IBM claimed the machine could search for 200 million moves per second, while Stockfish in the recent AlphaZero match could "only" search for 80 million per seconds on a modern multi-core machines. But... it was unclear how exactly IBM derived the number. There's no universe definition on how an engine calculates number of moves per second. How it's done is implementation dependent. It's not always possible to compare engines by number of moves searched per seconds. There was a somewhat related discussion on Rybka:
Deep Blue is out-dated, it was made before this century. Please note I was too young for the project, so my understanding might not be 100% correct.
Deep Blue had a paper. I just had a quick 10 minutes look. Let's give a try:
- 2018, nobody uses human grandmaster games for tuning
- Deep Blue didn't focus on deep search as much as modern engines like Stockfish (https://en.wikipedia.org/wiki/Deep_Blue_(chess_computer)#Aftermath supports my argument)
- Deep Blue was running on a dedicated machine. We don't do that in 2018.
- IBM had 32GB hash table, we can do better than that in 2018
- Deep Blue didn't have something known as null move pruning. (page 5 in the paper). I'm very confident all modern engines use it. It's a simple technique.
- Deep Blue didn't use late move pruning (I failed to see in the paper)
- Deep Blue used MPI for parallel search. This doesn’t happen anymore.
- Deep Blue was running evaluation on hardware, make no sense in 2018. If Stockfish did that, you wouldn't be able to use the engine on an iPhone.
- Deep Blue had very basic move ordering (http://www.talkchess.com/forum/viewtopic.php?t=64021&postdays=0&postorder=asc&topic_view=flat&start=10 by Alvaro Cardoso supports my argument)
- Deep Blue had a 5-piece tablebase. Stockfish has 6.
- Deep Blue pruned less than modern engines.
- The algorithms was hardware specific (e.g. generating moves). We don't do that anymore in 2018. We have efficient 64 bits operations.