Quote:
Originally Posted by LoRd_MuldeR
"Auto-detection" is working as expected, probably. That is because auto-detection has not been implemented to simply run as many encoder instances in parallel as there are (logical) CPU cores available. Instead, it has been optimized to run as many encoder instances in parallel as give the maximum throughput (or, equivalently, minimize overall encoding time). Note that running too many instances in parallel can easily result in HDD thrashing and thus slow things down considerably, as all those parallel instances also need to read their respective source file and write their respective output file. The following "cores to instances" curve has been derived experimentally, with the help of various LameXP users:
https://i.imgur.com/6XYwr03.png |
wow. that's interesting. thanks for clearing up.
in my case it was faster using more threads manually, but that doesn't mean it's the same for everybody. probably depends on a lot of components.