One of the common arguments against solar as an energy source is that it’s just too variable. You can never count on it when you need it. What if clouds roll in and out? 
One counter-argument might be – well, you never know when anyone will turn on their AC, either, at least not minute-by-minute. The grid is a balancing act; unpredictable, random loads have the same effect as unpredictable, random generators.
To which one might then counter yes, but there are so many AC units out there, they average out, more or less, turning on and off at random times and smoothing things out in aggregate.
To which the solar advocate might reply OK, then with enough solar the peaks and valleys of generation should cancel out too, as clouds move out of one area into another. Does this seem likely out in practice?
To find out, I grabbed 5 minute data from about 40 Enphase systems in the twin cities on a highly variable, sporadically cloudy day. Because we don’t yet have a whole lot of solar here, and I didn’t want the one or two large commercial systems in the group to swamp the smaller residential systems, first I normalized them all to a % of their max output. (This might be cheating a little, but with a lot more systems randomly distributed in size and geography, the swamping-out effect should be minimiized.) Here’s what just 4 of those systems looks like; each is indeed pretty messy and unpredictable at the 5-minute range:
The caveats might be that this is a very wide geographic range – I grabbed systems from all of the twin cities and suburbs. And that’s probably larger than the various sub-grids within the cities; what the variability is within those subgrids is, or how this solar variability affects them, I’m not sure. And of course my initial normalization of all systems to the same size could be argued with.
There have been much more rigorous papers and presentations written on this as well, see for example “Quantifying PV Power Output Variability” by Thomas E. Hoff and Richard Perez in 1999, and “Implications of Wide-Area Geographic Diversity for Short- Term Variability of Solar Power” by Andrew Mills and Ryan Wiser at LBNL in 2010. But with the advent of 5-minute monitoring from systems like Enphase, I wonder if even better results could be found from this wealth of data.
 I’ll submit that a sporadically cloudy day is more trouble to a grid operator than a generally cloudy day. We often know if a day will be cloudy well ahead of time, and that doesn’t yield the minute-to-minute variations of a sporadically cloudy day. The grid is better, I think, at responding to these longer-term variations.