dragontamer5788 wrote: ↑Tue Aug 13, 2019 1:14 am
Or are all of the cores unnecessarily searching nodes because they don't have any idea of what to actually do?
Nice description of "Lazy SMP".
Well... the description unfortunately applies to ABDADA and YBWC as well. Which is why all of my chess work is entirely focused on trying to discover a work-efficient search methodology.
I hoped that my current methodology would be 100% work efficient, but alas... a fully work efficient methodology would have limited parallelism (average case speedup of fractions of a percent), and require an exponential amount of RAM to store the huge amount of parallel processing (roughly O(ply^(depth/2)) storage costs). Needless to say, that's unworkable as ply-searched grows. Again, MY method has this problem, maybe someone else's future discovery can figure out a way around these problems.
Still, I think I've discovered something that is more work efficient than current methodologies. So I'm willing to code it up and test it. I think I've come up with a better speculative execution heuristic than YBWC or ABDADA.
Either way, these scores are quite conflicting. So if I planned to buy 2x 7742 cpu's what speeds can I expect in SF dev, especially in Windows? Will I get 190 Mn/s? 230 Mn/s? 275 Mn/s?
Either way, these scores are quite conflicting. So if I planned to buy 2x 7742 cpu's what speeds can I expect in SF dev, especially in Windows? Will I get 190 Mn/s? 230 Mn/s? 275 Mn/s?
The high end results were all on Linux and my guess is that Linux Huge pAges were enabled - which make a big big difference.
On a Linux box with a Threadripper 3970x 32C- ( btw the entire computer incljuding RTX 2060 Super with 128 GB ram was 33% less then just one 7742 CPU - getting 2x 7742 is a lot of cheese to throw on something just for chess 8>0 ..)
I think the Patrick @ ServeTheHome.com results on Ipman's web page are pretty close to what you should get on a Linux box, maybe subtract up to about 10% for a Windows box - I really don't know since I don't use Windows.
Werewolf wrote: ↑Sun Aug 11, 2019 9:59 pm
Surpassing Deep Blue's 200 Million nps.
But 22 years later...
But now, Stockfish at 1 Mn/s is stronger than Deep Blue at 200 Mn/s.
Stockfish at 1 kn/s is stronger than Deep Blue.
Is it? I’m not sure that’s right
Stockfish should be GM level even at 100 n/s classical time control.
Probably 0.5 knps at classical time control, if my old inferrence works. Super GM level at some 5 knps.
I often say that Stockfish reach 2600-2700 Elo with 1 second by move on an average home machine (around 2 kn/s) or a fast phone against human on a long game (40 moves in 2h). This view set SF speed around 10 kn/s.
I do not believe this is correct. At 5K nodes per second, you get to search 5K nodes and you are done. Not much of a search. I have the ability in Crafty to search a specific number of nodes (approximately). For a simple test, I took the newer was positions (201 total) and ran them against standard Crafty at 1 second per move on fairly good hardware (60M nodes per second). It correctly solved all 201 positions. I then re-ran the same test, except telling crafty to stop after searching 5K nodes per second (5K nodes total here) it only solved 114 out of the 201 positions. That hardly sounds like a GM to me. If you stretch it to 60 seconds, normal Crafty still gets all 201. Crafty searching 300K nodes (60 seconds x 5K nodes per second) gets 171 right.
That 5K nodes per second is a REAL restriction. Many were doing 5K nodes per second in the 70's and 80's. And far beyond. Without a GM being produced.
I am certain that the 2700 Elo at 2K nodes per second is a wild exaggeration of reality. Maybe 500K nodes per second, possible. Certainly not 2K.
Werewolf wrote: ↑Sun Aug 11, 2019 9:59 pm
Surpassing Deep Blue's 200 Million nps.
But 22 years later...
But now, Stockfish at 1 Mn/s is stronger than Deep Blue at 200 Mn/s.
Stockfish at 1 kn/s is stronger than Deep Blue.
Is it? I’m not sure that’s right
Stockfish should be GM level even at 100 n/s classical time control.
Probably 0.5 knps at classical time control, if my old inferrence works. Super GM level at some 5 knps.
I often say that Stockfish reach 2600-2700 Elo with 1 second by move on an average home machine (around 2 kn/s) or a fast phone against human on a long game (40 moves in 2h). This view set SF speed around 10 kn/s.
I do not believe this is correct. At 5K nodes per second, you get to search 5K nodes and you are done. Not much of a search. I have the ability in Crafty to search a specific number of nodes (approximately). For a simple test, I took the newer was positions (201 total) and ran them against standard Crafty at 1 second per move on fairly good hardware (60M nodes per second). It correctly solved all 201 positions. I then re-ran the same test, except telling crafty to stop after searching 5K nodes per second (5K nodes total here) it only solved 114 out of the 201 positions. That hardly sounds like a GM to me. If you stretch it to 60 seconds, normal Crafty still gets all 201. Crafty searching 300K nodes (60 seconds x 5K nodes per second) gets 171 right.
That 5K nodes per second is a REAL restriction. Many were doing 5K nodes per second in the 70's and 80's. And far beyond. Without a GM being produced.
I am certain that the 2700 Elo at 2K nodes per second is a wild exaggeration of reality. Maybe 500K nodes per second, possible. Certainly not 2K.
All my tests show that SF at 50K/sec - might be somewhere at the GM level - whether it is a Super GM or a more ordinary GM, who knows. I would be interested in hearing from Larry Kaufman on this. The Revelation is a device that reportedly plays at GM level and it see around 100K/nps ( roughly - I do not own one) . Perhaps only a GM knows for sure.
"GM level" means that after a match is played we expect a 50% performance, I doubt any human in the world has any chance of getting close to that at 50K/sec.
Ovyron wrote: ↑Thu Jan 16, 2020 12:42 am
"GM level" means that after a match is played we expect a 50% performance, I doubt any human in the world has any chance of getting close to that at 50K/sec.
Why are you always so angry in your posts? I have no desire to interact with you at all. You definition of GM level is not quite accurate - there is wide range of GMs of level - from the very highest level, to those that are much weaker. Plus my words are couched - they were deliberately couched for people like you. In reality you should be banned from this site, but I guess there is no rule against someone who enjoys being unpleasant to others.
'' .... might be somewhere at the GM level - whether it is a Super GM or a more ordinary GM, who knows. I would be interested in hearing from Larry Kaufman on this. " did I write something in any form that was absolute? " ...might be ..." "would like to hear from Larry K" etc.
I'm sorry , but for my own sanity , i have to block you from me. Have a nice life.
さようなら
Sayōnara
You're overreacting, Mike. I don't think you would have interpreted his post that way if you had not seen his username.
He said that he doubt that any human in the world has any chance of getting a 50% perf against SF at 50Knps. That's not an angry statement, that's the obvious truth.
This talk about constraining SF to a low nps is a contrived way to talk about time odds (the only point of slowing down the engine is to make the human ponder). Now, the reference nps depends on hardware, but taking fishtest's 1.6mnps standard (most recent CPUs do quite better on a single-core), 50knps is about 1:30 time odds. In TCEC QL testing, Stockfish with 1:300 time odds had a big positive score against what are 2800+ 1CPU CCRL engines. Sure, it wasn't on a single-core, but it gives a rough idea of how the massive strength gap can more than overcome the time odds.
So Stockfish with 3 minutes + 1s/move on the clock on one core would easily crush any human having a few hours on the clock.
We know that human are relatively worse at blitz against engines than at classical TC, so while with a sufficiently short TC, you'd get Stockfish on such low node counts that its strength would plumet, the human wouldn't be able to compete at all.
I mean, at bullet 1m+0s SF level 1 on lichess can give trouble to an IM that isn't careful :
Alayan wrote: ↑Thu Jan 16, 2020 2:16 am
You're overreacting, Mike. I don't think you would have interpreted his post that way if you had not seen his username.
He said that he doubt that any human in the world has any chance of getting a 50% perf against SF at 50Knps. That's not an angry statement, that's the obvious truth.
This talk about constraining SF to a low nps is a contrived way to talk about time odds (the only point of slowing down the engine is to make the human ponder). Now, the reference nps depends on hardware, but taking fishtest's 1.6mnps standard (most recent CPUs do quite better on a single-core), 50knps is about 1:30 time odds. In TCEC QL testing, Stockfish with 1:300 time odds had a big positive score against what are 2800+ 1CPU CCRL engines. Sure, it wasn't on a single-core, but it gives a rough idea of how the massive strength gap can more than overcome the time odds.
So Stockfish with 3 minutes + 1s/move on the clock on one core would easily crush any human having a few hours on the clock.
We know that human are relatively worse at blitz against engines than at classical TC, so while with a sufficiently short TC, you'd get Stockfish on such low node counts that its strength would plumet, the human wouldn't be able to compete at all.
I mean, at bullet 1m+0s SF level 1 on lichess can give trouble to an IM that isn't careful :
Perhaps , and if he posted what he said without quoting me, there would been zero reaction from me for sure.
Your comment is interesting regarding the 1/300 time odds to say the least. i was not aware of those tests. Also, I may have a natural bias to understate the strength of Honey for no other reason not to claim something that may not be true. Anyway - it is true, that SF at very low nps , 1000 nodes per second or is is very, very strong.