Wolff introduced a variant of the SW algorithm, in which a site is chosen at random, and a single cluster is grown around the site. This tends to favor larger clusters, and thus smaller autocorrelation times. Wolff also generalized the cluster algorithms to continuous spin models.
Cluster algorithms work amazingly well for some models, e.g., 2-d
Ising model, where 
.
Even for a large 
lattice, 
, which
is 0(1000) times smaller than 
for the Metropolis algorithm.