What are NPS in the engine?

I recently became aware of all the other stuff engine offers besides analysis. I saw some engines are faster, slower, some of them show higher evaluation and so on. Also, both engine thought they're "better" in the example below. 

Last week I downloaded Stockfish and matched him with Houdini, to play some rapid match. I was surprised Houdini actually won more times than Stockfish although Stockfish is considered better (rating).

I also saw Houdini had more NPS than Stockfish, but Stockfish calculated significantly faster (same plys) or deeper (same time).

How come Houdini then won more times and what actually are NPS (nodes/s)?

I noticed that, too. Not sure what impact it has on the calculating.

If both engines are evaluating from the point of view of White, then nothing is wrong; It's just that both engines are saying that White is slightly better.

Stockfish 2.0.1 JA is an old version of Stockfish. Houdini 1.5a is actually rated higher than Stockfish 2.0.1 JA.

Make sure that both engines are configured with the same resources: Same number of threads, same amount of hash, etc.

Nodes per second can be interesting to watch, but it's pretty much useless to try to compare NPS between different engines. (It's because almost every engine programmer will define NPS differently.)

http://chessprogramming.wikispaces.com/Nodes+per+second

Thanks for that. But why is Stockfish calculating deeper, more plys, but still lose?

I don't know for sure, but my guess would be that maybe the Houdini evaluation knowledge is better/more complete than Stockfish, and/or Stockfish's higher pruning rate can sometimes overlook good moves. (Probably more the former than the latter case.)

P.S. - Is Stockfish actually calculating to deeper plies? In your screen shot, both engines are at ply # 20.

Yes, when I set them to play bullet games, Stockfish had 1 ply more in average than Houdini. When playing longer games, the difference was even 2 or 3 sometimes (those games were mostly draw though).