D-Wave processors are quantum annealers, which is a type of quantum computer. You can think of it as having a reduced instruction set relative to gate-model quantum computers. Quantum annealers _can_ be universal, but D-Wave's aren't yet. They also have noise issues but are far more robust to noise than gate-model quantum computers. Right now D-Wave processors provide a heuristic algorithm (quantum annealing) for an NP-hard optimization (Ising minimization, equivalently, quadratic unconstrained binary optimization (QUBO)). Quantum annealing does, in theory, scale better than classical alternatives like simulated annealing. There has been some evidence of this (https://arxiv.org/abs/1705.07452) and I'm sure more will be coming.

The number of qubits is increasing from 2000 to 5000. The connectivity is also increasing from 6 couplings per qubit to 15 couplings per qubit. The connectivity is important because it allows more complex problems to be solved. For example, more qubits means you can solve problems that have more variables. But you can't solve any interesting problems unless the variables interact with each other. More connectivity means you can solve problems with more interactions between the variables.