Lc0 v0.32.0 (Dev)

[trippytreez445]

> This is a custom-built version of the Leela Chess Zero (Lc0) chess engine, compiled from source on June 14, 2025. This build uses the `dag-preview` search algorithm as its default and is built specifically for CUDA 12.


https://www.mediafire.com/file/h0fee2at ... t.zip/file

[Stephen Ham]

Hi Jason,

Does evidence exist that this Leela 32.0 (DEV) iteration is stronger than "the latest stable version of Lc0" - 31.2? Will it be utilized in the next TCEC?

Thanks in advance.

-Steve-

[trippytreez445]

Hi Jason,

Does evidence exist that this Leela 32.0 (DEV) iteration is stronger than "the latest stable version of Lc0" - 31.2? Will it be utilized in the next TCEC?

Thanks in advance.

-Steve-

Hi Steve,

Lc0 v0.32.0-dev was built from the latest development source and includes experimental features like the dag-preview search algorithm. While it may offer improvements in some cases, there’s no official testing or proof that it outperforms the stable release (v0.31.2). This build was made for personal testing and will not be used in TCEC.

[Stephen Ham]

Hi Jason,

I tried to run your Leela 32.0 development, but received an "exception" - the engine won't work.

I have Windows 10x64 and a NVIDIA GeForce RTX 2080 Ti with 65,462MB. Is that incompatible with your Leela?

-Steve-

[Hai]

Does it work on macOS Apple silicone?

[cc2150dx]

Does nothing for me either:

| _ | |
|_ |_ |_| v0.32.0-dev+git.dirty built Jun 14 2025
Search algorithm: dag-preview
go nodes 100
Loading weights file from: C:\Users\Jason\Desktop\lc0-v0.32.0-dev+git/BT4-1740.pb.gz
Weights file has multihead format, updating format flag
Switching to [cuda-fp16]...
CUDA Runtime version: 12.2.0
Latest version of CUDA supported by the driver: 12.9.0
GPU: NVIDIA GeForce RTX 2060
GPU memory: 5.99969 Gb
GPU clock frequency: 1680 MHz
GPU compute capability: 7.5
L2 cache capacity: 3145728

C:\Users\Jason\Desktop\lc0-v0.32.0-dev+git>

[trippytreez445]

Hi Jason,

I tried to run your Leela 32.0 development, but received an "exception" - the engine won't work.

I have Windows 10x64 and a NVIDIA GeForce RTX 2080 Ti with 65,462MB. Is that incompatible with your Leela?

-Steve-

Hi Steve,

The version you tried was built for the RTX 40 series (compute capability 8.9). I’ve now built a new version targeting the RTX 20 series (compute capability 7.5) for you, hopefully it will fix the error you’ve been getting. Keep in mind this version is untested and actually crashed on my system with an RTX 4070.

You can download it here:
https://www.mediafire.com/file/ibnkzhts ... 0.zip/file

Please try it out and let me know how it goes. Also, let me know what errors you’re getting if any.

Best,
Jason

[jcgdijkstra]

Hi Jason,

I tried to run your Leela 32.0 development, but received an "exception" - the engine won't work.

I have Windows 10x64 and a NVIDIA GeForce RTX 2080 Ti with 65,462MB. Is that incompatible with your Leela?

-Steve-

Hi Steve,

A little off topic but I noticed you have the same GPU as I do. I have been competeting LC0 against Stockfish for years and find that, with various networks tried, it's about as a strong as SF13. However, I find no improvements in it's strength of play weather I play LC0 on my RTX 2080 TI or my previous GTX 1050 TI. Naturally, I get more NPS but in game performance it hardly makes a difference. May I ask what is your preferred network for your GPU in terms of strength?

Tnx,
John

[Damir]

https://ci.appveyor.com/project/LeelaChessZero/lc0

[Stephen Ham]

Hi Jason,

I tried to run your Leela 32.0 development, but received an "exception" - the engine won't work.

I have Windows 10x64 and a NVIDIA GeForce RTX 2080 Ti with 65,462MB. Is that incompatible with your Leela?

-Steve-

Hi Steve,

A little off topic but I noticed you have the same GPU as I do. I have been competeting LC0 against Stockfish for years and find that, with various networks tried, it's about as a strong as SF13. However, I find no improvements in it's strength of play weather I play LC0 on my RTX 2080 TI or my previous GTX 1050 TI. Naturally, I get more NPS but in game performance it hardly makes a difference. May I ask what is your preferred network for your GPU in terms of strength?

Tnx,
John

Hi John,

I'm able to use the BT4-spsa-1740 NN.

As a ICCF GM, I analyze potions over relatively longer periods of time. I generally perceive Leela as at least as good as SF, which seems primarily coded for speed chess and standard time-controls. Nonetheless, in positions with complex tactical solutions and endgames, alpha-beta coded engines remain superior.

-Steve-