5 million records sort txt file (not using any tecnique)?

If the whole file would fit in memory, I'd read it all in with File.ReadAllLines().
Make 1 pass through all of the lines and build a parallel array that has the parsed numeric keys.
Then use Array.Sort<TKey, TValue>(TKey[] keys, TValue[] items) to do the sorting.
(This will sort both arrays together by the sort order of the keys array.)
Then rewrite the file using the sorted array of the lines.

If the whole thing will not fit in memory, then you could read the file line-by-line, extract and parse the key into an array, and simultaneously construct a parallel array of objects with the start byte position in the file and the byte length of each record.
Do the Array.Sort() as above then write a new output file, copying each record from the input file by seeking to the record position and copying length bytes to the output file.

[edit: Matt T Heffron]
From your comment it looks like you can pull the whole thing into memory, so modifying the code in your comment. I'm a little unclear on the nature of your sort key. It appears from the example in the question that you have floating point keys (shown with the "," as the decimal separator). However in your comment's code, you appear to be using only the integer before the first comma as the sort key, but, again, the original example implies a secondary sort based on the second number. I'll show this as sorting by the first integer column case, ignoring the second column:

var lines = File.ReadAllLines(fileunordred);
int[] allCustomerIds = new int[lines.Length];  // make it the same length as lines
char[] splitter = new char[]{','};
for (int ix = 0; ix < ix.Length; ++ix)
{
  var splitLine = lines[ix].Split(splitter, 2);
  int customerId;
  if (!int.TryParse(splitLine[0], out customerId)
  {
    // error parsing the data, do something "sensible"
    allCustomerIds[ix] = -1;  // some value to indicate a "bad" row, to sort together
  }
  allCustomerIds[ix] = customerId;
}
Array.Sort(allCustomerIds, lines);
// the <int,int> is wrong 
// and <int,string> is unnecessary since the compiler can figure it out.

both arrays are now sorted in ascending numerical order by the customer id.
just use File.WriteAllLines("filename", lines) to make the sorted file.

[Edit #2: Matt T Heffron]
Mostly very similar to the above, but now it compares first by the first integer and then by the second integer within the group that has the same first integer.

var lines = File.ReadAllLines(fileunordred);
int[] allInfo = new int[lines.Length];  // make it the same length as lines
char[] splitter = new char[]{','};
for (int ix = 0; ix < allInfo.Length; ++ix)
{
  var splitLine = lines[ix].Split(splitter);
  int[] pair= new int[2];
  allInfo[ix] = pair;
  int id;
  if (!int.TryParse(splitLine[0], out id))
  {
    // error parsing the data, do something "sensible"
    id = -1;  // some value to indicate a "bad" row, to sort together
  }
  pair[0] = id;
  if (!int.TryParse(splitLine[1], out id))
  {
    // error parsing the data, do something "sensible"
    id = -1;  // some value to indicate a "bad" row, to sort together
  }
  pair[1] = id;
}
Array.Sort(allInfo, (a, b) => {
  int comp = a[0].CompareTo(b[0]);
  return comp == 0 ? a[1].CompareTo(b[1]) : comp;
});
//Now just rewrite the file from the integers, (the lines array IS NOT sorted)
using (var out = new StreamWriter("outputfilename"))
{
  foreach (var pair in allInfo)
  {
    out.WriteLine("{0},{1}", pair[0], pair[1]);
  }
}

(I haven't actually tried this but it should be close...)