Leela Chess Zero and what is the "Best" Graphics Card.

[mwyoung]

I dislike the sloppiness of the test and the sweeping conclusions derived from it.

"Just completed a test match playing Lc0 Cudnn fp-16 vs Lc0 Cudnn. At blitz time controls. And the results have some implications if more testing confirms the results."

I guess either you don't know how to read, or you are just trolling.

I made no sweeping conclusions. Just made a observation and if true and confirmed the implications.

It looks like people have looked into this, and some feel it is true, and Lc0 scales very badly with better hardware.

[Hugo]

what worth is the best graphics card, if the owner is not able to understand what it means to play leela (2threads) vs Stockfish (32threads) ponder ON, at a machine which only has 32 threads in total?
For Fake news of course its always good to have the biggest and fastest

C.K.

[mwyoung]

what worth is the best graphics card, if the owner is not able to understand what it means to play leela (2threads) vs Stockfish (32threads) ponder ON, at a machine which only has 32 threads in total?
For Fake news of course its always good to have the biggest and fastest

C.K.

You are the one posting fake news. How can it be fake. If I post all the conditions of a test, and the results. And test with ponder on, and off. Stream the test, and record the test. For all to examine.

If you don't like the results, post your own data. And do you own test. And stop trolling.....

[corres]

Assuming the same CPU frequency we know Stockfish yields only some tens Elo if we grow the number of physical cores from 16 cores to 32 cores, to 64 cores... meanwhile the nps grows together with near the number of cores. The enhancement in Elo is more lower if we use the virtual cores (HT) to grow the number of cores.
It looks like Leela (and other NN engine) powered by GPU does not behave in such a manner and its Elo changes more drastically in function of the power of GPU and nps of Leela respectively. The main issue is we do not know exactly how behave Leela when we use a more weaker hardware for testing. But it is obvious if we choose the same ratio to decrease the nps of Stockfish as it is the decrease in nps of Leela the yielded weaker system does not model the more stronger system and Stockfish gets a lot of Elo benefit over Leela.

[Laskos]

I dislike the sloppiness of the test and the sweeping conclusions derived from it.

"Just completed a test match playing Lc0 Cudnn fp-16 vs Lc0 Cudnn. At blitz time controls. And the results have some implications if more testing confirms the results."

I guess either you don't know how to read, or you are just trolling.

I made no sweeping conclusions. Just made a observation and if true and confirmed the implications.

Well, your sloppy test is not trolling, but as several people started to make conclusions based on it, and you yourself here, in this very post, state and reinforce their conclusions, based on your junk test, I felt obliged to spend 2 hours to make things a bit clearer.

It looks like people have looked into this, and some feel it is true, and Lc0 scales very badly with better hardware.

So, based on your test, could you sell your useless RTX 2080Ti at a discount price to me? You will fare very well with a second hand $150 price GTX 980Ti, as Lc0 anyway scales very badly with these gimmicks.

[mwyoung]

I dislike the sloppiness of the test and the sweeping conclusions derived from it.

"Just completed a test match playing Lc0 Cudnn fp-16 vs Lc0 Cudnn. At blitz time controls. And the results have some implications if more testing confirms the results."

I guess either you don't know how to read, or you are just trolling.

I made no sweeping conclusions. Just made a observation and if true and confirmed the implications.

Well, your sloppy test is not trolling, but as several people started to make conclusions based on it, and you yourself here, in this very post, state and reinforce their conclusions, based on your junk test, I felt obliged to spend 2 hours to make things a bit clearer.

It looks like people have looked into this, and some feel it is true, and Lc0 scales very badly with better hardware.

So, based on your test, could you sell your useless RTX 2080Ti at a discount price to me? You will fare very well with a second hand $150 price GTX 980Ti, as Lc0 anyway scales very badly with these gimmicks.

Still trolling...

This is why talkchess.com is a virtual ghost town today. Compared to what it use to be.

"In terms of Price and performance, with Lc0 being your only consideration."

So I guess you can read and you are just trolling...

Maybe you should get a better job, and buy one yourself. I am not in the computer hardware selling discount business.

[Hugo]

what worth is the best graphics card, if the owner is not able to understand what it means to play leela (2threads) vs Stockfish (32threads) ponder ON, at a machine which only has 32 threads in total?
For Fake news of course its always good to have the biggest and fastest

C.K.

You are the one posting fake news. How can it be fake. If I post all the conditions of a test, and the results. And test with ponder on, and off. Stream the test, and record the test. For all to examine.

If you don't like the results, post your own data. And do you own test. And stop trolling.....

you post the conditions, yes, but the conditions are wrong.
I post a fact ...32 theads, and you give 34 for simultaneous usage....you are the troll.
it would be like giving 6 threads each engine on a 8 threads machine and let them ponder.
But you simle ignore.
you dont care about right or wrong.
you ignore simple physical facts by your flat "knowledge" that is based on what you dream
this is dumb !
C.K.

[mwyoung]

what worth is the best graphics card, if the owner is not able to understand what it means to play leela (2threads) vs Stockfish (32threads) ponder ON, at a machine which only has 32 threads in total?
For Fake news of course its always good to have the biggest and fastest

C.K.

You are the one posting fake news. How can it be fake. If I post all the conditions of a test, and the results. And test with ponder on, and off. Stream the test, and record the test. For all to examine.

If you don't like the results, post your own data. And do you own test. And stop trolling.....

you post the conditions, yes, but the conditions are wrong.
I post a fact ...32 theads, and you give 34 for simultaneous usage....you are the troll.
it would be like giving 6 threads each engine on a 8 threads machine and let them ponder.
But you simle ignore.
you dont care about right or wrong.
you ignore simple physical facts by your flat "knowledge" that is based on what you dream
this is dumb !
C.K.

I test under all conditions. And what does this have to do with anything in this thread. The results were tested with ponder off.
If you don't like the conditions. Post your own data. Who is stopping you... All you want to do is troll.

https://www.youtube.com/watch?v=Q2n-nuE7Z2s

[Hugo]

its easy to escape arguments be accusing me to troll (whatever that means to you)

I know about what I am talking
I have been with computers and engines at the times were crafty and comet were plaiyng at FICS.
This is more than 20 years ago.
I have had and still have strongly equipped hardware, upto 32 cpu / 64 threads with 256GB RAM.
I did many tournaments and posts about that with strong hardware and long time controlls.
I was 3 years involved in the Doctor title of Kai Himstedt about cluster computing.
2006 I was 3rd at the ICGA world championships.
I was from the beginning of Komodo with Don Daily involved in intensive testing for his developements for many many years.
So I think I know what I am talking about, when I say 2 threads for Leela and 32 threads for SF with ponder ON is completely wrong!
You ignore that
I will continue

C.K.

what worth is the best graphics card, if the owner is not able to understand what it means to play leela (2threads) vs Stockfish (32threads) ponder ON, at a machine which only has 32 threads in total?
For Fake news of course its always good to have the biggest and fastest

C.K.

You are the one posting fake news. How can it be fake. If I post all the conditions of a test, and the results. And test with ponder on, and off. Stream the test, and record the test. For all to examine.

If you don't like the results, post your own data. And do you own test. And stop trolling.....

you post the conditions, yes, but the conditions are wrong.
I post a fact ...32 theads, and you give 34 for simultaneous usage....you are the troll.
it would be like giving 6 threads each engine on a 8 threads machine and let them ponder.
But you simle ignore.
you dont care about right or wrong.
you ignore simple physical facts by your flat "knowledge" that is based on what you dream
this is dumb !
C.K.

I test under all conditions. And what does this have to do with anything in this thread. The results were tested with ponder off.
If you don't like the conditions. Post your own data. Who is stopping you... All you want to do is troll.

https://www.youtube.com/watch?v=Q2n-nuE7Z2s

[mwyoung]

its easy to escape arguments be accusing me to troll (whatever that means to you)

I know about what I am talking
I have been with computers and engines at the times were crafty and comet were plaiyng at FICS.
This is more than 20 years ago.
I have had and still have strongly equipped hardware, upto 32 cpu / 64 threads with 256GB RAM.
I did many tournaments and posts about that with strong hardware and long time controlls.
I was 3 years involved in the Doctor title of Kai Himstedt about cluster computing.
2006 I was 3rd at the ICGA world championships.
I was from the beginning of Komodo with Don Daily involved in intensive testing for his developements for many many years.
So I think I know what I am talking about, when I say 2 threads for Leela and 32 threads for SF with ponder ON is completely wrong!
You ignore that
I will continue

C.K.

Good for you....

I have been doing computer chess testing since 1982. And posting results online since the 1990's.

I started testing chess computers. Because back in the day. Most of the companies lied about the results, and the strength of their chess products.

With you vast experience, and equipment you need to post more data here. The more test data the better...

I post data as one more data point...

My opinions don't have to be your opinions. My conclusions don't have to be your conclusions.

I find it offensive that trolls have made this place a ghost town. To what it was at the beginning.

what worth is the best graphics card, if the owner is not able to understand what it means to play leela (2threads) vs Stockfish (32threads) ponder ON, at a machine which only has 32 threads in total?
For Fake news of course its always good to have the biggest and fastest

C.K.

You are the one posting fake news. How can it be fake. If I post all the conditions of a test, and the results. And test with ponder on, and off. Stream the test, and record the test. For all to examine.

If you don't like the results, post your own data. And do you own test. And stop trolling.....

you post the conditions, yes, but the conditions are wrong.
I post a fact ...32 theads, and you give 34 for simultaneous usage....you are the troll.
it would be like giving 6 threads each engine on a 8 threads machine and let them ponder.
But you simle ignore.
you dont care about right or wrong.
you ignore simple physical facts by your flat "knowledge" that is based on what you dream
this is dumb !
C.K.

I test under all conditions. And what does this have to do with anything in this thread. The results were tested with ponder off.
If you don't like the conditions. Post your own data. Who is stopping you... All you want to do is troll.

https://www.youtube.com/watch?v=Q2n-nuE7Z2s