Velvet Chess Engine

[Graham Banks]

Restarted the 4CPU gauntlet with 4.1.0.
Same link.

[Darius]

Hi Martin,

I would like to thank you sincerely for the wonderful engine that is Velvet. I love looking at and analyzing chess games using it

I have created a post on my website regarding the Velvet engine.
https://www.chessengeria.com/post/velvet-4-1-0-released


I tried compiling different versions of Velvet to test for Apple Silicon processors. I was able to do this for the versions:
1.2.0
3.1.3
3.2.0
The above-mentioned versions of the Velvet engine work correctly.
The 4.1.0 version you shared on GitHub also works great on the MacBook Pro M1


For other versions, the following messages appear when Velvet is started after compilation is complete:









I compiled Velvet using the commands:
cargo build --release --bin velvet


If there is anything I can do differently to make some versions of Velvet work, I would appreciate knowing.
Thanks!

[Graham Banks]

Debug files for 5.1.0 attached.

[mhonert]

If there is anything I can do differently to make some versions of Velvet work, I would appreciate knowing.
Thanks!

Hi Darius,

sorry for the late reply. I just saw your post by chance today.
Thanks a lot for your kind words and the interesting blog post about Velvet!

The issue is related to a compiler optimization regression that started in Rust v1.59.
There was also a GitHub issue (https://github.com/mhonert/velvet-chess/issues/5) about it.
Velvet v5.0.0 and newer versions contain a workaround and should work fine with newer Rust versions.
If you would like to compile an older version of Velvet, you need to use Rust v1.58.1.

If you are using the Rust toolchain installer (rustup), you could for example run "rustup override set 1.58.1" from inside the source folder. That way the old version is only used for the Velvet source folder.

[Darius]

If there is anything I can do differently to make some versions of Velvet work, I would appreciate knowing.
Thanks!

Hi Darius,

sorry for the late reply. I just saw your post by chance today.
Thanks a lot for your kind words and the interesting blog post about Velvet!

The issue is related to a compiler optimization regression that started in Rust v1.59.
There was also a GitHub issue (https://github.com/mhonert/velvet-chess/issues/5) about it.
Velvet v5.0.0 and newer versions contain a workaround and should work fine with newer Rust versions.
If you would like to compile an older version of Velvet, you need to use Rust v1.58.1.

If you are using the Rust toolchain installer (rustup), you could for example run "rustup override set 1.58.1" from inside the source folder. That way the old version is only used for the Velvet source folder.

Martin,

Thank you very much for your reply. It is certainly useful information for me.
The engine you authored - Velvet - is amazing. It is one of several of my favorites. It plays beautifully, effectively and efficiently

I have written about it in several places on my blog:
https://www.chessengeria.com/post/velvet-4-1-0-released
https://www.chessengeria.com/post/velvet-5-0-0-released

Velvet also plays in the MCERL competition: https://www.chessengeria.com/mcerl

[mhonert]

Hi everyone,

today I released Velvet v5.2.0:
https://github.com/mhonert/velvet-chess ... tag/v5.2.0

Estimated strength increase is around 40 Elo in standard chess and 60 Elo in FRC.
Details can be found in the release notes.

Best regards
Martin

[Graham Banks]

Thanks, Martin.

[bastiball]

Hi everyone,

today I released Velvet v5.2.0:
https://github.com/mhonert/velvet-chess ... tag/v5.2.0

Estimated strength increase is around 40 Elo in standard chess and 60 Elo in FRC.
Details can be found in the release notes.

Best regards
Martin

Thanks!

[Gabor Szots]

Hi everyone,

today I released Velvet v5.2.0:
https://github.com/mhonert/velvet-chess ... tag/v5.2.0

Estimated strength increase is around 40 Elo in standard chess and 60 Elo in FRC.
Details can be found in the release notes.

Best regards
Martin

I have submitted my first bunch of games. Stay tuned to this week's update.

[bastiball]

https://www.computerchess.org.uk/ccrl/4 ... lvet_5_2_0

[mhonert]

Thanks a lot for the tests!

[rmangus]

Martin, I wanted to send you a quick message of appreciation for your chess engine "Velvet". I've been using it to analyze my games for about a month now. It finds very creative attacking ideas.

I discovered it through Stefan Pohl's EAS rating list. Velvet 4.1.0 still holds the top of this list, and is capable of extraordinary dynamic play with its ~3280 elo rating.

What inspired you to write Velvet? I'm afraid I don't have the technical knowledge to understand how these engines work, but Velvet has an uncanny ability for creating positional imbalances / complications. Was this your goal?

-Ryan

[mhonert]

Hello Ryan,

thanks a lot for your message!
It's great to hear, that you find my chess engine interesting.
Even though it started as a small hobby project, it's nice to know it's also helpful for chess players to analyze their own games.

But, the initial motivation for creating this engine wasn't really about chess.
I started it mainly to learn a new programming language (Rust) and later to understand more about machine learning and neural networks.
A long time ago, I had spent some time with chess engine development, so I already knew some basics about the search algorithm, move generation, etc.
Back then, my inspiration to create a chess engine came from an article in a german computer magazine.
There was an interview with Christian Donninger about a new version of "Nimzo" and an accompanying introductory article about chess engine development.

The engine's playing style is not the result of some deliberate design process (e.g. to mimic certain chess players), but I think, it rather
emerged from small design decisions I made along the way.
Several times, I had to abandon ideas or revert changes, which increased overall strength, but had a "negative" impact on the playing style.

If you see any areas to improve or features you think are missing, don't hesitate to let me know.

Thanks again for reaching out!

[rmangus]

Hey Martin, thanks for the reply.

Would it be somehow possible to compile a version of Velvet 4.1.0 with the multi-pv analysis feature that was added in 5.0.0?

I still use 4.1.0 for analysis since (as you mentioned) the newer versions have more "objective" evals, which put a damper on Velvets dynamic play.

Maybe an easier alternative would be to compile 5.3.0 with the neural network from 4.1.0? From my limited knowledge changing an engines neural network seems to influence eval the most. Many of the engines I've tested in the past gain strength from larger self play generated networks, but lose style as a result. This way I might get all the latest fixes and features while maintaining a similar style.

Also, I did have some ideas for future releases. I think a UCI limit strength feature would be cool. Many commercial engines have this feature to give players a fighting chance if they want to spar against an engine.

Another feature that might compliment this is the "variety" UCI option that Komodo and several other engines have. It increases the likelihood of an engine diverging from it's best move if it sees several other options within a certain centi-pawn threshold (I assume). This gives more varied games within the same time controls.

I've no idea what's involved in implementing these features, those were just some things that came to mind.

Best of luck in all your endeavors!

[mhonert]

Hi everyone,

today I released Velvet v6.0.0:
https://github.com/mhonert/velvet-chess ... tag/v6.0.0

Estimated strength increase is around 35 Elo in standard chess.
Details can be found in the release notes.

Best regards
Martin

[Gabor Szots]

Hi Martin,

Good news. Tournament just started. Not likely to reach this week's update though.

[mhonert]

Hi everyone,

thank you for the tests! I updated, the readme of the project with the latest results.

Two days ago, I released the new Velvet v7.0.0:
https://github.com/mhonert/velvet-chess ... tag/v7.0.0

The estimated strength increase is around 65 Elo.
Additional details can be found in the release notes.

[Gabor Szots]

Hi everyone,

thank you for the tests! I updated, the readme of the project with the latest results.

Two days ago, I released the new Velvet v7.0.0:
https://github.com/mhonert/velvet-chess ... tag/v7.0.0

The estimated strength increase is around 65 Elo.
Additional details can be found in the release notes.

Hi Martin,

Gauntlet for the blitz list with Velvet 7.0.0 in progress.

[mhonert]

Thanks
Looking forward to the results.