Commercial Release of Ethereal 13.00 (NNUE) for AVX2 Systems

[Damir]

Hi Andrew,
It would be nice if you can put more buying options like paypal on your list...
When only Credit Card, it would seem you are trying to avoid paying taxes....

I do not have a credit card (and dont want to get one). So, without the possibility to pay with PayPal or GooglePay, I cannot buy Ethereal 13 and cannot test it.

I am in the process of getting Paypal to work. Originally I was planning to use Paypal, as it has greater name recognition, and charges lower fees on my end, where as Stripe takes a fair chunk for non US transactions. Paypal integration seemed like it was only meant for sending physical products, not making digital purchases. As a last resort, I might end up adding a Paypal link which will not be a "sale" page, but simply a money transfer page. Downside is that I will not be able to automate the process as well.

The more buying options you have the better. Than people can choose what buying option suits them best. You have nothing to lose and everything to win...

[noobpwnftw]

Do you happen to sell bundles with a GUI or a game database of some sort?

[mclane]

I like the idea very much to have a new commercial engine in the field !!

Always seeing Komodo as the only one fighting against stockfish or LC0 is IMO a limited experience.

Now another commercial.

Lets hope all the others come back soon.
Miss them.

[Thomas Lagershausen]

Can we have some informations about the nnue?

Maybe the size of the net is interesting and what style of play can we see from this net.

[MikeB]

This is pretty awesome Andrew,, quick question, I use both a laptop with zen3 and a desktop with zen2, will I be able to use the binaries in both machines ( licensing wise) and will I be allowed to download the two different binaries, one for my zen3 laptop and one for my zen2esktop. Always happy to support developers who give to the community.

[AndrewGrant]

This is pretty awesome Andrew,, quick question, I use both a laptop with zen3 and a desktop with zen2, will I be able to use the binaries in both machines ( licensing wise) and will I be allowed to download the two different binaries, one for my zen3 laptop and one for my zen2esktop. Always happy to support developers who give to the community.

Yes the engine binaries and the Network copies belong to you. You may use them on as many of your machines as you please. The download link provided will expire after ~7 days, so you should download it onto both machines, or just be ready to transfer the files between your machines later.

[MikeB]

This is pretty awesome Andrew,, quick question, I use both a laptop with zen3 and a desktop with zen2, will I be able to use the binaries in both machines ( licensing wise) and will I be allowed to download the two different binaries, one for my zen3 laptop and one for my zen2esktop. Always happy to support developers who give to the community.

Yes the engine binaries and the Network copies belong to you. You may use them on as many of your machines as you please. The download link provided will expire after ~7 days, so you should download it onto both machines, or just be ready to transfer the files between your machines later.

Very good, the check is in the mail in a manner of speaking.

[AndrewGrant]

Can we have some informations about the nnue?

Maybe the size of the net is interesting and what style of play can we see from this net.

Ah, yes, that got away from me.

Ethereal is using the HalfKP paradigm, with a 40960x256 -> 512x32x32x1 Network. This is the textbook approach, but with some changes. Firstly, not all weights are quantized to int8 / int16 for the input layer. Instead, the network goes like this: int16_t => int16_t => (int32_t -> float_t) => float_t => float_t. This approach allows us to never have to pack the data downwards, saving many operations, and also lets us take a slightly more expensive approach to the later layers in exchange for massively increased precision. If I eventually add support for AVX (not avx2) machines, it will be a significant gain as AVX does not have 256-bit vector support for integer types in a meaningful way.

During training the Network actually has 43850 input parameters, using a few factorization of the board to aid in training without having tens of billions of positions. In practice, each Net was trained somewhere between 2 and 4 billion positions total, evaluated by Ethereal / Ethereal NNUE. The networks are trained using a modified form of the Adam optimizer, which allows better performance for datasets with extremely sparse input fields. For example, with a Batch Size of 16384, only about 50% of the 43,850 parameters are used on average.

Data generation for a given network takes about 3 weeks, completed on a 104 core machine. From there, processing that data down into a list of FENs and then into the format used by Ethereal's NNTrainer takes another 12 hours or so. Finally, training the actual Network can take a few days, with many stops and starts to drop the learning rate and find a global optima.

The trainer itself is a fully original work, written in C and making use of all 104 threads. It includes some AVX2 and even AVX512 code for use in updating the network parameters. This toolkit was used in training the Halogen networks as well. It is fairly flexible and trying things like HalfKA, changing layer sizes, adding layers, changing activation functions, or adding more factorizers is only a few minutes of effort in the code. It rivals speeds of GPU based trainers, by leveraging massive SMP and efficient implementations.

[emadsen]

Ethereal 13 has been released, with both a free (Standard) version and a commercial NNUE (Efficiently-Updated Neural Network) version which requires an AVX2 compatible system. The free version is available on Github, and offers minor elo gains over the previous version, with increased functionality for analysis. Ethereal continues its legacy of stability and reliability for analysis and engine play

Congrats Andrew! Big achievement. I'm glad you've pursued commercializing your engine. It makes sense. Ethereal offers original analysis with world-class strength. Good to see competition with Komodo, another engine I admire.

Purchased.

[JohnW]

I do not have a credit card (and dont want to get one). So, without the possibility to pay with PayPal or GooglePay, I cannot buy Ethereal 13 and cannot test it.

Just get a pre-paid Visa card at a store.