We need more GPUs to improve the BIG LC0:)

[Hai]

http://157.230.189.191:8080/

clent_linux / client.exe --hostname=http://157.230.189.191:8080 --user=[username] --password=[password]

Does somebody need help on how to donate gpu power?

[Damir]

Why offer more GPU support, when Lc0 networks have not been updated for almost a week ?

[Nordlandia]

Does somebody need help on how to donate gpu power?

What about stockfish Hai?

Someone has to contribute to stockfish concurrently.

[Robert Pope]

Why offer more GPU support, when Lc0 networks have not been updated for almost a week ?

This is a different network. If your GPU can't go towards the Lczero.org net, why not support this one for a bit?

[Hai]

http://157.230.189.191:8080/

clent_linux / client.exe --hostname=http://157.230.189.191:8080 --user=[username] --password=[password]

Does somebody need help on how to donate gpu power?

http://157.230.189.191:6006
http://157.230.189.191:8080/training_runs
http://157.230.189.191:8080/matches/


These new 40 x 256 networks are much better than the old 20 x 256.
40 x 256 is strong but has much lower kn/s.
It needs only ~ 1/3 of your memory.
This way you can avoid memory problems.

Example:
You are using 20 x 256
You have only 16 gb memory
You are doing analysis
Your gpu(s) need 10 minutes to fill the memory
But you would like to do at least 30 minutes of analysis!!!

Solution 1:
Use 20 x 256
Go and pay for 48 gb memory and do your 30 minutes of analysis

Better Solution 2:
Use 40 x 256
Stay with your 16 gb memory, don't pay a lot of money and do your 30 minutes of analysis

Note that 40 x 256 can run 3 times longer which leads to a ~ +75 elo increase!

[Hai]

TCEC game 79
LC0 blundered with 124. Qg1.

My LC0 wants to play Kg5!! or Bd8 or Kh4 or Bf6 or Qh2 depends on depth.

Then I changed to multi pv 20 and saw on 11 Qg1.
Kg5!!, Bd8, Kh4, Bf6, Qh2 are also in multi pv the best moves.

Note that I'm using id 24 the 40 x 256 networks and not tcecs weak 20 x 256!

[AdminX]

http://157.230.189.191:8080/

clent_linux / client.exe --hostname=http://157.230.189.191:8080 --user=[username] --password=[password]

Does somebody need help on how to donate gpu power?

http://157.230.189.191:6006
http://157.230.189.191:8080/training_runs
http://157.230.189.191:8080/matches/


These new 40 x 256 networks are much better than the old 20 x 256.
40 x 256 is strong but has much lower kn/s.
It needs only ~ 1/3 of your memory.
This way you can avoid memory problems.

Example:
You are using 20 x 256
You have only 16 gb memory
You are doing analysis
Your gpu(s) need 10 minutes to fill the memory
But you would like to do at least 30 minutes of analysis!!!

Solution 1:
Use 20 x 256
Go and pay for 48 gb memory and do your 30 minutes of analysis

Better Solution 2:
Use 40 x 256
Stay with your 16 gb memory, don't pay a lot of money and do your 30 minutes of analysis

Note that 40 x 256 can run 3 times longer which leads to a ~ +75 elo increase!

Interesting, I will donate GPU time during my sleeping hours starting tonight.

[Hai]

http://157.230.189.191:8080/

clent_linux / client.exe --hostname=http://157.230.189.191:8080 --user=[username] --password=[password]

Does somebody need help on how to donate gpu power?

http://157.230.189.191:6006
http://157.230.189.191:8080/training_runs
http://157.230.189.191:8080/matches/


These new 40 x 256 networks are much better than the old 20 x 256.
40 x 256 is strong but has much lower kn/s.
It needs only ~ 1/3 of your memory.
This way you can avoid memory problems.

Example:
You are using 20 x 256
You have only 16 gb memory
You are doing analysis
Your gpu(s) need 10 minutes to fill the memory
But you would like to do at least 30 minutes of analysis!!!

Solution 1:
Use 20 x 256
Go and pay for 48 gb memory and do your 30 minutes of analysis

Better Solution 2:
Use 40 x 256
Stay with your 16 gb memory, don't pay a lot of money and do your 30 minutes of analysis

Note that 40 x 256 can run 3 times longer which leads to a ~ +75 elo increase!

Interesting, I will donate GPU time during my sleeping hours starting tonight.

Thank you very much
Elo is increasing very fast.