink bridges with the RTX 40 series onwards. As such, neither technology is being utilized with this mod. “Neural rendering is a terminal stage. It takes a finished frame and returns a finished frame, so it can be picked up and executed somewhere else entirely,” the modder explains. So far, Guibout has demonstrated the technique in two games: The Blood of Dawnwalker and Cyberpunk 2077. Benchmarks for Blood of Dawnwalker at 1080p below detail the difference in performance. When rendering at full resolution with DLAA, there’s no performance cost on the second card, and the gap increasingly widens as DLSS’s upscaling modes get stronger. | DLSS mode | FPS DLSS 5 NR off | FPS DLSS 5 on the render card | FPS DLSS 5 on the second card | |---|---|---|---| | DLAA | 67-70 | 44 | 67-70 | | Quality | 98-99 | 54-55 | 91 | | Performance | 127-131 | 59 | 106-107 | | Ultra Performance | 172 | 69-71 | 157 | GitHub https://github.com/maohgad-web/Neural-coprocessor We’ve seen similar double-GPU setups like this before, back when people were using a dedicated card for 32-bit PhysX support https://www.pcguide.com/news/3-year-old-rtx-3050-saves-the-rtx-5090-in-these-32-bit-physx-benchmarks/ . For context, the RTX 50 series launched without support, but Nvidia did end up adding it with a driver update. Another use case is with the popular Lossless Scaling tool https://steamcommunity.com/app/993090/discussions/0/594016978256007124/ , where gamers are offloading the upscaling & frame gen workload for even more FPS.