KataGo V1.5

[Vargo]

40 games test : KG 1.5.0 v. KG 1.4.5

gpu : 1x1080 , network : g170e-b20c256x2-s5303129600-d1228401921 , 1s/move for both, same config file.

KG 1.5.0 wins 22-18 (55%) All games by resignation, no error, no duplicate game.

Probably not a very meaningful test, but ... KG goes on

Stats :

k1.jpg (447.72 KiB) Viewed 16797 times

k2.jpg (442.62 KiB) Viewed 16797 times

[And]

These games are probably on CUDA? it would be interesting to see also on OpenCL ("New in this release is support for FP16 tensor core GPUs in OpenCL, roughly doubling performance." https://github.com/lightvector/KataGo/releases/)

[And]

"Also as of this release, there is a pure-CPU implementation which can be compiled via -DUSE_BACKEND=EIGEN for cmake. There are no precompiled executables for it right now because the implementation is very basic..." https://github.com/lightvector/KataGo/releases/

who knows how, please do it!

[iopq]

"Also as of this release, there is a pure-CPU implementation which can be compiled via -DUSE_BACKEND=EIGEN for cmake. There are no precompiled executables for it right now because the implementation is very basic..." https://github.com/lightvector/KataGo/releases/

who knows how, please do it!

Have you tried OpenCL on the iGPU? It should be way faster than Eigen

[And]

my AMD Athlon X4 950 and AMD Ryzen 5 processors do not have an integrated graphics card. the thing is, I want to use KataGo in the background while doing something else. but when OpenCL is working, something else is not comfortable to do, because slows down! when using cpu-only I can select for example only 2 cores. with OpenCL it doesn't work out for me

[goame]

my AMD Athlon X4 950 and AMD Ryzen 5 processors do not have an integrated graphics card. the thing is, I want to use KataGo in the background while doing something else. but when OpenCL is working, something else is not comfortable to do, because slows down! when using cpu-only I can select for example only 2 cores. with OpenCL it doesn't work out for me

You should buy a strong RTX 3000 GPU.
This gives endless fun.

[And]

my AMD Athlon X4 950 and AMD Ryzen 5 processors do not have an integrated graphics card. the thing is, I want to use KataGo in the background while doing something else. but when OpenCL is working, something else is not comfortable to do, because slows down! when using cpu-only I can select for example only 2 cores. with OpenCL it doesn't work out for me

You should buy a strong RTX 3000 GPU.
This gives endless fun.

if there was extra money, I'd rather buy a T-50. and, no doubt, I would have received much more pleasure!

[iopq]

my AMD Athlon X4 950 and AMD Ryzen 5 processors do not have an integrated graphics card. the thing is, I want to use KataGo in the background while doing something else. but when OpenCL is working, something else is not comfortable to do, because slows down! when using cpu-only I can select for example only 2 cores. with OpenCL it doesn't work out for me

Then just run the processor as an OpenCL device, it will still be faster than the Eigen backend.

[And]

iopq is it really possible? how to do it?

[iopq]

iopq is it really possible? how to do it?

I found one question on SO about this
https://askubuntu.com/questions/821341/ ... plications

if you're on windows I don't know

[And]

iopq is it really possible? how to do it?

I found one question on SO about this
https://askubuntu.com/questions/821341/ ... plications

if you're on windows I don't know

yes, windows. but AMD A8-5600K with integrated graphics

[And]

benchmark leela-zero-0.17-cpuonly-win64 elf_v2 AMD Athlon X4 950
--benchmark -v500 -t4 -w networkfile.gz
1m 11s
10.2 average depth, 19 max depth
396 non leaf nodes, 1.28 average children
506 visits, 176493 nodes, 505 playouts, 8 n/s

benchmark leela-zero-0.17-win64 elf_v2 GT 610
--benchmark -v500 -w networkfile.gz
1m 07s
8.1 average depth, 17 max depth
397 non leaf nodes, 1.29 average children
513 visits, 179993 nodes, 512 playouts, 10 n/s

low-power processor ~ low-power graphics card

[And]

"borrow_tensorflow" version: x1400 speed up from 1.5.0 (70% of libtensorflow backend). (https://github.com/lightvector/KataGo/i ... -670855071)

can someone compile? (windows)

[Fine Mouche]

i try install it i follow this video : https://www.youtube.com/watch?v=uYT-32lCK5k but the neural network i DL is a .zip not a .bin.gz do i need dezip it ? if yes they are some others files not just the .bin.gz, what i do ? (the inner of the .zip : https://zupimages.net/viewer.php?id=20/34/zboc.png)

And what i must do for run the benchmark it's not clear since he already do it before the video and just comment :/

[And]

Fine Mouche https://github.com/lightvector/KataGo#how-to-use