29,90 €
Within the current Fritz18 updates you get a second Fritz 18 engine, the new engine Fritz18 Neuronal. This engine achieves an increase in playing strength of about 120 ELO points compared to the classic Fritz18 engine with short thinking times!
The classic Fritz18 StandardEngine has a manually optimized rating function and achieves a playing strength of about 3300 ELO depending on the hardware. In the new Fritz 18 Neuronal engine, the rating function has been replaced by a fast neural network (NNUE), which makes it play 120 ELO points better, i.e. it reaches about 3,400 ELO. NNUE stands for "Efficiently updatable neural network", the abbreviation being read from right to left because the idea was first published in Japanese in connection with shogi.
Just try the new engine and compare the results with your standard engine. Open a game and run the standard engine (in the example Fritz 18). Then go to the Engine menu and add the Kiebitz item:
In the engine selection dialog, click Fritz 18 Neuronal.
The second engine appears on the screen and the parallel evaluation with the two engines looks like this:
In the example, the previous default engine, Fritz 18, runs in the lower window. The new Fritz 18 Neuronal engine shows its evaluations in the upper window. Both calculate in parallel and the same time at the same position, that their during the game view always run along. Just have a look and you will see that engines can be of different opinion.
Tip: If you have installed the new engine, the same possibility to analyze with two engines running in parallel is also available under our database program ChessBase 16.
The special feature of NNUEs is that they are relatively small and can be efficiently evaluated incrementally during the search. NNUEs do not require a GPU - the instruction set extensions of modern CPUs (SSE2, AVX2 or AVX-512) are sufficient. The fastest version of Fritz 18 Neuronal, which uses the AVX-512 instruction set, achieves about the same number of nodes (NPS) as Fritz 18, the AVX2 version (all modern CPUs) is only 5% slower.
For training the net, a total of about 5 million games were played and 400 million rated positions were generated from them for training the net. The first net was trained on the basis of evaluations from 200,000 selfplay games of Fritz 18, then in further 80 training iterations a total of about 300 intermediate versions of the net were tested and step by step the best net was chosen.
Note: The engine runs exclusively under a Windows 64 bit system!
Back to List
to send a comment please fill the form:
Name *
Email Address *
Subject
Feedback *
