LCZero is using my cores, not my GPU.

[duncan]

Probably makes little sense as I know little about either game. I was attempting to compare elo gains from leela zero to lczero.
lczero gets about 30 per day. do you know how many leela zero gets a day.?

10600 Elo gained in about 160 days would be about 66.

But I don't think such a comparison makes much sense.

and what about elo gains against other go programs?

[Gian-Carlo Pascutto]

and what about elo gains against other go programs?

There aren't enough to be useful as a baseline, Leela Zero is far stronger than anything else that is publicly available.

[Vinvin]

The last line of the report was this:
(1269/5310) KWG=32 KWI=2 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=16 SA=0 SB=0 STRM=1 STRN=1 VWM=8 VWN=2 0.7962 ms (10.5 GFLOPS)

How do I start up with those settings?

They're saved in the leelaz_opencl_tuning file and remembered.

All for naught, it is clear to me that my GPU is not being used.
The end-around I found on the net failed, and my back-off to an older driver is hung for an hour now.

I guess it is time to buy a nice Nvidia card.

Some numbers as references for the speed :
On my old GTX 750 Ti : 0.7 Kn/s (700 n/s)
On a GTX 1080 Ti : around 2.5 Kn/s
On the TCEC server (43 cores @2.8 GHz) : 5 Kn/s

2 live video to watch LC0 playing :
https://www.twitch.tv/ccls
and :
https://www.twitch.tv/tcec_chess_tv

[duncan]

and what about elo gains against other go programs?

There aren't enough to be useful as a baseline, Leela Zero is far stronger than anything else that is publicly available.

so leela zero is stronger than Finearts ?

https://qz.com/936654/googles-alpha-go- ... r-fineart/

FineArts first encountered Ke on foxwq.com during November, taking a win and a loss. But after an update in February, the algorithm has since won 10 straight games (link in Chinese) against the world’s No. 1 player.

As Tencent’s tech blog explains (link in Chinese), FineArts works in a similar way to AlphaGo. Both AIs comprise two computer systems modeled on the human brain, which can be trained on large data sets. One part of the system, the “policy network,” predicts which of the possible moves are the likeliest to be played. The other, the “value network,” then evaluates which of those is likeliest to win.

[Gian-Carlo Pascutto]

and what about elo gains against other go programs?

There aren't enough to be useful as a baseline, Leela Zero is far stronger than anything else that is publicly available.

so leela zero is stronger than Finearts ?

I said "publicly available".

We have no way to test against Fineart. Of course, they are free to setup Leela Zero in a handicapped manner in a private match and claim superiority. This is an approach that has been proven successful in the past and is hence likely to be tried again.

[Milos]

Of course, they are free to setup Leela Zero in a handicapped manner in a private match and claim superiority. This is an approach that has been proven successful in the past and is hence likely to be tried again.

Lol, Deep Mind rings a bell .

[op12no2]

My CPU training is working. Thought I'd try the GPU (it's not up to much but I could upgrade it) and get this:-

Code: Select all

Args: [C:\Users\Colin\Projects\leela\gpu-win/lczero --weights=networks\bb479ee94f3fb39fefb5a5b6db75bd57fd614dbd275424173
0d7cb377ebb5f68 -t1 --randomize -n -v800 -lC:\Users\Colin\Projects\leela\gpu-win/logs-274216/20180425162018.log --start=
train 274216-0 1]                                                                                                       
Logging to C:\Users\Colin\Projects\leela\gpu-win/logs-274216/20180425162018.log.                                        
Using 1 thread(s).                                                                                                      
Detecting residual layers...v2...128 channels...10 blocks.                                                              
Initializing OpenCL.                                                                                                    
Detected 1 OpenCL platforms.                                                                                            
Platform version: OpenCL 1.1 CUDA 6.5.51                                                                                
Platform profile: FULL_PROFILE                                                                                          
Platform name:    NVIDIA CUDA                                                                                           
Platform vendor:  NVIDIA Corporation                                                                                    
Device ID:     0                                                                                                        
Device name:   GeForce 210                                                                                              
Device type:   GPU                                                                                                      
Device vendor: NVIDIA Corporation                                                                                       
Device driver: 342.01                                                                                                   
Device speed:  1402 MHz                                                                                                 
Device cores:  2 CU                                                                                                     
Device score:  1111                                                                                                     
Selected platform: NVIDIA CUDA                                                                                          
Selected device: GeForce 210                                                                                            
with OpenCL 1.1 capability.                                                                                             
Loaded existing SGEMM tuning.                                                                                           
Wavefront/Warp size: 32                                                                                                 
Max workgroup size: 512                                                                                                 
Max workgroup dimensions: 512 512 64                                                                                    
BLAS Core: Piledriver                                                                                                   
id name lczero v0.7                                                                                                     
id author The LCZero Authors                                                                                            
option name Threads type spin default 1 min 1 max 4                                                                     
option name Quiet type check default false                                                                              
option name FPU Reduction type string default 0.000000                                                                  
option name Puct type string default 0.850000                                                                           
option name SlowMover type spin default 89 min 1 max 2147483647                                                         
option name Go Nodes Visits type check default true                                                                     
uciok                                                                                                                   
Created dirs data-274216-0                                                                                              
Found 0 existing chunks in data-274216-0/training                                                                       
2018/04/25 16:20:24 exit status 3221226505

Any ideas about the last line error; it's something to do with the output chunker whatever that is...?

It was a clean install on Windows via the zip from the Leela website, then running:-

.\client.exe -debug

[Albert Silver]

The last line of the report was this:
(1269/5310) KWG=32 KWI=2 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=16 SA=0 SB=0 STRM=1 STRN=1 VWM=8 VWN=2 0.7962 ms (10.5 GFLOPS)

How do I start up with those settings?

They're saved in the leelaz_opencl_tuning file and remembered.

All for naught, it is clear to me that my GPU is not being used.
The end-around I found on the net failed, and my back-off to an older driver is hung for an hour now.

I guess it is time to buy a nice Nvidia card.

Some numbers as references for the speed :
On my old GTX 750 Ti : 0.7 Kn/s (700 n/s)
On a GTX 1080 Ti : around 2.5 Kn/s
On the TCEC server (43 cores @2.8 GHz) : 5 Kn/s

2 live video to watch LC0 playing :
https://www.twitch.tv/ccls
and :
https://www.twitch.tv/tcec_chess_tv

On my laptop I am getting about 2k:

26 01:05 131k 2k +0.20 e2-e4 e7-e5 Ng1-f3 d7-d6 Nb1-c3 Ng8-f6 h2-h3 Bf8-e7 d2-d4 e5xd4 Nf3xd4 O-O Bc1-f4 d6-d5 Nd4-b5 Nb8-a6 e4xd5 c7-c6 d5xc6 b7xc6 Qd1xd8

I ran the start position for a minute in Arena and this is what it gave. Hardware is i7-6600QM (16GB) and GTX980M (8GB)

[Gian-Carlo Pascutto]

Device vendor: NVIDIA Corporation
Device driver: 342.01
...
Selected platform: NVIDIA CUDA
Selected device: GeForce 210
with OpenCL 1.1 capability.

Are you able to update your video driver? No idea if that's the cause, but this is extremely old.

[op12no2]

Are you able to update your video driver? No idea if that's the cause, but this is extremely old.

I'll give it a whirl.

If I was to buy a video card just to train Leela, what would be recommend? I can get it through work; so may as well.