Fritz 17

[Ovyron]

The most ironic thing would be that all these discussions about piracy would be irrelevant because weights that are better than Fat Frit's exist and are freely available already.

[Nay Lin Tun]

Why not?

Well, it depends whether all he cares about is the NN engine.
If that's the case, he should just go with Leela.
But if he's interested in Fritz17, etc., that's different.

I guess it will depend on Fat Fritz results.

Of course you may wonder whether
any new CPU engine better than SF or not?
any new NN engine better than Leela or not?

The chance is pretty, pretty slim!! ( not worth hyping). The best you can get is pretty close result.

[Ozymandias]

Thx for speedy response, Graham and Ozymandias. Sorry but I have no idea what fp16 is. Not asking for a lengthy explanation; just a quick pointer for if I buy this thing. Cheers....

It's what makes the new nVidia cards so fast. They're faster than previous generation even without it, but once you take advantage of this, it more than doubles the performance for RTX cards. Look at the 2nd and 3rd graphs here.

[Stonker]

Many thanks. Very informative.

[Stonker]

Why not?

Well, it depends whether all he cares about is the NN engine.
If that's the case, he should just go with Leela.
But if he's interested in Fritz17, etc., that's different.

I guess it will depend on Fat Fritz results.

Yes, I am mainly interested in FF, as I have the Komodo subscription and obviously SF. The selling point for me is the fact that this will all be available via a simple download/install from Chess base instead of the (to me) complex 'pick and mix' of various ever changing ingredients that seems to be necessary with the other NNs. I just don't want to get into all that. I just want to befuddle my brain with the 64 squares!

[chrisw]

We buy FatFritz and everything else is not our problem.

Some people are claiming that only one person needs to buy Fat Fritz and then they can distribute its weights so anybody else can use it without buying it. I don't think I've ever seen a precedent for a case like this, it would be piracy to distribute anything on the disk that they sell, except for that weight file?

Inversion of claim is not the same as the claim.
The claim is that data produced by mechanical process can not be copyrighted.
If and when some network weights data carries with it a legal notice of a legally accepted copyright registration, together with the reference number and the jurisdiction, then people can take a look at the claim and decide.

[George Tsavdaris]

Thx for speedy response, Graham and Ozymandias. Sorry but I have no idea what fp16 is. Not asking for a lengthy explanation; just a quick pointer for if I buy this thing. Cheers....

But is the NN engine all you care about, not the Fritz17 AB engine? If that's the case, it's not worth buying.

Why not? For me seeing this result makes it interesting.
This result came from an independent from Chessbase or whatever related to Fat Fritz person, who got(bought or got for free from Albert, don't know exactly) the Fat Fritz net before its release.

FAT FRITZ(2 x RTX 2070) v STOCKFISH 201019(28 threads i9-7960X) 5min + 5s/move time control generic opening book (8 ply reversed colors):
FatFritz - Stockfish_201019 result: 13.0-7.0, +6 -0 =14.

PGN: https://gofile.io/?c=66J2Cj

[shrapnel]

I would like to know whether it's worth getting Fat Fritz with only a GTX 1660 GPU. I have Ryzen 1700 with 32GB RAM.
Does it just run a little slower or not at at all?

Yes - it will be fine.
Albert says that you can actually use fp16 according to a well-known tester. Not quite as fast as the RTX cards, but 1.7 times faster than plain CUDA.

Hi Graham, I don't think that's strictly accurate.
My understanding is that to be able to utilise cuddn-fp 16 you need to have an RTX 20 Series Card, like 2060, 2070, 2080 Super and of course the 2080 Ti Card.
He has an old GTX 1660, which will have only have cudnn as backend option,not cuddn-fp-16, unless of course he has the new GTX1660 Super,which he hasn't specified.
Even my previous 1080 Ti GTX which I had before I purchased my 2080 Tis refused to run if I chose cuddn-fp16 as Backend, I know this for my own experience. It ran only on plain cuddn.
You experts know perfectly well that Fat Fritz will run slow as molasses on an old GTX 1660, and he would feel cheated if he was buying Fritz 17 only for the NN part.

[Ozymandias]

My understanding is that to be able to utilise cuddn-fp 16 you need to have an RTX 20 Series Card, like 2060, 2070, 2080 Super and of course the 2080 Ti Card.
He has an old GTX 1660, which will have only have cudnn as backend option,not cuddn-fp-16, unless of course he has the new GTX1660 Super,which he hasn't specified.
Even my previous 1080 Ti GTX which I had before I purchased my 2080 Tis refused to run if I chose cuddn-fp16 as Backend, I know this for my own experience. It ran only on plain cuddn.
You experts know perfectly well that Fat Fritz will run slow as molasses on an old GTX 1660, and he would feel cheated if he was buying Fritz 17 only for the NN part.

The whole GeForce 16 series is Turing based. I don't know what card you have the GTX 1660 confused with, but they were released this year.

[shrapnel]

My understanding is that to be able to utilise cuddn-fp 16 you need to have an RTX 20 Series Card, like 2060, 2070, 2080 Super and of course the 2080 Ti Card.
He has an old GTX 1660, which will have only have cudnn as backend option,not cuddn-fp-16, unless of course he has the new GTX1660 Super,which he hasn't specified.
Even my previous 1080 Ti GTX which I had before I purchased my 2080 Tis refused to run if I chose cuddn-fp16 as Backend, I know this for my own experience. It ran only on plain cuddn.
You experts know perfectly well that Fat Fritz will run slow as molasses on an old GTX 1660, and he would feel cheated if he was buying Fritz 17 only for the NN part.

The whole GeForce 16 series is Turing based. I don't know what card you have the GTX 1660 confused with, but they were released this year.

My bad. If they are Turing based, then it will be able to utilize cuddn-fp16 as backend.
But still the performance will suffer and that's bad because lc0/fat fritz needs all the boost it can get, to be a Match for the top AB Engines.