Which used GPU is best for 4K high-FPS gaming and local AI: RX 9070 XT, RX 7900 XTX, or RTX 5080?

For gaming value and local AI capacity, I’d lean toward the RX 7900 XTX. It is generally faster than the RX 9070 XT in conventional rasterized games and its 24 GB of VRAM gives it more room for larger models and higher-resolution workloads. The RTX 5080 is a poor value only if its price is much higher, but at €850 it becomes a serious option. The main reasons to choose it are CUDA and the better-supported NVIDIA AI ecosystem, plus stronger ray tracing, upscaling, and frame generation. The 9070 XT is the sensible budget choice, but the XTX seems like the better compromise here.

The 7900 XTX can handle models around the 35B range with suitable quantization and is still an excellent rasterization card. Its extra VRAM matters more than a small gaming benchmark difference because models that fit entirely in VRAM are much more convenient than ones that spill into system memory. However, AMD’s ROCm support can require more setup and is less broadly compatible than CUDA. Check the exact software and model runtimes you plan to use before assuming the XTX will be plug-and-play.

At €850, the RTX 5080 is probably the strongest all-rounder if gaming is the priority. It offers the best ray-tracing performance and the most mature support for CUDA-based AI tools, while DLSS and frame generation are useful for demanding 4K titles. Sixteen gigabytes can still run fairly capable local models, but it limits the size and flexibility of what you can load compared with 24 GB. For AI-first use, I’d take the 7900 XTX; for gaming-first use with occasional AI, I’d take the 5080.