I'm running an undervolted RTX 3080 Ti with an i5-14600KF and 32GB of DDR4 on a Seasonic 850W 80+ Gold power supply. I also have a spare RTX 3060 12GB and a Hyper Pro 700W 80+ power supply. Seasonic's calculator estimates roughly 916W if both GPUs are used together. I'd like to use the 3080 Ti for gaming and the 3060 for frame-generation or neural-rendering features, possibly with a DLSS-related mod. Is it safe or practical to power the second GPU from the separate 700W unit while it remains installed in the same motherboard?
4 Answers
Technically, a dual-PSU setup can power the extra card, but both supplies need to start and shut down together. You would need a proper dual-PSU synchronizer, and the motherboard’s slot power also has to come from the same system arrangement. Do not rely on manually switching the second PSU on and off.
A second GPU usually won’t help with normal gaming, SLI, or frame generation. Modern games generally don’t split those workloads across two unrelated GPUs, and your 3080 Ti should already be capable of handling frame generation by itself. This setup is more likely to become a compatibility and latency-tuning project than a straightforward performance upgrade.
The sensible option is to replace the current setup with a quality 1000W-or-larger PSU that has enough PCIe power connectors, then sell or repurpose the spare parts. Alternatively, keep the 3060 for a separate secondary system or use it for a dedicated encoding task. Running both cards in one case is unlikely to provide the frame-generation benefit you’re expecting.
The Seasonic 850W is the only unit in this pair I’d consider trusting for a demanding gaming system. The Hyper Pro 700W is not a good candidate for powering an expensive GPU unless its exact model and protections are verified; an 80+ label only describes efficiency, not build quality or safety. I would avoid putting the 3080 Ti or 3060 on it.
The 850W unit is a Seasonic 80+ Gold model, but I’m worried it may be borderline because this is for a heavily modded Skyrim setup using DLSS-related neural rendering.