News about Rebel or Chess System Tal ?

Are there any news regarding the current development of Rebel and Chess System Tal ?

Any "speculations" whether Niemann has cheated or not have not much to do with computer chess in my opinion.

I would be especially happy to test the new chess engine from Chris. I hope that this will be the case sometime.

I am at +35 compared to R15.1, need at least 15 more for a release.

Currently at 80-90, heading for ~100. It will be the last Rebel anyway. Thereafter I will keep working on nnue development in favor of Chess System Tal.

I'm excited about Rebel's continued development. You will have your reasons for not continuing this development. I assume that you and Chris will manage to further improve CS Tal.

I hope that CS Tal will eventually be available and not a "legend" like Ferret.

Dio wrote:

You will have your reasons for not continuing this development.

Sure, basically because I got the most out of the Toga search improving it with 80-100 elo. Furthermore the Toga source code is unstable likely because it was developed by a different compiler (GCC) by various other people. Rebel is developed under Visual Studio from Microsoft, it compiles error free but also comes with alarming warnings. A few examples:

1. Not able to do a full optimization of the code for maximum speed, the engine then often will rush to its maximum depth and play a bad move.

2. Go back to previous versions who had full Polyglot book support. It played legal moves from the book that were not in the book. I had to remove the code.

In other words, it's time to give up, small miracle I got that far.

I still like to do nuue evaluation, hence in favor of Chris, after all I owe him a lot.

Chris was a few years earlier in the business of computerchess and delivering software for homecomputers in “UK” and europe then Ed.

It was a time where machine hardware was a limited resource.
Ram and rom space was small.
CPUs slow.

People like David Broughton and chris w. , richard Lang, thomas Nitsche+Elmar Henne, Martin Bryant, later Ed too, concentrated on intelligent strategies to overcome these limitations.

B strategy or tapered foward pruning was one of those methods.
With the hardware development making huge progress brute force ideas were able to fight back.

The next (and last) Rebel will be the next step in the cooperation between Chris and me. We have been working very hard to replace Toga and port the code base of Chess System Tal instead. The advantages of this move are, 1) clean compiles, 2) a stronger search and last but not least we now work together on creating stronger nets and I can do what I like most (and was my only goal to return to computer chess) getting the most out of NNUE programming.

On the latter, the last Rebel will come with the same version 10 net architecture as Rebel 15.1 but much improved, peppered with special HCE ProDeo evaluation stuff.

In the meantime Chris has created another even more brilliant net architecture, version 20 [!], currently already 30-35 elo stronger than net version 10, with no end in sight.

So this will be the new reality, 1+1 = 3, team work.

I like to present the first beta result.

Code:

No. Name                 Win Draw Loss Unf.  Score Games       %
----------------------------------------------------------------
  1 BETA-ONE           +1525 =4180 -1295 *0 3615.0  7000   51.6%
  2 Revenge_3.0         +230 =650 -120   *0  555.0  1000   55.5%
  3 Berserk_10          +292 =521 -187   *0  552.5  1000   55.2%
  4 Koivisto_8.0        +200 =611 -189   *0  505.5  1000   50.5%
  5 Ethereal_13.75-NNUE +185 =605 -210   *0  487.5  1000   48.8%
  6 Seer_2.6.0          +167 =614 -219   *0  474.0  1000   47.4%
  7 RubiChess_20220813  +122 =605 -273   *0  424.5  1000   42.5%
  8 SlowChess_2.9        +99 =574 -327   *0  386.0  1000   38.6%

No idea how much elo that is.

But net version 20 beats them all.

Looking great.. you know that i any time can test your engine for my lists.. good luck further!

Ipman.

Ipmanchess wrote:

Looking great.. you know that i any time can test your engine for my lists.. good luck further!

Ipman.

Bonjour!

Chess System Tal
Pre-release version for testers will be available via GitHub soonishly. It’s about 25 years since I last did a release, and the procedure then (commercial days, and with distribution) was to ship the release under embargo of a few days, that way no distribution company got any advantage over any other and they all could begin sales on the same day in that knowledge.
I’m proposing something similar with testers, an information embargo for one week. Does that sound about right? I’ll need tester GitHub handles to give you early access.

If I understood it correctly, as a tester you have to have your own github account to be able to test the pre-release version.

I guess, that no tester has such an account.

Since all testers are actually registered in the forum (Graham, Stefan, Ipman, Werner, Thorsten, me..),
I would rather prefer a download link via PM...

I have a github account from betatesting cstal from February 2022.
Its Mclanecxantia.
Just create one jörg burwitz alias diogenes.

Can't wait... Great news! Both engines!

Dio wrote:

If I understood it correctly, as a tester you have to have your own github account to be able to test the pre-release version.

I guess, that no tester has such an account.

Since all testers are actually registered in the forum (Graham, Stefan, Ipman, Werner, Thorsten, me..),
I would rather prefer a download link via PM...

Okay, I can cope with that, but I was generally going to use GitHub to distribute, I guess you can still get on to a public GitHub page without an account, but, then so can everybody else, and that means testers don't have a week or so of exclusive sight.

matejst wrote:

Can't wait... Great news! Both engines!

AVX2?

avx2 ,avx512 ,vnni256 ,vnni512

Unfortunately, not yet, Chris. And not soon, I am afraid. Ipman could perhaps make an SSE compile, if you allow it.

matejst wrote:

Unfortunately, not yet, Chris. And not soon, I am afraid. Ipman could perhaps make an SSE compile, if you allow it.

It won’t be open source. It would have to be me that converted all the simd, and I have no SSE nor AVX512 to development test on. Well, I sort of have SSE, but the work to convert suffer from manana-itis

Most programmers don't have a system who handle avx512 ,but still most already can create one!
I tell them make a few ,i test them out and see if there is any gain in speed.
You can check on OpenBench where different programmers test their engine ,all open source and can check how they have done this..

If it's not possible for you ,then i think a pop & avx2 version would be a good start.

These days it's prefered to have a cpu that handle at least avx2/bmi2 to run these engines with net
but i test also on a little laptop and they just doing well ..okay a bigger slowdown but the ranking order from these engines is still same.

Ipman.

matejst wrote:

Unfortunately, not yet, Chris. And not soon, I am afraid. Ipman could perhaps make an SSE compile, if you allow it.

What does a laptop (new or refurbished) costs where you live? I like to donate 100 euro.

Ipmanchess wrote:

Most programmers don't have a system who handle avx512 ,but still most already can create one!
I tell them make a few ,i test them out and see if there is any gain in speed.
You can check on OpenBench where different programmers test their engine ,all open source and can check how they have done this..

If it's not possible for you ,then i think a pop & avx2 version would be a good start.

These days it's prefered to have a cpu that handle at least avx2/bmi2 to run these engines with net
but i test also on a little laptop and they just doing well ..okay a bigger slowdown but the ranking order from these engines is still same.

Ipman.

That's because the open source community guys are mostly using the same or variants of the same code, and that's all been nicely done, interconvertable, SSE, AVX2, AVX512, by SF/Sopel. Whereas mine is hand-written using mostly AVX2 simd instructions and its all a bit messy. Okay, it's not impossible to convert mine to SSE and AVX512 but wide register programming requires a different mindset and mine is flipped back to "normal" for a while now, so the thought leads to instant pro-crastination and do something else instead.

Admin wrote:

matejst wrote:

Unfortunately, not yet, Chris. And not soon, I am afraid. Ipman could perhaps make an SSE compile, if you allow it.

What does a laptop (new or refurbished) costs where you live? I like to donate 100 euro.

Thank you very much for such a gracious offer, Ed, but you know I can't accept. I already bought a refurbished laptop, but I did not check if it had avx2 beforehand -- and, alas, it does not.

matejst wrote:

Admin wrote:

matejst wrote:

Unfortunately, not yet, Chris. And not soon, I am afraid. Ipman could perhaps make an SSE compile, if you allow it.

What does a laptop (new or refurbished) costs where you live? I like to donate 100 euro.

Thank you very much for such a gracious offer, Ed, but you know I can't accept. I already bought a refurbished laptop, but I did not check if it had avx2 beforehand -- and, alas, it does not.

We will make you an offer you cannot refuse!

A laptop for testing is IMO contraproductive. You can never exchange components and its also difficult with an nvidia graphic card.

Lc0 needs a good GPU.

a laptop is quite suitable for tests, especially since it consumes much less energy. My tests run on a laptop. That Lc0 requires a powerful graphics card is not correct, 90 percent of my tests with Lc0 run on CPU with the DNNL version (1 thread).

This might be true jörg. I also have a low energy laptop with 8 cores and the ryzen GPU can do lc0 too.
But my point is that a normal pc is more flexible.

You can at any time exchange any component.
Change graphic card. Change HDD. Change motherboard  or CPU or whatever.