When the tracks are monitored, they bypass the ASIO Guard and are only processed with the Real-Time path (the base buffer that you set in your driver).
The Real-Time path and the ASIO Guard path are processed in parallel. During playback, the tracks that are already recorded (and not monitored) use the ASIO Guard path, and the monitored instruments that you play in real time use the Real-Time path.
This is why the delay added by ASIO Guard is only happening when you hit Play initially, and not occurring an additional time on monitored track.
You computer does not work more or less harder before the project starts playing, it always goes as fast as it can in order to render the audio in real time, just that it starts doing so in advance so that the CPU has more heardroom to process everything in time.
For example, with a given workload, if you decrease the buffer size from 1024 to 256, the performance meters will go up because the CPU has 4x less time to process each audio block.
The performance meter only represents what the CPU can achieve at any given moment, based on its frequency and current load. It does not represent how hard it is working.
If you open Task Manager you’ll see that the CPU usage remains the same wether you are using a buffer of 1024 or 256.
If you put 5 times more plugins on your tracks, then only now the CPU will work harder, as it has more stuff to process, and the CPU usage will increase. More processing to do means more time needed to process one audio block. If the buffer is too low and the processing takes longer than this, then the audio will dropout since the block won’t be entirely processed in time, before the next block starts. In this case, the Real-Time or ASIO Guard meter maxes out, whichever comes first.
I hope this helps!
Take a look at the following topic :