The argmax type options give the time index which is related to time, of course, by a factor of a TR usually. So let's say a cue happens at a particular time index and we want to look at only a certain length of time after that cue, then we can write a 3dTstat command like this to get the time index relative to that cue :

set cue = 12
set duration = 10

@ endcue = $cue + $duration

3dTstat -argmax -prefix timetopeak_cue12 dset+orig"[${cue}..${endcue}]"

The output voxels have intensities corresponding to the relative index following the cue that has the maximum intensity. So for example if a peak occurred two sub-bricks after the cue (the third sub-brick of the piece of the dataset we are looking at), then the output will have an intensity of 2 at that voxel. If you want that in time, you can use 3dcalc to multiply the dataset by the TR value. You can use this method to look for peaks in EPI datasets or tent / spline deconvolution results, but be careful there is often a lot of noise in this data making this approach less reliable.