I think that in order to make a good comparison it would be interesting to perform the same test. I used to calculate nps the command
bench 1024 16 16.
Perhaps to make a better comparison you could use the same command as on the page:
http://ipmanchess.yolasite.com/amd--int ... ckfish.php
bench 1024 16 26 default depth nnue vs bench 1024 8 26 default depth nnue
With this command the nps I think is somewhat lower, for example with 16 threads I think it is something like 16 Mnps.
On the page you can see that processors that have 8 cores/16 threads show significantly more nps when using 16 threads vs. 8 threads.
If there is something else that we may need to be taken into account, in this ranking they only take into account the nps, however in the bench command it also shows the time to complete the test. Perhaps when a processor with hyperthreading or smt exceeds the number of threads to cores for chess, the number of nodes per second increases but the time to complete the test does not decrease. So perhaps some people may be right that increasing the number of threads above 8 will not give more power to the engine. Here it would be interesting to see a match between stockfish 8 threads vs stockfish 16 trheads.
All this doesn't explain why the 5800U doesn't show more nps with 16 threads. The 5800U in turbo mode should have 8 cores at 3.40 GHz and the 5800H 8 cores at 3.80 GHz.
