Komodo or Stockfish or Houdini ?

Sort:
Avatar of peperoguez
Silver-Myth wrote:

I'll wait until Google's Deep Mind Alpha Zero is available.

Lc0 is stronger than A0

Avatar of scythe_makes_right
peperoguez wrote:
Silver-Myth wrote:

I'll wait until Google's Deep Mind Alpha Zero is available.

Lc0 is stronger than A0

Not even close. Lela had about twice as many wins as Stockfish at the most recent computer championship. Alpha Zero won many times the games that Stockfish did in their heads up matchup (despite a very high draw rate).

Avatar of mjhuan

stockfish

 

Avatar of chessroboto

Why no love for Houdini 6 Pro? Is it that useless for analysis and preparation for human play?

Avatar of drmrboss
scythe_makes_right wrote:
peperoguez wrote:
Silver-Myth wrote:

I'll wait until Google's Deep Mind Alpha Zero is available.

Lc0 is stronger than A0

Not even close. Lela had about twice as many wins as Stockfish at the most recent computer championship. Alpha Zero won many times the games that Stockfish did in their heads up matchup (despite a very high draw rate).

Do you know the hardware specifications between A0 and SF8?

Do you know the strength of engine is determined by the power of hardware.

Do you know Stockfish running on home 4 cores PC can be easily crushed by the same Stockfish running on 40 cores PC. ( just e.g).

Avatar of jjupiter6

sound67 wrote:

Stockfish 10 is still the strongest of the "regular engines". Lc0 only makes sense when you invest in a higher-quality graphics card, only then it can show its great strength. I tested Komodo 13 and Komodo 13 MC against Stockfsh in 30 games at tournament level (2h/40m), and Stockfish didn't lose a single game.

What difference does a graphics card make?

Avatar of DiogenesDue
sound67 wrote:

For one reason or another, Lc0 uses the graphics card for calculations. I didn't believe it at first either, but the guys who are really into this "self-learning" engine actually buy graphics cards you would normally use for a hi-def ego shooter. There's a version that works without a dedicated graphics interface (my PC only has an on-board Intel chip), but I tried this and it cannot win against e.g. Komodo, no matter how much processor power you have.

You previous two posters do realize that graphic cards do what they do by having powerful floating point operations processors in them, right?  Well...those GPUs are also good for AI/neural network operations/calculations.  In fact, the processors on the graphics cards, transistor for transistor, are better suited to the task than the processors on the motherboard.

I guess some people might think that "graphics card" means that the cards' resources are somehow limited to graphics only, but those resources are available to any software that can can communicate with the card via a driver.

Avatar of pfren
jjupiter6 έγραψε:

 

sound67 wrote:

 

Stockfish 10 is still the strongest of the "regular engines". Lc0 only makes sense when you invest in a higher-quality graphics card, only then it can show its great strength. I tested Komodo 13 and Komodo 13 MC against Stockfsh in 30 games at tournament level (2h/40m), and Stockfish didn't lose a single game.

 

What difference does a graphics card make?

 

 

Most AI applications (including LC0) use the GPU processing power for their computational needs.

Avatar of drmrboss
jjupiter6 wrote:

 

sound67 wrote:

 

Stockfish 10 is still the strongest of the "regular engines". Lc0 only makes sense when you invest in a higher-quality graphics card, only then it can show its great strength. I tested Komodo 13 and Komodo 13 MC against Stockfsh in 30 games at tournament level (2h/40m), and Stockfish didn't lose a single game.

 

What difference does a graphics card make?

 

 

Because Neural networks requires multiple parallel calculations( programmers use matrix calculations to simulate NN connections), GPU are excel at matrix calculations.



, CPU are good in liner algebric calculations(  long sequense of input and output).

 

 

 

Of course Leela can run in CPU but x50 times slower and it cost approximately -200 to -300 elo weaker. 

Avatar of ivica403

saulo_09 je napisao/la:

I have Lucas chess and Arena..I have say Lucas for me is more appeal  then Arena, is simple and u can  easily compare 2 engines with kibitzers. I use Stockfish 10 and Komodo 9.02.

tray tarasch chess gui..the best for chess ..

Avatar of Warbler
Earth64 wrote:

Which engines do you prefer , Komodo, Stockfish or Houdini ?

What about Fritz or Rybka?

I guess to me it would be a question of what happens in games against each other.

Avatar of drmrboss
Warbler wrote:
Earth64 wrote:

Which engines do you prefer , Komodo, Stockfish or Houdini ?

What about Fritz or Rybka?

I guess to me it would be a question of what happens in games against each other.

You can install in your computer or watch computer chess games.

Stockfish will crush Fritz like 80-20 and Rybka like 90-10 in a 100 games match( Cos they are not  even top 10 engines).

 

In this games, SF beat Komodo 7-1, and Houdini 4-2.

Avatar of Virmus

I use whichever engine is the fastest on my machine. Currently it's Stockfish_10_64_bmi2

Avatar of drmrboss
rychessmaster1 wrote:
drmrboss wrote:
Warbler wrote:
Earth64 wrote:

Which engines do you prefer , Komodo, Stockfish or Houdini ?

What about Fritz or Rybka?

I guess to me it would be a question of what happens in games against each other.

You can install in your computer or watch computer chess games.

Stockfish will crush Fritz like 80-20 and Rybka like 90-10 in a 100 games match( Cos they are not  even top 10 engines).

 

In this games, SF beat Komodo 7-1, and Houdini 4-2.

 

But lost to my girl LEELA

No, Leela is currently no 3. SF beat Leela 4.5 - 3.5.

SF lose to Leelenstein- 3-5

SF drew to Leela Dev T60. 3-3

 

In fact all these 3 leela derivatives and  SF are in very close strength. < 10 games is very small sample size, we probably need minimum 500 games to see the difference in strength.

Avatar of drmrboss
sound67 wrote:

SugaR NN 1.1, a variant of Stockfish 10, is currently the highest rated engine. 

Because CCRL use Stockfish 10, which is 1 year old Stockfish.  Sugar is released approx in April 2019.

If CCRL use latest Stockfish 11 developmental version(+50 elo stronger now), I am pretty sure that SF 11 will be better than sugar.

https://github.com/glinscott/fishtest/wiki/Regression-Tests

Avatar of Virmus

SugaR NN is Stockfish

Avatar of Ronbo710

Love to go Stockfishing happy.png 

Avatar of Hedgehog1963

Using Komodo 11.2.2 I can turn the engine strength down to 1250 strength in Handicap and Fun" mode in Fritz 12.  Gives me a chance.

Avatar of pfren
Hedgehog1963 έγραψε:

Using Komodo 11.2.2 I can turn the engine strength down to 1250 strength in Handicap and Fun" mode in Fritz 12.  Gives me a chance.

 

You can easily turn down the playing strength in almost any UCI engine, Stockfish included. The latter has 20 "skill levels".

Avatar of drmrboss
Howhorseymove wrote:

Has anyone notice though that when you turn down the level that it will play “weird” chess by making a random move that serves no purpose and which a human would not play?

It is ususal to most programs.

 

Programmers has to intentionally add random bad moves to make it weaker. Or they just limit search depth or search nodes to make it weaker.