The annual daylight simulation uses Daysim which simplifies the way it calculates direct sun. See this discussion:
In your first simulation the values are very closed but since the legend lower and higher bounds are not set it looks like that. I would also check the sun position and normal and diffuse radiation values.