How you measure this depends on what your equipment can do.
Active differential probes could be a possibility, though I myself would prefer to use a scope with two channels that can internally subtract the two. This way, I can compare the performance of the individual lines of a differential pair, which can sometimes illustrate issues such as differential-to-common-mode performance of the link. It is important in that case to use phase-matched cables to connect the DUT to the source and scope, since you want to measure the line, and not the mismatch of your cables.
analogsystemsrf suggests at least 10x faster sample speed, which is a good rule-of-thumb, but unfortunately not always possible - the fastest scope ont he market samples at 256 gigasamples/second, which would mean you can only measure up to 25 Gbit/s with it. However, 25 gigasamples/s is much more reasonable and should be achieved by most scopes capable of measuring a 2.5 GT/s link.
It could be interesting to look at scopes that have built-in standard compliance tests. Many vendors sell options to have the scope measure a number of things automatically for you and compare it to the requirements set up by the standard used.