More On SmarThink

     In the previous post I wrote, "Surprisingly, I found that SmarThink's selections did well in matching the GM's opening moves." That was in the Nisipeanu vs. Cornette game. 
     After that post I played an online game using the Grob Attack and because it's not heavily analyzed, I thought the game would be good to analyze with SmarThink, Stockfish 7 and Komodo 8, especially the opening, and compare the analysis just to see it there were any major differences. In all, I spent a couple of hours over the last couple of days looking at the game with all three engines and was kind of disappointed with the performance of SmarThink.
     One thing I noticed is that SmarThink is much slower than either Stockfish or Komodo and, even though in some positions it was analyzing deeply, it was not finding the best lines...this probably had something to do with its speed. On two occasions SmarThink, like I did, missed mates in 10 and 12 moves. Both of the other engines found them in seconds. Even after several minutes SmarThink was showing an evaluation of about 125.0, but still had not found a mate. 
     In any case, I was really looking for opening moves that the other two engines didn't put a high importance on, but SmarThink didn't seem to be up to par. Some of the "improvements" it suggested did not work out well when analyzed with the other two engines. 
     In the end, while I think this engine is very strong, it can't compare to either Stockfish or Komodo, but then what engine can? We'll still go with Komodo for opening analysis and positional play and Stockfish for tactics and endings. 
     That said, SmarThink is a good engine to practice against. I lost a bunch of 5-minute games to it, but it felt more like I was playing against a very strong human than when playing against the other two.  It was taking its time and several times I thought it was going to run out of time,  but then it started blitzing out some pretty strong moves. 
     Also, using SmarThink in the Fritz 12 training modes ("Friend" where the program automatically adjusts its level of play to match that of the opponent and the "Sparring" mode where the program plays a reasonably strong game, but at the same time makes tactical errors) did make it feel like I was actually facing a human.