Go forward to Article Caching.
Go backward to Sorting.
Go up to The Summary Buffer.

Asynchronous Article Fetching
=============================

   If you read your news from an NNTP server that's far away, the
network latencies may make reading articles a chore.  You have to wait
for a while after pressing `n' to go to the next article before the
article appears.  Why can't Gnus just go ahead and fetch the article
while you are reading the previous one? Why not, indeed.

   First, some caveats.  There are some pitfalls to using asynchronous
article fetching, especially the way Gnus does it.

   Let's say you are reading article 1, which is short, and article 2 is
quite long, and you are not interested in reading that.  Gnus does not
know this, so it goes ahead and fetches article 2.  You decide to read
article 3, but since Gnus is in the process of fetching article 2, the
connection is blocked.

   To avoid these situations, Gnus will open two (count 'em two)
connections to the server.  Some people may think this isn't a very nice
thing to do, but I don't see any real alternatives.  Setting up that
extra connection takes some time, so Gnus startup will be slower.

   Gnus will fetch more articles than you will read.  This will mean
that the link between your machine and the NNTP server will become more
loaded than if you didn't use article pre-fetch.  The server itself will
also become more loaded--both with the extra article requests, and the
extra connection.

   Ok, so now you know that you shouldn't really use this thing...
unless you really want to.

   Here's how:  Set `gnus-asynchronous' to `t'.  The rest should happen
automatically.

   You can control how many articles that are to be pre-fetched by
setting `nntp-async-number'.  This is five by default, which means that
when you read an article in the group, `nntp' will pre-fetch the next
five articles.  If this variable is `t', `nntp' will pre-fetch all the
articles that it can without bound.  If it is `nil', no pre-fetching
will be made.

   You may wish to create some sort of scheme for choosing which
articles that `nntp' should consider as candidates for pre-fetching.
For instance, you may wish to pre-fetch all articles with high scores,
and not pre-fetch low-scored articles.  You can do that by setting the
`gnus-asynchronous-article-function', which will be called with an
alist where the keys are the article numbers.  Your function should
return an alist where the articles you are not interested in have been
removed.  You could also do sorting on article score and the like.