I haven't stated it explicitly here yet, but "north" locations default to all white if no tile and "south" locations default to all black if no tile.
Fixed the northern extent and flipped the polarity below "sea-level":
David Glick has pointed out there's a 13n1e which goes against my earlier comment about 8n being the "north"-most extent. Recalculating some pictures now...
Here's a 1/64th scale version (which this site will further scale to fit the card):
Full 1/64th scale version at:
Here is a (hopefully bug-free) version of where the tiles are:
Hmm, looks like a bug is causing the symmetry, not the data.
27 x 81 x 2048 x 2048 = 9,172,942,848
With the extra 13 tiles, that means the pixel count is actually 943,718,400 (although that's just the locations for which there are tiles; arguably there are "pixels" in the white areas above too)
Without requiring contiguity, here are the tile locations:
Doing a brute-force retrieval resulted in 225 tiles instead of 212.
Attempting a brute-force retrieval without requiring contiguity.
I'm suspicious I haven't got it right yet, though, as there's nothing in the lower half.
First attempt at showing which locations in the overall extent actually have tiles:
There are, however, 212 actual tile locations that don't 404. That's still 889,192,448 pixels!
The full extent seems to be 8n and 19s in the vertical and 33w and 48e in the horizontal.
That means an 81 x 27 tile extent (or 2,187 possible tile locations).
I wrote a quick Python script that starts at 1n1e and successfully attempts to load in each of four directions (keeping track of where it's already been) to explore the entire contiguous tile set.
Each tile has a file name of the form
<number>(n|s)<number>(w|e).png
The drawing is divided into 2048 x 2048 pixel tiles.
XKCD's Click and Drag is a wonderfully massive drawing with all sorts of hidden gems in it if you're willing to, well, click and drag.