I run similar setup except I also have a 980Ti GPU. A lot depends on the quality, size and format of the photos.
I work with RAW format at 20 megapixels. At around 180 photos the ram usage goes above 30gb. At mesh creation (decimation stage) that goes above 45gb.
Don't take my words as a guide, but I would tackle it this way:
1.Create 5 chunks with 900 photos each
2.Do a sparse cloud for each, alignment correction, marker detection if any, scale bars etc (ensure your geo coordinates are ok)
3.Create dense clouds of each chunk
4.See below for alternatives
-Attempt to create a mesh of each chunk (You might go over 140gb here during decimation stage)
or
-Use the python script PS120_split_in_chunks_dialog.py to split your chunk into smaller pieces. Build a mesh for each of them (I would say 2 by 2 would be enough). Don't need to recreate a dense cloud, also do not merge them instantly. Once smaller chunks are complete realign those chunks using your camera or markers. Finally merge smaller chunks.
5.Align big 5 chunks
6.Merge Meshes of big 5 chunks together.
7.Texture (Disable color correction as it might take days for this data set)
Warning: Make sure your SSD has enormous amount of free space. When using python script mentioned above you will duplicate dense clouds and store them on SSD.
Python script I mentioned can be taken from Agisoft Python page:
http://wiki.agisoft.com/wiki/PythonP.S. It has interpolation enabled by default, look through the code close to the bottom if you want different settings for mesh generation (low, medium, high quality etc.)