I'm shopping for a used GPU in Germany and am comparing an RX 9070 XT for €625, an RX 7900 XTX 24 GB for €700, and an RTX 5080 16 GB for €850. The system would use a Ryzen 7 9800X3D, 64 GB of DDR5-6000 memory, and an 850 W power supply. I'll be gaming at 1440p and 4K on a 240 Hz OLED, mainly playing Counter-Strike, League of Legends, and Warzone. I also do professional work involving AI, statistical analysis, local inference, and hosting LLMs. Keeping sensitive project data local is important to me. The RTX 5080 is currently my least appealing option because of its smaller VRAM capacity, but its gaming performance, upscaling, frame generation, ray tracing, CUDA support, and broader AI compatibility are attractive. The RX 7900 XTX offers 24 GB of VRAM and may be more useful for larger local models, while the RX 9070 XT appears to offer the best gaming value and newer FSR features. Which card would you choose at these used prices, and how much does the 7900 XTX's extra VRAM matter in real-world local LLM and AI workloads?
4 Answers
The RX 9070 XT is the value option, but the price gap to the 7900 XTX is only €75. That makes the XTX more attractive if you can use the extra VRAM. The 9070 XT benefits from newer architecture and may have better support or performance for newer FSR features, while the XTX gives you more raw memory capacity. For Counter-Strike, League, and Warzone, all three cards should be extremely fast; the bigger differences will show up in demanding 4K titles, ray tracing, AI software support, and model size.
At €850 used, the RTX 5080 is worth serious consideration rather than being automatically overpriced. It is the best all-around gaming card here and should pair well with a 4K high-refresh display. However, the right answer depends heavily on the models you actually plan to run. If your priority is gaming with occasional AI experimentation, choose the 5080. If running larger local LLMs is a core requirement, the 7900 XTX’s 24 GB may matter more than its lower gaming feature set. Also verify the used card’s warranty, condition, power connector, and seller reliability before deciding.
At these prices, the RTX 5080 is much more reasonable than it would be new. It is the strongest gaming option, particularly for 4K ray tracing, DLSS, and frame generation, and CUDA generally makes local AI software easier to use than AMD’s ROCm ecosystem. Sixteen gigabytes can still handle a useful range of local models, but you’ll need to be more selective with model size, quantization, and context length. For a workload split between gaming and AI, I’d lean toward the 5080 unless you already know you need models that exceed its VRAM.
The RX 7900 XTX is the better choice if local AI and larger models are a major priority. Its 24 GB of VRAM gives you considerably more room for 30B–35B-class quantized models and larger contexts, and it is generally faster than the 9070 XT in traditional rasterized games. The trade-off is weaker ray tracing and a less convenient software experience for AI, since many tools are designed around CUDA first. At €700, it offers a strong balance if you value capacity more than maximum upscaling and ray-tracing performance.
That is exactly what makes the decision difficult. I don’t replace my main PC every few years, so I’m worried that 16 GB could become restrictive for local models even if the 5080 is clearly faster in games.