Modern Times wrote:Seems that on the hardware that 99% of us can afford and run, Stockfish is stronger in game-play. But on TCEC like hardware, with long time control, the tables are turned. For analysis purposes that makes Komodo the weapon of choice, but if you are serious about things you'll use at least one other engine as well.
Hello Ray,
A question:
As a serious tester for years now, which would get your vote?
The one strongest on "normal" hardware or the strongest on "ultimate" hardware?
Blitz games ? Gimme a break !
Everyone knows that Komodo can only beat Stockfish in 30 Minutes games and above.
Anything shorter, and Stockfish will win always.
Pointless testing.
Gimme a break .......... cut out your nasty comments .......
Modern Times wrote:Seems that on the hardware that 99% of us can afford and run, Stockfish is stronger in game-play. But on TCEC like hardware, with long time control, the tables are turned. For analysis purposes that makes Komodo the weapon of choice, but if you are serious about things you'll use at least one other engine as well.
Agreed Ray, although 2 or 3 % better is just marginally whether for Stockfish or Komodo in TCEC final
I think that up to 8 cores latest SF is better and above, because of (still) better scaling with more cores above 8, Komodo gets the upperhand
I haven't noticed any difference in win-loss ratio or ELO going from blitz 3m2s to LTC 30m20s between same opponents in my own testing, nor did I notice such on the CEGT, CCRL or FastGM lists, just to mention some
Modern Times wrote:Seems that on the hardware that 99% of us can afford and run, Stockfish is stronger in game-play. But on TCEC like hardware, with long time control, the tables are turned. For analysis purposes that makes Komodo the weapon of choice, but if you are serious about things you'll use at least one other engine as well.
Agreed Ray, although 2 or 3 % better is just marginally whether for Stockfish or Komodo in TCEC final
I think that up to 8 cores latest SF is better and above, because of (still) better scaling with more cores above 8, Komodo gets the upperhand
I haven't noticed any difference in win-loss ratio or ELO going from blitz 3m2s to LTC 30m20s between same opponents in my own testing, nor did I notice such on the CEGT, CCRL or FastGM lists, just to mention some
grts Bram
I'm quite sure that it has been mentioned and agreed many times that Komodo also has a better evaluation function. To simply say it is just better scaling is sort of a slap in the face to both Mark and Larry. I happen to believe that Stockfish also scales very well at VLTC, so I'm not entirely convinced that Komodo can be that much better in that department. Stockfish uses a ton of time which is absolutely the pivotal thing when playing very long games. There was never any reason to assume that it played worse at VLTC; as long as time is used the engines should shine.
Komodo 9.2 absolutely destroys Stockfish 6 from the rating lists that I've seen. Thus I can't see SF development being that much better than K9.3 on mediocre hardware...
beram wrote:First test with Komodo 9.3 is done with one of the strongest latest Stockfish compiles SF 121115IP (Ipman compile)
SF came out as a winner but +20 - 16 is pretty close I would say
Privat testbook of 25 positions 2 runs 100games
Both engines with standard setting, 512 Mb Hash, ponder off
As a serious tester for years now, which would get your vote?
The one strongest on "normal" hardware or the strongest on "ultimate" hardware?
Later.
Well, I'm a very weak chess player myself, so either one is overkill for my purposes. The level of play for both is way over my head.
In both cases, leave it overnight on your 4,6 or 8 core home machine, with multipv perhaps and you'll get an extremely high quality analysis from either of them.
beram wrote:First test with Komodo 9.3 is done with one of the strongest latest Stockfish compiles SF 121115IP (Ipman compile)
SF came out as a winner but +20 - 16 is pretty close I would say
Privat testbook of 25 positions 2 runs 100games
Both engines with standard setting, 512 Mb Hash, ponder off
beram wrote:First test with Komodo 9.3 is done with one of the strongest latest Stockfish compiles SF 121115IP (Ipman compile)
SF came out as a winner but +20 - 16 is pretty close I would say
Privat testbook of 25 positions 2 runs 100games
Both engines with standard setting, 512 Mb Hash, ponder off
6th cycle the latest SF 021215 against Komodo 9.3 (ct=0) on my AMD X6 1090T
This time +21 -12 =67 far more draws but lesser wins for Komodo with contempt set to zero
3 main uses of computerchess:
1).To challenge yourself against. Well, Komodo should be the same as sf at ANY TC.
2).To use for analysis, correspondence etc. Well, we already agree that Komodo is now top.
3).For comp vs comp shorter TC's. No normal consumer should buy it for that purpose. It is anti-intellectual to do so. It's like an irrelevant curiosity, That's fine, but not more than that.
I can't think of any other important value of chess engines.