If there is a sudden increase in CPU usage to 98%, say in the last 6 hours, then that increase would show as a peak in the resource graphs when viewing data for the last 6 hours.
However, if you change to view the last 30 days, that peak would be reduced due to the averaging that takes place. Therefore there is no way, using the graphs, to identify surges in CPU usage that did not happen very recently.
I suggest that you allow viewing historic resource usage data at the same scale as the "last 6 hours" view, but at any specified time interval within the last 30 days, so that these spikes can be seen fully.
Note to readers: somebody has already made a suggestion titled "Expose Graph Data through API" which would also solve the problem, so feel free to vote for that too, or instead.