Looking at some solutions they seem to include their own double parsing implementation. I built a home made serializer for csv files, I am using the default .net parsing functions and I find that parsing numbers/dates is by far the slowest part of the process on large files.