As ever, it depends, and really this question is unanswerable without knowing more.
It depends on what your users want to take away from the graph. Some may be concerned about the spikes (which would otherwise be lost if data points were too widespread), where as others may interested in an overview in which case something like the following may suit better - calculated as a rolling average, while yet again others may be interested in a flat-line average.
As a generalisation, compare for example the needs of technical, marketing and management. Tech might be interested in spikes; Marketing in variations at time of day; Management in the comparison between this time yesterday or last week.
So your data points depend entirely on what your users need to know. If your users have different needs then provide that via different sets of information - either different graphs or overlaying as appropriate (like in the example above), but always do as much work as is necessary for your users to see the takeaways without having to manually process the information. This may include providing actual numbers (max, min, average, total, comparison with other time period etc etc).
Remember - graphs are not there just for the sake of being there - filling a gap and looking pretty - they should serve the purpose of communicating required information easily to the user.