update README to mention the new topologies

Author: exaexa

README.md

@@ -350,7 +350,7 @@ limited by the amount of computers you can attach to the analysis.
 The training coordinator node has to keep the full SOM topology in the memory.
 Because we store a "generic" all-to-all distance topology file, this puts a
 limit on the size of the self-organizing map that you can train. If the SOM
-size is K nodes (typically, K=`som-x`×`som-y`), you need to fit (K²+K×D)×4
+size is K nodes (typically, K=`grid-x`×`grid-y`), you need to fit (K²+K×D)×4
 bytes into the memory.
 
 On a modest computer with 16GB of memory, this thus allows you to train a SOM
@@ -392,7 +392,13 @@ generated by `remotesom generate`:
   with **squared** distance to each centroid. (Viewed as a matrix, this would
   be a symmetric matrix with non-negative elements and zeroes on the diagonal.)
 
-You can supply any data in a compatible format.
+Additionally, `remotesom` has out-of-the-box support for hexagonal, toroidal
+and circular grid topologies. To get a complete list of the related options,
+see the list of available options for the `generate` command:
+
+```sh
+remotesom generate --help
+```
 
 #### Can I use a different transport of the SOM training commitments than TLS?