Keep in mind that the incoming audio needs to be accumulated in a buffer for a certain amount of time before the data can be computed and the display updated. In contrast with the buffers you probably know from soundcards, this block-processing is not just a computer technicality and only a source of undesirable latency, but an integral part of the process related to the mathematical aspects involved (Time</link>-frequency product uncertainty principle).
As such it determines both the precision of the analysis and the maximum display rate, and should be adjusted depending on the specifics of your application.
In order to maintain a sufficiently responsive display refresh rate, blocks overlap by 75 %.
The default setting is 8192 samples, which corresponds to a length of roughly 180ms at 44.1kHz sampling rate. This value constitutes a good compromise between precision and responsiveness for most situations. However, if you need to measure a particular frequency with great precision, you should raise the analysis block size. On the other hand, if you need to follow rapid spectrum variations, this value should be lowered.