SNPs are much denser than RFLPs and VNTRs therefore the DNA resolution is much greater with SNPs. VNTRs were historically used for linkage mapping while SNPs allowed for association studies (e.g. GWA studies). Therefore your question goes down to what are the differences between linkage mapping and association studies.
They are both forward genetic methods which aim to identify a genetic modification provoking a trait.
The two major differences between both approaches are 1) linkage mapping requires a pedigree but can be performed using a fairly small sample size while 2) association studies can be performed at the population level but usually require a much larger sample size.
Both approaches require markers but for different reasons. For linkage mapping, markers, such as microsatellites, are used to delimit the genomic region associated with the trait and take advantage of recombination events in the genome to pin down that region. This, of course, requires both 1) recombinant individuals in the pedigree and 2) at least one marker that is co-inherited with the trait. Linkage mapping allows to find virtually any genetic modifications, such as a genomic deletion, but only in the case of a genetic modification showing high penetrance.
For association studies, markers are the actual genomic modification themselves (usually SNPs) and are used directly in statistical tests (often a logistic or linear regression) for association with the trait. This means that either 1) the causal marker is available or 2) at least one marker is highly co-inherited with the causal mutation (i.e. in linkage disequilibrium). Association studies allows to find both low- or high-penetrance mutations, given a sufficiently big sample size. Also an association study can be conducted in the presence of a non-discrete (i.e. linear) phenotype which one cannot use in linkage mapping.
Another difference is that linkage mapping does not require any a priori knowledge of the genomic modification causing the trait while association studies do (as markers are statistically associated with the phenotype).
Usually both methods will highlight, with varying resolution based on the phenotype and markers used, the genomic region causing a trait but rarely the actual causal mutation will be found with 100% certainty.