Should we come up with a way to benchmark these AI/ML tools in flame ?
in the resolve discord we found that for some tools the M3Ultra was like 2-2.5x as fast as a M2U , namely depthmaps and “Magicmask” .
In general computing tasks the M3U is maybe 20% faster than a M2U - however in ML there seems to be a much bigger difference.
With these tools becomming the modern main bottleneck for many that do timeline work and with these ML tools finally running on GPU on macOS maybe its time for a flame-ML benchmark archive.