Fat Fritz 2

[Werewolf]

How do you install it? I've done a F17 update and nothing is installed

[Werewolf]

Oh goodness. Do you actually have to buy it for 99 euros? Has the interface changed since Fritz 17?

[Gabor Szots]

Oh goodness. Do you actually have to buy it for 99 euros? Has the interface changed since Fritz 17?

1 - Yes.
2 - No.

[Modern Times]

Oh goodness. Do you actually have to buy it for 99 euros?

I suspect a lot of money was spent on renting hardware to create that double-size network. They have to recover the cost somehow.

[Daniel Shawul]

https://de.chessbase.com/post/fat-fritz ... e-nummer-1

Commercial version of SF? I smell SF dev here .

Not really.
I am testing Fat Fritz 2 right now. Fat Fritz 2 uses a nnue net, which is 2x bigger, than the Stockfish nnue-nets. Because of this, the binary is more than 40MB in size and the nps are measureable lower on my system (AMD Ryzen 12core CPU):

singlethread nps:
Fat Fritz 2 avx2: 1.15 MN/s
Stockfish 210131 avx2: 1.8 MN/s

22 Threads (11 cores, with huge hashtables (8 GB)):
Fat Fritz 2 avx2: 16 MN/s
Stockfish 210131 avx2: 18 MN/s

Fat Fritz uses the SF binary, true, but a complete different nnue-net (double size!). So, FF2 is definitly more than a Stockfish clone like Eman, RaubFish, CorChess and all the others...


More information here:
https://en.chessbase.com/post/92134

"Fat Fritz 2 is trained and developed using this NNUE technology, running inside the open source Stockfish binary, but with a completely new and different neural network that differs significantly from Stockfish's on a couple of key points:
The first is that the evaluations and positions did not come from a brute force engine such as Stockfish or Komodo, but rather came from our original neural network Fat Fritz.
The second key difference is that the neural network in Fat Fritz 2 is double the size of Stockfish 12, using 512 neurons instead of 256, offering the opportunity for double the knowledge and understanding."

IMHO these 2 differences makes Fat Fritz 2 a very interesting new approach. Of course based on Stockfish and the nnue-papers. But definitly more than a simple Stockfish Clone. As far as I know, until now, nobody built such a big nnue-net and nobody built a nnue-net based on evaluations given by a Lc0 neural net.
For me Fat Fritz 2 is much more interesting and innovative than Fat Fritz 1.

The fact that it uses the Stockfish backend gives it a huge push to the very top. This is worse than the first fat fritz that used leela backend, well because the Stockfish search took decades to build and optimize. I am sure if all NNUE used the Stockfish backend they will all be congregated around Stockfish, but in reality most are atleast 200 elos off. That is what Fat Fritz gains compared to other NNUE engiens.

Second that it uses 512 neurons instead of 256 (whatever that means) is not interesting at all.
There are far more interesting fundamental architectural changes to the NNUE architecture being experimented with the SF developers.

Training a NNUE net from A0 style res nets, such as an Lc0 net, is not new either. Dkappe already trained several nets with this approach,
and often his nets require different scaling than SF style nets.

I see it as a ripoff of the latest and greatest open source project with some bizarre claims of originality that could convince the regular user.
What was the claimed novelity of the first Fat Fritz ? Yes, I remember. It uses awesome human games and supervized training it that made it so
so so aggresive. Yeah right.

[dkappe]

Training a NNUE net from A0 style res nets, such as an Lc0 net, is not new either. Dkappe already trained several nets with this approach,
and often his nets require different scaling than SF style nets.

I see it as a ripoff of the latest and greatest open source project with some bizarre claims of originality that could convince the regular user.
What was the claimed novelity of the first Fat Fritz ? Yes, I remember. It uses awesome human games and supervized training it that made it so
so so aggresive. Yeah right.

I’m sorry that you feel that my nets are a ripoff.

[Daniel Shawul]

Training a NNUE net from A0 style res nets, such as an Lc0 net, is not new either. Dkappe already trained several nets with this approach,
and often his nets require different scaling than SF style nets.

I see it as a ripoff of the latest and greatest open source project with some bizarre claims of originality that could convince the regular user.
What was the claimed novelity of the first Fat Fritz ? Yes, I remember. It uses awesome human games and supervized training it that made it so
so so aggresive. Yeah right.

I’m sorry that you feel that my nets are a ripoff.

I have no idea how you reached to that conclusion. I simply stated that training a NNUE net from a ResNET is not new with fat fritz and that you did it already ...

[Frank Quisinsky]

No understandings!
The "net" can be offer without Stockfish!

Or is the "Net" the reason to offer the strongest chess program as commercial product?!

Again, absolutely no understandings!

[amchess]

what's the github repository link?

[Guenther]

"Fat Fritz 2 is trained and developed using this NNUE technology, running inside the open source Stockfish binary, but with a completely new and different neural network that differs significantly from Stockfish's on a couple of key points:
The first is that the evaluations and positions did not come from a brute force engine such as Stockfish or Komodo, but rather came from our original neural network Fat Fritz.
The second key difference is that the neural network in Fat Fritz 2 is double the size of Stockfish 12, using 512 neurons instead of 256, offering the opportunity for double the knowledge and understanding."

IMHO these 2 differences makes Fat Fritz 2 a very interesting new approach. Of course based on Stockfish and the nnue-papers. But definitly more than a simple Stockfish Clone. As far as I know, until now, nobody built such a big nnue-net based on evaluations given by a Lc0 neural net.

Yes I agree. It was inevitable that "Stockfsh clone" would be the response from a few people, but this is very much more than that.

Of course it was inevitable, people can think on their own. It would have been a surprise, if people in majority would have said,
'Bravo Chessbase, well done, another step in the right direction etc.'

Actually the opposite is happening and it is quite understandable.

It would have been better also you hadn't said above in your position, already provided with a version of it, while it is not even
officially released and may be it was even sent for free from CB? What happened to independent testing?

And this

but this is very much more than that.

is completely unproven.
BTW only you and Pohl have used the word 'clone' so far, very telling.