HI so based on your description I would think that your best bet would bet Time Over Threshold (TOT)
This would be a sliding window where you would set the value of 80%.
So if the CPU is greater than 80% over a five minute windows send an alarm.
This of course could be longer period.
So say you normally see s a spike for 20 Minutes
You could set the TOT to alarm if this goes on for 40 minutes.
That way the 20 minute that is normal is not alarmed on.
As your sounds like short term spike I do not believe Time To threshold TTT would be of much help.
Usually this is used to track thinks such as disk growth rate or network utilization rate to know when the resource is going to run low.
This tracks qos growth over time to provide a predictive warning on when there might be an issue.
I hope this helps