The new Fritz 9 engine

By Peter Schreiner

Our latest Fritz 9 program comes with a completely rewritten chess engine. The author is Frans Morsch of the Netherlands, who has been improving his chess algorithms ever since the first version of Fritz appeared fourteen years ago.

Already the previous iteration of Fritz was well respected for its balanced style of play, in all phases of the game. It hardly showed any weaknesses and was a very unpleasant opponent in over-the-board games. Last year in Bilbao, running on a simple notebook, it scored 3.5/4 points against the world class GMs Topalov, Ponomariov and Karjakin. Fritz 8 was also considered one of the most reliable and precise analysis partners by most of the top players.

Beautiful 3D graphics of Fritz 9

The improvement of the Fritz 9 engine is for the most part the result of implanting a huge amount of chess knowledge into the program. The new code allows this to be done with great efficiency. The program is so designed that additional knowledge will hardly slow it down at all – which ensures that it will not sacrifice tactical strength for general chess principles.

The properties of a chess position are largely determined by the pawn structure, and this is where Fritz 9 has received much more generalised chess knowledge than its predecessor. Another touchy point is king safety. Unlike human beings chess programs tend to be oblivious to dangers looming against the king until these dangers are revealed in concrete variations which the program can calculate. This weakness is not immediately obvious when you play computers against each other, but when they encounter strong opponents it may spell their downfall.

Fritz 9 has also been greatly improved in this area and can detect long-term threats against the king long before they appear in the search tree. This does not just lead to better defensive play, it also causes the program to sense potential dangers to the enemy king and lauch devastating attacks based on these weaknesses. So it becomes more and more difficult to play “anti-computer chess” against the program, while at the same time it has become an even more dangerous opponent for human players.

Finally the search algorithms have been greatly improved, with the new data structure allowing the engine to use its chess knowledge very deep in the tree. That is probably the most significant improvement. Fritz can now find meaningfull moves even in very balanced positions when there are no tactical motifs to guide its choice.

There was always a danger that these improvements would lead to better strategic play against human opponents, but it could fail when Fritz was up against other computers. Fortnunately this was not the case: first tests are showing that it is considerably stronger than its predecessor against other chess programs as well.

Let us take a look at some concrete positions and see how Fritz 9 handles them.

The Fritz 9 engine quickly finds Kramnik’s move, 12… Nh5!, which according to experts is the most accurate way to achieve equality. Fritz 8 switches to 12…Qc7 and stays with this positionally inferior move.

Here Fritz 9 finds Kasparov’s brilliancy, 42...Bxe3!!, instantaneously. More remarkable is that it gives the full line and displays a very high evaluation in favour of Black.

The game continued 43.fxe3 [43.Ne2 Nh2+ 44.Ke1 Rxg2 45.Qxe3 Nf3+ 46.Kf1 Rg1+!! 47.Nxg1 Rd1+] 43...Rdxg2! 44.Qc3! [44.Qxb6 Rh2 45.Ne2 Rgg2+–] 44...Rh2 45.Ne2 Kh7! [45...Rgg2 46.Qc8+?=] 46.Qc8?! [46.Qb4! f5! 47.Qb5 (47.Qf8 Rh1+ 48.Kf2 Nd2!) 47...f4! 48.Qb4 Nd2+! 49.Qxd2 (49.Ke1 f3 50.Kxd2 Rxe2+ 51.Kc3 Rxe3+ 52.Kd4 f2) 49...Rh1+ 50.Kf2 f3!+–] 46...Rh1+ 47.Kf2 Nd2! [47...Nd2! 48.Ng3 Rh2+ 49.Ke1 Nf3+ 50.Kf1 Rxb2] 0-1. Note that Fritz 9 has understood all of the above analysis and evaluates it correctly in under ten seconds.

In the third game of the match Hydra vs the 32-processor hardware-enhanced Hydra chess machine (which defeated Adams by a 5.5:0.5 score) Hydra played the very deep move 26.Ra6!, which won it universal praise. Fritz 9 finds this move in about 15 seconds on a Centrino notebook, giving it a substantially positive value (+0.79), while its predecessor sticks to the inferior 26.Qf3 with a slight negative score. Incidentally the game ended after 26…Qb7 27.Rd6 Be7 28.Bxh6 1-0.

Fritz finds the black brilliancy almost instantaneously: 33…Nf2!! On modest hardware the program needs just three seconds, with full understanding of all the intricacies involved coming after eight seconds.

The process of discovery: Fritz 9 finds 33…Nf2!! in Ivanchuk-Jussupow
in three seconds and understands it in eight.

Fritz 8 required considerably longer to find the win. The game ended with the moves 34.Bxf4 Qxf4 35.Ne6 Qh2 36.Rdb1 Nh3 37.Rb7+ Kg8 38.Rb8+ Qxb8 39.Bxh3 Qg3 0-1.

Benchmarks

The speed of hardware has great influence on the playing strength of the program. “Engine” – “Chess Benchmark” will run an internal test on your system and give you values for the speed relative to a benchmark processor (P3, 1.0 GHz) and the average nodes-per-second count. You can compare these values for different hardware configurations. Here are some reasonable values we have registered for common systems:

Note that Pentium processors with “Hyperthreading” are simply simulating two processors. This will cause the benchmark program to find two processors, but the benchmark will not be as high as you would expect from a dual system. A dual core processor, on the other hand, will (or should) actually double the speed of the program, as is the case of the dual Athlon in the table above. The values for this system were supplied by John Nunn.

Fritz Chess Benchmark, running on a 2.0 GHz Centrino notebook

The program that actually performs the benchmark terst is called Fritz Chess Benchmark.exe and can be found in the Programs directory of your C drive (in \ChessBase\ ChessProgram9). It is about 450 KB in size and runs independently of Fritz. So you can copy it on a memory stick and take it with you when you are shopping around for a new computer.

Running the benchmark takes just one minute, so it is possible to test a number of candidate machines in a short period of time. You should do this before you decide on one. It is quite surprising to see that sometimes computers which have nominally identical hardware can display considerable variance in the chess benchmark.

• "Where's my Fritz ?" Previous article on the program