lkaufman wrote: If so, a gain of 10 or perhaps 12 looks like a good prediction.
We shall see.
lkaufman wrote:
Can you tell us what is the percentage increase in nodes per second you get going from 2.2.2. to 2.3? I wonder if my 10% increase in Linux applies to the Windows version.
2.3 shows a decrease in kn/s for me vs 2.2.2
Not even close Ray. nps in the sse42 for the new Stockfish 2.3 is much faster than for 2.2.2 sse42 exe. Large margin. Not sure how you come up with the nps for each. But the easy and reliable way is to check it in console mode. Just type in - go depth 22, hit the enter key, and wait for it to finish. Then check the other and compare.
When Stockfish will be set to use Gaviota or Nalimov tablebases? This is big (and only, according to me) disadvantage compared with Houdini, Rybka and Critter.
Load both engines in Shredder GUI, and set infinite analysis on the start position. You can see the kn/s of both. That's the way I do it. Stockfish bench seems to be the least reliable thing to do.
So far Stockfish has gotten 4 wins 1 draw and 0 losses vs Shredder 8 at 5 minute chess. The style of play and tactical shots are amazing. It is a Lamborgini and Shredder is a Volkswagen bus.
Thank you,
Tim Frohlick
A New Way Comes Upon Earth.
God is an infinitely variable Constant.
Man marks his ground with ideologies.
Galaxies are the dreidels of God.
War is a punishment for implacability.
Peace flows from forgiveness of sins.
Modern Times wrote:Load both engines in Shredder GUI, and set infinite analysis on the start position. You can see the kn/s of both. That's the way I do it. Stockfish bench seems to be the least reliable thing to do.
I experienced the same under arena. According to the n/s 2.3 seems a little bit slower than 2.2.2 (around 40 knodes/s). In the Stockfish benchmark, 2.3 is a lot faster. But how can you say the benchmark is less reliable? I mean, it could be, but how do you prove it?
So you, or Marco, or anyone could try to explain these differences.
But nodes per second doesn't matter much, if SF 2.3 is playing stronger. I am still waiting for the IPON/CCRL test results
Modern Times wrote:Load both engines in Shredder GUI, and set infinite analysis on the start position. You can see the kn/s of both. That's the way I do it. Stockfish bench seems to be the least reliable thing to do.
I experienced the same under arena. According to the n/s 2.3 seems a little bit slower than 2.2.2 (around 40 knodes/s). In the Stockfish benchmark, 2.3 is a lot faster. But how can you say the benchmark is less reliable? I mean, it could be, but how do you prove it?
So you, or Marco, or anyone could try to explain these differences.
But nodes per second doesn't matter much, if SF 2.3 is playing stronger. I am still waiting for the IPON/CCRL test results
Bench measures mostly midgame positions to a fixed depth, so is quite reliable. Are you analyzing from the start position? It's quite possible the optimizations help more in the mid/endgame. There are more evaluation features in 2.3, so they may be slowing things down a little in the early game.
I haven't profiled things lately though, should probably take a look .