diff --git a/doc/ug/lb.tex b/doc/ug/lb.tex
index 764441305bf..23099aa2539 100644
--- a/doc/ug/lb.tex
+++ b/doc/ug/lb.tex
@@ -71,7 +71,14 @@ \section{Setting up a LB fluid}
 expressed in \var{agrid}, all times in \var{tau} and so on.  Therefore
 changing \var{agrid} and \var{tau}, might change the behaviour of the
 LB fluid, \eg at boundaries, due to characteristics of the LBM
-itself. Currently it is not possible to precisely give a parameter set
+itself. There is a slight difference between the GPU and CPU implementations
+in that the CPU places the LB nodes at 0.5, 1.5, 2.5... whereas in the GPU
+code they are located at 0, 1, 2...  The shift allows for the use of Verlet
+lists in the CPU implementation.  This will cause the results to differ
+on the length scale of the LB grid spacing.  This can be
+eliminated by shifting all positions by +0.5 in x,y and z in
+the CPU input script.
+Currently it is not possible to precisely give a parameter set
 where reliable results are expected, but we are currently performing a
 study on that. Therefore the LBM should \emph{not be used as a black
   box}, but only after a careful check of all parameters that were