Although I would not buy a Mac, I am interested in the speed of KataGo on Apple M4 Max with 40 GPU cores and 16 CPU cores. Speed must be expressed as visits per second, which is new and reused playouts per second. The minimum speed on a typical position, such as a quiet opening position, should be observed. That is, there are no or only few reversions, which would dramatically increase speed. Also, please state the name / number of the neural net you are using! If you can, also tell us the watt under load.

The GPU of Apple M4 Max seems to be much faster for intensive tasks than (especially much) earlier Apple chips. So it should be interesting to compare how much the remaining gap to Nvidia GPUs has shrunk.

I think nobody should buy a Mac only to run KataGo but, if somebody buys a new Macbook Pro for other reasons anyway, it would be really nice to share some data for KataGo!

See here for previous speed comparisons:
https://home.snafu.de/jasiek/AI_Compute ... ocId629239

See: https://talkchess.com/viewtopic.php?p=975077#p975077

IIUC, the comparison there is:

"Lc0 engine [...]
The 4090 gets around 10,000 nps.
MacBook 16-inch M4 MAX = 40 GPU cores = 9326 nps [estimated due to M4 Pro 20 GPU cores result]"

It is ambiguous whether the reference is to the desktop or laptop 4090. For KataGo, they cite some numbers but do not state the used computer.

Now, I do not know whether I should be impressed by the Lc0 comparison or whether it is just too different from KataGo.