Moderator: Ras
.Fully agreedJJJ wrote:I m sorry, but I think Stockfish 6 will be better at 120/40 than Komodo 8.
The Komodo dev who played TCEC is less good than stockfish 6, because stockfish 6 is probably around 20 elo better than the previous one in TCEC final.
So, I don't know how far Komodo 9 have advanced, but based on the only tournament we know with the Komodo 1365, I d say Komodo is not ready.
And if you want to see Komodo 9 be the number one on list, I think you ll need a Komodo 9 with at least 20 / 25 more elo than Stockfish 6.
Code: Select all
1 Stockfish 6 MP : 3505 160 (+ 76,= 81,- 3), 72.8 %
Houdini 4 MP : 20 (+ 5,= 14,- 1), 60.0 %
Gull 3 MP : 20 (+ 6,= 13,- 1), 62.5 %
Komodo 1318 MP : 20 (+ 7,= 13,- 0), 67.5 %
Rybka 4.1 MP : 20 (+ 11,= 9,- 0), 77.5 %
Chiron 2 MP : 20 (+ 16,= 4,- 0), 90.0 %
Hannibal 1.5 MP : 20 (+ 9,= 11,- 0), 72.5 %
Texel 1.05 MP : 20 (+ 11,= 9,- 0), 77.5 %
Fire 4 MP : 20 (+ 11,= 8,- 1), 75.0 %Plain and simple, the hardware listed here isn't good enough. How many times do people need to be told that Komodo thrives with high-core count hardware (scalability tests prove this, and we are not even looking into how Komodo has won two out of the last three TCEC tournaments). Moreover, Komodo is better with Intel CPUs as opposed to AMD. This is what I understood from Larry anyways. Komodo 8 with my 20 core PC beats Stockfish 6 where I play (anywhere from 4 cores to 24 at 120 minute 60 second increment). Stockfish 6 is not 50 Elo better. That is complete garbage.beram wrote:I have gathered some results for SF6 against Komodo 8 at LTC
Just to gave more data about how SF6 really scores against Komodo 8 on these rarely tested longer time controls
In four matches 508 games, SF6 scores 57,1%, which is about 50 ELO better
- SF6 vs K8 at LTC games 508 games, avg 57,1%
TC 60m15s on AMD FX8320 @8cpu 188 games, +46 =120 -22 56,38%
TC 40/40 CCRL @4cpu 70 games +13−5=52 55,70%
TC 30m5s, Tom Casanovas @6cpu 150 games +40 =97 -13 59%
TC 40/20 CEGT @4cpu 100games +24 =67 -9 57,50%
That last Komodo development version that won TCEC was only about 14 Elo stronger than Komodo 8; it was nothing significant. Why do you think that Komodo 9 has yet to be released? It has been a struggle just trying to achieve 30 Elo for the Komodo team.JJJ wrote:I m not surprised at all because when the last dev of Komodo was winning, it was close on TCEC. And it was better than Komodo 8 and on higher time control.
If you look here, Komodo dev is doing close than the last DEV of stockfish :
http://www.clemens-keck.de/livegames/
120 min + 30 ? Would Komodo 8 won Stockfish 6 on my 4 core intel i5-3570 3,40 GHz at this time control ?APassionForCriminalJustic wrote:Plain and simple, the hardware listed here isn't good enough. How many times do people need to be told that Komodo thrives with high-core count hardware (scalability tests prove this, and we are not even looking into how Komodo has won two out of the last three TCEC tournaments). Moreover, Komodo is better with Intel CPUs as opposed to AMD. This is what I understood from Larry anyways. Komodo 8 with my 20 core PC beats Stockfish 6 where I play (anywhere from 4 cores to 24 at 120 minute 60 second increment). Stockfish 6 is not 50 Elo better. That is complete garbage.beram wrote:I have gathered some results for SF6 against Komodo 8 at LTC
Just to gave more data about how SF6 really scores against Komodo 8 on these rarely tested longer time controls
In four matches 508 games, SF6 scores 57,1%, which is about 50 ELO better
- SF6 vs K8 at LTC games 508 games, avg 57,1%
TC 60m15s on AMD FX8320 @8cpu 188 games, +46 =120 -22 56,38%
TC 40/40 CCRL @4cpu 70 games +13−5=52 55,70%
TC 30m5s, Tom Casanovas @6cpu 150 games +40 =97 -13 59%
TC 40/20 CEGT @4cpu 100games +24 =67 -9 57,50%
It should be noted that these listed time controls are still faster than Komodo's optimum time control which is 120 minute 30 second increment. Komodo is all about super long LTC.
Sedat Canbaz has really nice hardware i7 980x 3.33Ghz and i7 920 3.2GhzAPassionForCriminalJustic wrote:Plain and simple, the hardware listed here isn't good enough. How many times do people need to be told that Komodo thrives with high-core count hardware (scalability tests prove this, and we are not even looking into how Komodo has won two out of the last three TCEC tournaments). Moreover, Komodo is better with Intel CPUs as opposed to AMD. This is what I understood from Larry anyways. Komodo 8 with my 20 core PC beats Stockfish 6 where I play (anywhere from 4 cores to 24 at 120 minute 60 second increment). Stockfish 6 is not 50 Elo better. That is complete garbage.beram wrote:I have gathered some results for SF6 against Komodo 8 at LTC
Just to gave more data about how SF6 really scores against Komodo 8 on these rarely tested longer time controls
In four matches 508 games, SF6 scores 57,1%, which is about 50 ELO better
- SF6 vs K8 at LTC games 508 games, avg 57,1%
TC 60m15s on AMD FX8320 @8cpu 188 games, +46 =120 -22 56,38%
TC 40/40 CCRL @4cpu 70 games +13−5=52 55,70%
TC 30m5s, Tom Casanovas @6cpu 150 games +40 =97 -13 59%
TC 40/20 CEGT @4cpu 100games +24 =67 -9 57,50%
It should be noted that these listed time controls are still faster than Komodo's optimum time control which is 120 minute 30 second increment. Komodo is all about super long LTC.
Code: Select all
1 Stockfish 6 MP : 3505 160 (+ 76,= 81,- 3), 72.8 %
Houdini 4 MP : 20 (+ 5,= 14,- 1), 60.0 %
Gull 3 MP : 20 (+ 6,= 13,- 1), 62.5 %
Komodo 1318 MP : 20 (+ 7,= 13,- 0), 67.5 %
Rybka 4.1 MP : 20 (+ 11,= 9,- 0), 77.5 %
Chiron 2 MP : 20 (+ 16,= 4,- 0), 90.0 %
Hannibal 1.5 MP : 20 (+ 9,= 11,- 0), 72.5 %
Texel 1.05 MP : 20 (+ 11,= 9,- 0), 77.5 %
Fire 4 MP : 20 (+ 11,= 8,- 1), 75.0 %
20 games is not statistically significant. It's almost meaningless. As you can see even Houdini 4 and Gull did better than Komodo 1318 on such few games. On the top of that, if I'm not mistaken, Sedat uses Arena GUI which introduces more noise (Arena steals time to the engines and some engines will forfeit much more on time than others while if they had ran into cutechess-cli no loss on time would have occurred. Even discarding the losses on time, how do we make sure Arena stole an even amount of time between engines? We can't, hence the noise introduced...)beram wrote:
Sedat Canbaz has really nice hardware i7 980x 3.33Ghz and i7 920 3.2Ghz
And he run a Gladiators 2015 tournamen recently which he published on Rybka forum and on CSS, but for to me unknown reasons not here
http://www.sedatcanbaz.com/chess/?page_id=1675
The Komodo version 1318 is supposed to b e +10 ELO above Komodo 8 and played in stage 2 of latest TCEC
On TC 60m15s 6core (3core each, ponder on), SF6 beat the hell out of that Komodo version +7 =13 - 0
Individual statistics:
Code: Select all
1 Stockfish 6 MP : 3505 160 (+ 76,= 81,- 3), 72.8 % Houdini 4 MP : 20 (+ 5,= 14,- 1), 60.0 % Gull 3 MP : 20 (+ 6,= 13,- 1), 62.5 % Komodo 1318 MP : 20 (+ 7,= 13,- 0), 67.5 % Rybka 4.1 MP : 20 (+ 11,= 9,- 0), 77.5 % Chiron 2 MP : 20 (+ 16,= 4,- 0), 90.0 % Hannibal 1.5 MP : 20 (+ 9,= 11,- 0), 72.5 % Texel 1.05 MP : 20 (+ 11,= 9,- 0), 77.5 % Fire 4 MP : 20 (+ 11,= 8,- 1), 75.0 %