The first series (1996) was a PR scam. It is incorrect to say he was playing 'Deep Blue'. It is far, far more correct to say he was playing a team, comprising many of the top players, who used Deep Blue to test their moves before implementing them. The programming on the machine changed daily. In the second series, the program was - according to IBM - only changed between games ... although there was a serious question of a mid-game change (Game 1) that led to the computer's loss.
That said, yeah, a lot of modern machines leave their predecessors in the dust, computationally. Chess algorithms ... not so much. Deep Blue's 'innovation', such as it was, was simply to numerically rate a sequence of moves, discarding the lowest scoring, and then continuing its computation from that point. (...and it was a supercomputer) Contrasting with the previous 'Brute force, try all possibilities, select the best after _n_ moves.' As chess is, practically, a finite game, once computers reach the level of _n_ that is about the end point of all games, they aren't going to lose anymore. A lot of the modern chess programs that are free/cheap follow the brute force model, not the more analytical method pioneered by Deep Blue. The top machines do have better coding that Deep Blue ... more importantly, the number of plies has improved, due to better weighting (far more situational / far less point oriented).