AutoCAD's 64-bit precision has been discussed and documented for years and hailed as the most technically precise system for many years since the eighties. This was long before people realized that some would actually need even more precision in the floating point calculation provided by the then math coprocessors and then in the 32-bit platform. There was an example demonstrated by Jon Walker the founder of Autodesk with a AutoCAD drawing showing a to scale of the universe. In this drawing of the universe you could zoom in on the moon and see a moon lander with a plaque and text on it. The drawing has many named views for the locations in the solar system and I have resaved it in AutoCAD 2004 format.
Screen capture of the solar drawing. All images can be clicked on for a larger version in a pop up window. You can also download the drawing from Here
To demonstrate the limitations of the 64-bit floating point I created a polyline with an arc (bulge) next to the moon lander plaque. Now because we are 1.499E+08 (149900000.0) from 0,0 or located at the coordinates of X=-1.205E+08 Y=8.921E+07 if we try to offset this polyline the small radius bulge will round to 0 due to the limit of digits in the floating point of the processor for calculations. So the limitation is if you have an object a very great distance from the origin coordinate of 0,0 the limit of 15-16 digits calculations could result in unexpected results. Of course by simply moving this object temporarily closer to the origin of 0,0 the offset works perfectly because less digits are used in this calculation and no rounding issues.
What does this all mean?
If you are designing a small micro circuit board on the surface of mars, move the part closer to the 0,0 when using operations that require system calculations such as Bhatch, Offset, and so on. Well AutoCAD uses the most precision it can and has since the eighties but with new technology in processors coming we will continue researching to see if we can get even more precision in the extreme conditions.