Puget tested topaz and the 5000 series was slower than 4000 for some reason  https://www.pugetsystems.com/labs/articles/nvidia-geforce-rtx-5090-amp-5080-ai-review/#:~:text=Overall%2C%20the%20Topaz%20Video%20AI,the%20RTX%204080%20&%204080%20SUPER, but that's from a year ago.

Its unlikely unlikely a 5080 is 50% faster than a 4070ti in this program, it's just another program and uses the GPU like anything so we can use general benchmarks to get an idea of it's performance and a 4070ti is always about 30% slower than a 5080 (for cuda, but topaz runs in tensor cores).

For Topaz the 5080/5090 has some weird optimization issue making it artificially slower according to puget's data.  But thise benchmarks are old and I assume topaz has fixed the issue by now. If they did then a 5080 will be proportionally faster in topaz like it is in other tensor workloads, which is hard to find benchmarks for.

A 5080 has 336 fifth-generation tensor cores and a 4070ti has only 264 fourth-generation cores, so it has 27% more cores and they are newer and better cores too. I would suspect it should be around this much faster or more for optimized workloads, probably not 50% faster though like chatgpt thought.