I currently have a Gigabyte RTX 5080 and an older Asus ROG Strix RTX 3070 Ti. I'm interested in trying DLSS 5-style neural rendering in some older games, possibly through Lossless Scaling or a ReShade-based solution. Since the 5080 is much faster but the 3070 Ti is still available, could I use one GPU to render the game and the other to handle frame generation or neural processing? Which card should do each task, and would this setup actually improve performance without losing frames to GPU-to-GPU overhead? I'm also wondering whether PCIe lane configuration, cooling, or driver compatibility would make a major difference.
5 Answers
The practical recommendation is to use the RTX 5080 for DLSS 5 or other neural processing and only use the 3070 Ti for game rendering if you’re specifically testing an experimental setup. For ordinary gaming, the 5080 should handle everything by itself. Also check your motherboard and CPU PCIe lane layout, since the second card may run with limited bandwidth, and make sure your case and power supply can handle the heat and power draw of both GPUs.
A second GPU can sometimes handle frame-generation calculations through Lossless Scaling, but the results may not be impressive. Lossless Scaling does not have the same motion-vector information that native DLSS has, so image quality and artifact control can be worse. Since the RTX 5080 already supports advanced frame-generation features, adding the 3070 Ti is likely to create more complexity than useful performance.
Be careful with the RTX 3070 Ti as the processing card. Current dual-GPU DLSS 5 experiments have mainly been tested on RTX 40- and 50-series cards. Ampere lacks native FP8 matrix support, which DLSS 5 relies on heavily, and early attempts on RTX 30-series hardware have shown extremely poor performance. The 3070 Ti may technically work in some configurations, but it could end up taking seconds to produce a frame instead of improving frame rate.
If you want to experiment with a dual-GPU setup, the most promising arrangement would probably be having the RTX 3070 Ti render the game while the RTX 5080 handles the neural post-processing. There are early ReShade-based projects attempting something similar, but they still require manual setup and troubleshooting. Lossless Scaling may not integrate cleanly with that workflow, so there probably isn’t a simple plug-and-play configuration yet.
Two GPUs can be installed successfully, but cooling is a serious concern. With both cards under load, the system can become much hotter and louder, especially if they are close together. It may be worth testing for curiosity, but unless a specific mod supports your hardware well, the extra power use, PCIe overhead, and troubleshooting could outweigh any performance gain.
That’s exactly what I’m trying to determine. The hardware can be installed, but I don’t want to run both cards if the added complexity gives worse performance than simply using the 5080.