If anyone has trouble with quartz oscillator stability after upgrading to an ATMEGA328P, replacing C8 and C9 with 10 pF capacitors might be worth a try.
The reason I specifically recommended an ATMEGA328P-PU in my previous post is that, in my clock at least, an ATMEGA328P-PN was unable keep time. But after a couple months of use, the same clock with an ATMEGA328P-PU chip started losing time also. I suspected unstable oscillation of the quartz crystal.
My clock came with 20 pF capacitors for the oscillator, and the ATMEGA328P datasheet reports pin capacitance on XTAL1 and XTAL2 as 18 pF and 8 pF. Assuming a stray capacitance of 2 pF, the total load would be
{ ( 20 + 18 ) * ( 20 + 8 ) } / { ( 20 + 18 ) + ( 20 + 8 ) } + 2 = 18.1 pF
Since the crystal is designed for a 12.5 pF load, a load of 18.1 pF seemed a bit high. I decided to replace the 20 pF caps with 10 pF caps which would reduce the load to
{ ( 10 + 18 ) * ( 10 + 8 ) } / { ( 10 + 18 ) + ( 10 + 8 ) } + 2 = 13.0 pF
After that change, both an ATMEGA328P-PU and an ATMEGA328P-PN worked perfectly, and I hope the oscillator will continue to run stably.
As a side note, the crystal in the Ice Tube Clock has a frequency tolerance of 20 ppm which amounts to less than 2 seconds per day. But even with the ATMEGA168V, I've notice a time drift of greater than 2 seconds per day,
as have others. The ATMEGA168V datasheet, does not state pin capacitance, so using the somewhat standard assumption of 5 pF on each pin, the load would be
{ (20 + 5) * (20 + 5) } / { (20 + 5) + (20 + 5) } + 2 = 14.5 pF
which seems close enough. But if the actual pin capacitance on the ATMEGA168V is more akin to the closely related ATMEGA328P, that might explain why time drifts as much as it does...