You're right.

I was looking at P-1 save file size, not LL save file size. [i](Ack! Another blindspot discovered. Maybe this cheese really is too ripe.)[/i]

[quote="NickGlover"]M18128533 uses 1024K FFT. This means it uses 1,048,576 elements to compute the FFTs for this exponent. This comes out to 18,128,533 / 1,048,576 = 17.3 bits per element. This actually means that each element stores either 17 or 18 bits of the residue.

I think in George's implementation all the elements are stored in 32-bit unsigned ints. When saving a residue, I'm guessing the program just writes all the 32-bit ints to disk rather then pulling out the releveant 17 or 18 bits from each element. So, this would mean that the save file size for M18128533 is 1024K * 32 bits = 1024K * 4 bytes = 4096KB. It looks like save files are actually 22 bytes larger. Presumably, this is to store other necessary data like the actual exponent, iteration count, etc.[/quote]
And to think all I did was just look at the size of a save file of a similar exponent I had in progress... ;)

A follow-up: in P-1 stage 2 there is only 1 save file, but it is slightly larger. I am testing M17717807, and the save file mH717807 is 4.22 MB (4,429,512 bytes).

Error writing intermediate file
 

I am also now getting this error message. I have lots of space on my hard drive. Does this invalidate my analysis? What can I do to fix it?
Thanks
Steve