Nvidia vs AMD for Davinci's AI features

Hi everyone,
in THIS article (by Puget) I read: NVIDIA's Tensor cores can be very useful for many of the new AI-based features in DaVinci Resolve. So, it seems that Tensor cores are the best option for Davinci AI features under Windows. But let's compare some specs:
Nvidia 4090 has 512 Tensor cores and does FP32 82.58 TFLOPS;
Nvidia 3090 has 328 Tensor cores and does FP32 35.58 TFLOPS;
AMD RX 7900 XTX (without Tensor cores) does FP32 61.42 TFLOPS.
Now, my question is: despite that Tflops can be not the direct expression of GPU AI abilities, is it possible that RX 7900 XTX is at least at the level of 3090 in AI performance for Davinci's AI features?

Thanks in advance

Sadly it doesn't seem the Puget has posted the detailed breakdown of their benchmarks for the new cards

https://www.pugetsystems.com/labs/artic ... -creation/

Normally they breakdown Face Refinement and Optical Flow

https://www.pugetsystems.com/pic_disp.php?id=73914

They are in the process of rebuilding their benchmark. Hopefully they add more of the new tools.

It's a minefield but you can extrapolate some idea about how performance compares from Puget user benchmarks:

https://benchmarks.pugetsystems.com/benchmarks/