lc0-win-20180512-cuda90-cudnn712-00

[shrapnel]

Hello,

I thought I installed everything needed but couldn't find/locate the 2 following dll's:
cublas64_92.dll
cudart64_92.dll

Where can I find them/download them from ?

rgds

Hi Lion
Just make sure you have the following FIVE files in your Leela CUDA folder, cublas64_92.dll,cudart64_92.dll,
cudnn64_7.dll, lc0.exe and weights.txt.
Simple.
Hope this helps.
Regards

[pohl4711]

Thank you !

I have now found them and copy them to the same folder as LC0. Unfortunately, I still dont see any increase of nps....?
Any idea what I am still missing?

rgds

Nothing. With Cuda 9.2 Leela is only 5% faster, than with Cuda 9.0. No big difference.

I think he meant only 5% faster than the Open CL (non-CUDA) LCZero. So something is wrong.

No. I meant Cuda 9.0/9.2. Both Cuda-Versions are around 4xfaster than the normal GPU-Leela.
Between Cuda 9.0 and 9.2 is only 5% speed-difference. What did you expect from a small Cuda-update???

[Werewolf]

Nothing. With Cuda 9.2 Leela is only 5% faster, than with Cuda 9.0. No big difference.

I think he meant only 5% faster than the Open CL (non-CUDA) LCZero. So something is wrong.

No. I meant Cuda 9.0/9.2. Both Cuda-Versions are around 4xfaster than the normal GPU-Leela.
Between Cuda 9.0 and 9.2 is only 5% speed-difference. What did you expect from a small Cuda-update???

You're missing the point. It is about what LION meant, not you. I think lion means he sees no speedup at all from Open CL.

[pohl4711]

Then he should follow my step-by-manual:


How to run Leela CUDA (should be around 4x faster, than Leela GPU(!)) in Fritz/Shredder GUI:

Important: Leela CUDA needs 3 .dll-files from the NVIDIA-developer-website, otherwise it can not run!
And - of course - you need a NVIDIA CUDA GPU in your PC...

1) Update your Nvidia driver for your GPU to the latest version here: http://www.nvidia.de/Download/index.aspx
After that: reboot your PC.
2) Get LC0.exe for the cuDNN version from: https://crem.xyz/lc0/ and unpack it and put it in your Leela CUDA folder.
3) Install CUDA v9.2 from https://developer.nvidia.com/cuda-92-download-archive (you need an account) and get cublas64_92.dll and cudart64_92.dll from the bin directory of CUDA v9.2 and copy it to your Leela CUDA-folder (Alternatively if you don't want to install CUDA, you can just unpack the installer and search in the unpacked folder and files for the .dlls with <Strg>+F in the Windows-Explorer, that was, what I did). Put the 2 files in your Leela CUDA-folder.
4) Get cuDNN for CUDA 9.2 from https://developer.nvidia.com/rdp/cudnn-download (you also need an account for this) and get cudnn64_7.dll from the bin directory of the zip and extract it and put it in your Leela CUDA-folder.
5) Download the latest network from http://lczero.org/networks and unpack it (Leela CUDA will autodetect it in any form either as weights.txt or however it is named) and put it in your Leela CUDA-folder.
6) Create the UCI-Engine in Fritz/Shredder.

Done.

7) I strongly recommend to change the UCI-settings of Leela CUDA to this:

Scale thinking time = 2.500000
FPU reduction = -0.068300
Cpuct MCTS = 3.168000

Why these values? Because FPU reduction and Cpuct were optimized with the CLOP-tool to these values.
And raising Scale thinking time to 2.5 makes Leela moving slower, which means a better use of the game-time.
In my testing, Leela plays definitly stronger with this settings.

[Werewolf]

Then he should follow my step-by-manual:


How to run Leela CUDA (should be around 4x faster, than Leela GPU(!)) in Fritz/Shredder GUI:

Important: Leela CUDA needs 3 .dll-files from the NVIDIA-developer-website, otherwise it can not run!
And - of course - you need a NVIDIA CUDA GPU in your PC...

1) Update your Nvidia driver for your GPU to the latest version here: http://www.nvidia.de/Download/index.aspx
After that: reboot your PC.
2) Get LC0.exe for the cuDNN version from: https://crem.xyz/lc0/ and unpack it and put it in your Leela CUDA folder.
3) Install CUDA v9.2 from https://developer.nvidia.com/cuda-92-download-archive (you need an account) and get
cublas64_92.dll and cudart64_92.dll from the bin directory of CUDA v9.2 and copy it to your Leela CUDA-folder.
(Alternatively if you don't want to install CUDA, you can just unpack the installer and search in the unpacked
folder and files for the .dlls with <Strg>+F in the Windows-Explorer, that was, what I did). Put the 2 files in
your Leela CUDA-folder.
4) Get cuDNN for CUDA 9.2 from https://developer.nvidia.com/rdp/cudnn-download (you also need an account for this)
and get cudnn64_7.dll from the bin directory of the zip and extract it and put it in your Leela CUDA-folder.
5) Download the latest network from http://lczero.org/networks and unpack it (Leela CUDA will autodetect it in
any form either as weights.txt or however it is named) and put it in your Leela CUDA-folder.
6) Create the UCI-Engine in Fritz/Shredder.

Done.

7) I strongly recommend to change the UCI-settings of Leela CUDA to this:

Scale thinking time = 2.500000
FPU reduction = -0.068300
Cpuct MCTS = 3.168000

Why these values? Because FPU reduction and Cpuct were optimized with the CLOP-tool to these values.
And raising Scale thinking time to 2.5 makes Leela moving slower, which means a better use of the game-time.
In my testing, Leela plays definitly stronger with this settings.

OK that's really helpful - I didn't know this existed.

Thanks!

[shrapnel]

Then he should follow my step-by-manual:


How to run Leela CUDA (should be around 4x faster, than Leela GPU(!)) in Fritz/Shredder GUI:

Important: Leela CUDA needs 3 .dll-files from the NVIDIA-developer-website, otherwise it can not run!
And - of course - you need a NVIDIA CUDA GPU in your PC...

1) Update your Nvidia driver for your GPU to the latest version here: http://www.nvidia.de/Download/index.aspx
After that: reboot your PC.
2) Get LC0.exe for the cuDNN version from: https://crem.xyz/lc0/ and unpack it and put it in your Leela CUDA folder.
3) Install CUDA v9.2 from https://developer.nvidia.com/cuda-92-download-archive (you need an account) and get cublas64_92.dll and cudart64_92.dll from the bin directory of CUDA v9.2 and copy it to your Leela CUDA-folder (Alternatively if you don't want to install CUDA, you can just unpack the installer and search in the unpacked folder and files for the .dlls with <Strg>+F in the Windows-Explorer, that was, what I did). Put the 2 files in your Leela CUDA-folder.
4) Get cuDNN for CUDA 9.2 from https://developer.nvidia.com/rdp/cudnn-download (you also need an account for this) and get cudnn64_7.dll from the bin directory of the zip and extract it and put it in your Leela CUDA-folder.
5) Download the latest network from http://lczero.org/networks and unpack it (Leela CUDA will autodetect it in any form either as weights.txt or however it is named) and put it in your Leela CUDA-folder.
6) Create the UCI-Engine in Fritz/Shredder.

Done.

7) I strongly recommend to change the UCI-settings of Leela CUDA to this:

Scale thinking time = 2.500000
FPU reduction = -0.068300
Cpuct MCTS = 3.168000

Why these values? Because FPU reduction and Cpuct were optimized with the CLOP-tool to these values.
And raising Scale thinking time to 2.5 makes Leela moving slower, which means a better use of the game-time.
In my testing, Leela plays definitly stronger with this settings.

Excellent Tutorial.
It should be saved in Notepad.
Herr Pohl, can you also Post a Tutorial about how to build 20x256 NNs OR point us to the relevant URL ?

[pohl4711]

The step-by-step manual to get Leela Cuda running (and normal Leela GPU in FritzGUI (which needs the polyglot-adapter in older Fritz-versions, because Fritz timeouts engines, which took too much time to initialize)) can be found.on my website in the LC Zero-section.
https://www.sp-cc.de/long-thinkingtime.htm

Creating own networks (20x256) is not my business, sorry. Can not help here.

Stefan (SPCC)

[Werewolf]

Have you done any work getting LC0 running in Aquarium? It doesn't have a command line. The inbetween utility didn't work for me, but I may have done it wrong.

[pohl4711]

Have you done any work getting LC0 running in Aquarium? It doesn't have a command line. The inbetween utility didn't work for me, but I may have done it wrong.

Checkout the manuals for Fritz/ShredderGUI on my website. Fritz has no command-line, too. What works in Fritz/Shredder should work in Aquarium, too. In my 2 manuals, I describe the use of polyglot-adapter, too.
Perhaps you try this out. I do not own Aquarium, so I cant say more...

https://www.sp-cc.de/long-thinkingtime.htm

[Lion]

Then he should follow my step-by-manual:


How to run Leela CUDA (should be around 4x faster, than Leela GPU(!)) in Fritz/Shredder GUI:

Important: Leela CUDA needs 3 .dll-files from the NVIDIA-developer-website, otherwise it can not run!
And - of course - you need a NVIDIA CUDA GPU in your PC...

1) Update your Nvidia driver for your GPU to the latest version here: http://www.nvidia.de/Download/index.aspx
After that: reboot your PC.
2) Get LC0.exe for the cuDNN version from: https://crem.xyz/lc0/ and unpack it and put it in your Leela CUDA folder.
3) Install CUDA v9.2 from https://developer.nvidia.com/cuda-92-download-archive (you need an account) and get cublas64_92.dll and cudart64_92.dll from the bin directory of CUDA v9.2 and copy it to your Leela CUDA-folder (Alternatively if you don't want to install CUDA, you can just unpack the installer and search in the unpacked folder and files for the .dlls with <Strg>+F in the Windows-Explorer, that was, what I did). Put the 2 files in your Leela CUDA-folder.
4) Get cuDNN for CUDA 9.2 from https://developer.nvidia.com/rdp/cudnn-download (you also need an account for this) and get cudnn64_7.dll from the bin directory of the zip and extract it and put it in your Leela CUDA-folder.
5) Download the latest network from http://lczero.org/networks and unpack it (Leela CUDA will autodetect it in any form either as weights.txt or however it is named) and put it in your Leela CUDA-folder.
6) Create the UCI-Engine in Fritz/Shredder.

Done.

7) I strongly recommend to change the UCI-settings of Leela CUDA to this:

Scale thinking time = 2.500000
FPU reduction = -0.068300
Cpuct MCTS = 3.168000

Why these values? Because FPU reduction and Cpuct were optimized with the CLOP-tool to these values.
And raising Scale thinking time to 2.5 makes Leela moving slower, which means a better use of the game-time.
In my testing, Leela plays definitly stronger with this settings.

Thank you so much for your time and help.
Unfortunately I still don’t see any nps improvement.
Also the engine seems a little buggi as when analyzing the depth seems to be at around 1 or 2 instead of I.e. 23