"Fat fritz" stronger than Stockfish ?

[smatovic]

So, is ChessBase planning to distribute the source for LC0, er DeusX, er Fat Fritz?

They did per request:

viewtopic.php?f=10&t=71831&start=10#p813130

--
Srdja

[dannyb]

Found on http://vlasak.biz/evcnews.htm

Eine große Überraschung ist der Motor, bei dem es sich um Fat Fritz handeln wird, modifizierte LC0.
ChessBase tötet mehrere Fliegen mit einem Schuss.
- Er wird Vasek Rajlich los, der den Motor nicht entwickelt, sondern nur seinen Namen wieder verkauft hat.
- Gibt TCEC eine echte Chance, in die Premium Division aufzusteigen.
- Es wird sich großartig verkaufen.

Wenn der Benutzer keine teuren Grafikkarten hat, ist die Engine auf einem herkömmlichen Prozessor leider sehr schwach.
Über die Alternative ist noch nichts geschrieben.

Neueren Trends zufolge wird nur Windows 10 unterstützt.

So a gpu is needed to run Fritz 17? is it confirmed? if that's the case I think I'll skip it for now. In my opinion it's still too soon to sell gpu-only engines

[pohl4711]

So a gpu is needed to run Fritz 17? is it confirmed? if that's the case I think I'll skip it for now. In my opinion it's still too soon to sell gpu-only engines

viewtopic.php?f=10&t=71831&start=10#p813130

As you can read there, ChessBase sent 2 links to the lc0-GitHub on request for the FatFritz SourceCode. This means, that FatFritz is 100% identical to lc0 (except some different paramter-settings (perhaps)). The only difference is the neural-net, which ChessBase/A. Silver built for FatFritz.
So, FatFritz will run on GPU (CUDA or OpenCL) or CPU (blas). If not, you can use lc0-blas instead, set the parameters like FatFritz and use the neural-net of the Fritz17-package. Then you have a FatFritz running on CPU.
But, this must be clear: On CPU or non-CUDA GPUs, lc0 (and FatFritz) is running much slower than on a NVIDIA GPU (especially a new RTX Card, which allows 16bit floatingpoint calculations (fp-16 is running around 3x faster, than fp-32)). So, for valid testings and using, a NVIDIA RTX Card is highly recommended. But not necessary.

[smatovic]

Any chance of Fat Fritz being sold as a standalone Program in the near future ?

According to Amazon in Germany Fritz 17 will ship on 12. November and will 'include' Fat Fritz.

https://www.amazon.de/Fritz-17-ganz-gro ... 866817282/

If that means that ChessBase will 'ship' Fat Fritz or grant access via their Engine Cloud remains imo open.

--
Srdja

[chrisw]

Any chance of Fat Fritz being sold as a standalone Program in the near future ?

According to Amazon in Germany Fritz 17 will ship on 12. November and will 'include' Fat Fritz.

https://www.amazon.de/Fritz-17-ganz-gro ... 866817282/

If that means that ChessBase will 'ship' Fat Fritz or grant access via their Engine Cloud remains imo open.

--
Srdja

It seems to be established that the neural net weights are not protected by copyright. If that’s the case, they could be freely shared on the internet (and tested of course via LC0.exe). Assuming the weights file is part of the Fat Fritz distribution.

[smatovic]

Any chance of Fat Fritz being sold as a standalone Program in the near future ?

According to Amazon in Germany Fritz 17 will ship on 12. November and will 'include' Fat Fritz.

https://www.amazon.de/Fritz-17-ganz-gro ... 866817282/

If that means that ChessBase will 'ship' Fat Fritz or grant access via their Engine Cloud remains imo open.

--
Srdja

It seems to be established that the neural net weights are not protected by copyright. If that’s the case, they could be freely shared on the internet (and tested of course via LC0.exe). Assuming the weights file is part of the Fat Fritz distribution.

Yea, reminds me of someone (Ronald?) who said that you can not copyright a bunch
of numbers, and that's what neural network files are, despite the nn structure.

There are funny stories about illegal numbers btw. https://en.wikipedia.org/wiki/Illegal_number

That's why my guess is that they won't ship Fat Fritz, but give access via Engine
Cloud.

--
Srdja

[chrisw]

Any chance of Fat Fritz being sold as a standalone Program in the near future ?

According to Amazon in Germany Fritz 17 will ship on 12. November and will 'include' Fat Fritz.

https://www.amazon.de/Fritz-17-ganz-gro ... 866817282/

If that means that ChessBase will 'ship' Fat Fritz or grant access via their Engine Cloud remains imo open.

--
Srdja

It seems to be established that the neural net weights are not protected by copyright. If that’s the case, they could be freely shared on the internet (and tested of course via LC0.exe). Assuming the weights file is part of the Fat Fritz distribution.

Yea, reminds me of someone (Ronald?) who said that you can not copyright a bunch
of numbers, and that's what neural network files are, despite the nn structure.

Yes, it was a very long long thread in which I argued the weights were a protected intellectual property and he argued they were not. In the end, I conceded defeat, something which does not happen often.
His argument (he is legally qualified btw, I understand) was the weight generation was a mechanical process, no human hand was involved and human hand is a necessary condition for IP copyright protection.

There are funny stories about illegal numbers btw. https://en.wikipedia.org/wiki/Illegal_number

That's why my guess is that they won't ship Fat Fritz, but give access via Engine
Cloud.

--
Srdja

[Ovyron]

However, if you're buying those numbers from Chessbase then distributing them would be piracy, lest I can argue that I can buy Fritz 17 and distribute it because at its core it's just 0s and 1s in some order of some binary file that can't be copyrighted because it was produced by some mechanical process from source code, which wouldn't be different from the mechanical process used to generate the neural net weights.

[chrisw]

However, if you're buying those numbers from Chessbase then distributing them would be piracy, lest I can argue that I can buy Fritz 17 and distribute it because at its core it's just 0s and 1s in some order of some binary file that can't be copyrighted because it was produced by some mechanical process from source code, which wouldn't be different from the mechanical process used to generate the neural net weights.

Are you sure about that? You may be confusing the 1s and 0s that contain within them the representation of protected IP, with 1s and 0's that contain within them the representation of unprotected IP.
If something on the CD is not protected by IP Law, why can't the purchaser copy that part of the CD and transmit it to others?
Put it another way, A sends B a type written text from Chaucer in Sans Serif font, that A typed out onto a piece of paper. The text is not protected because Chaucer is long dead. What stops B from photocopying the piece of paper and sending 1000 copies of it to anyone B wants?

[Ovyron]

Are you saying that it'd be fine to distribute the commercial opening chess book Goi 6.2.1 CTG (which costs $22) because the chess moves it contains and move rankings it contains can't be copyrighted?

I always thought distributing something like that would be piracy.