You really don't want a 1.3 billion point table anywhere -- Even with the best possible
performance, you're looking at a 30+ hour load time (and hours to do any meaningful
analysis).
If your points don't vary in value all that much, you might be able to map 1Kx1K pixel
chunks into a multi-point table (where all the points with the same value are merged into
a single feature) -- this will still be moderately unusable as tens of thousands of features
with tens of thousands of verticies.
Your best bet is to apply some classification scheme and convert the result into a polygon
table. Still unwieldy due to the vertex count, but better than the other options.
- V