How can I switch off hyperthreading?

[Laskos]

Nice news, Houdini 3 again +20 elo stronger !

Oops, +10% means +7 Elo, not +20 ...

A bit more than 7 points, probably 10-20. It is a double effect: more solutions found and in shorter time.

Kai

[Houdini]

Nice news, Houdini 3 again +20 elo stronger !

Oops, +10% means +7 Elo, not +20 ...

A bit more than 7 points, probably 10-20. It is a double effect: more solutions found and in shorter time.

Kai

One cannot reliably extrapolate tactical test suite results to Elo strength.
Adding more threads will enlarge the width of the search - great for tactical puzzles but not necessarily for real game play.

Robert

[kranium]

the significant improvement in hyperthreading has been known for some time...
Microsoft worked very closely w/Intel during the development of WIndows 7, and significantly beefed up support for HT.

http://www.dailytech.com/Windows+7+to+O ... e15150.htm

a large proportion of the best engines on playchess are now running w/ hyperthreading enabled,
why?, the operators noticed a significant performance improvement

there's no Houdini 'magic' here...regardless of what the author may want you to believe...
all SMP engines should benefit equally (if run on a modern Intel architecture w/ Windows 7)

Houdini: 0 net ELO gain when tested against other SMP opponents on same architecture

[Vinvin]

Houdini: 0 net ELO gain when tested against other SMP opponents on same architecture

How many thousands games did you play to get an error bar < 1 ?

[syzygy]

Houdini: 0 net ELO gain when tested against other SMP opponents on same architecture

How many thousands games did you play to get an error bar < 1 ?

The fact that H2 apparently only gains 8% by HT already disproves his theory. Different engines benefit to different degrees from HT. Add to that the fact that different engines incur different amounts of search overhead when the number of threads doubles.

[Vinvin]

And the last test, pretty convincing from my point of view. I took "Hard test set talkchesscom-2012" by Vincent Lejeune http://talkchess.com/forum/viewtopic.php?t=44531 consisting of 218 hard positions.

Tactical Mode, Hash 2GB, time: 60s/position

4 threads: 131/218 Average time: 11.04s
8 threads: 142/218 Average time: 10.05s

Pretty conclusive, not only significantly more solutions with 8 threads, but 10% faster to solutions. HT with its 27-36% speed-up on my comp seems clearly beneficial.

Kai

As tactical suite is not very accurate to estimate the strength of an engine, I've another system to evaluate the speed : In the starting position, with 8 best lines, measure time to reach depth 22 (it's about 1 min on your PC), make this operation 10 times with and without hyper-threading, make the average for both and report the time here ...

[Laskos]

And the last test, pretty convincing from my point of view. I took "Hard test set talkchesscom-2012" by Vincent Lejeune http://talkchess.com/forum/viewtopic.php?t=44531 consisting of 218 hard positions.

Tactical Mode, Hash 2GB, time: 60s/position

4 threads: 131/218 Average time: 11.04s
8 threads: 142/218 Average time: 10.05s

Pretty conclusive, not only significantly more solutions with 8 threads, but 10% faster to solutions. HT with its 27-36% speed-up on my comp seems clearly beneficial.

Kai

As tactical suite is not very accurate to estimate the strength of an engine, I've another system to evaluate the speed : In the starting position, with 8 best lines, measure time to reach depth 22 (it's about 1 min on your PC), make this operation 10 times with and without hyper-threading, make the average for both and report the time here ...

Well, tactical suite is not very accurate in determining the strength generally, but everything else being equal, time to correct move (time to solution) and the number of correct moves are showing the strength on a given hardware. Again, everything else being equal (and it's not exactly the case, as more threads mean a bit different shape of the tree, but it's pretty close).

Kai

[Lavir]

As tactical suite is not very accurate to estimate the strength of an engine, I've another system to evaluate the speed : In the starting position, with 8 best lines, measure time to reach depth 22 (it's about 1 min on your PC), make this operation 10 times with and without hyper-threading, make the average for both and report the time here ...

That doesn't work so well. Depending on the pruning, SMP randomization and all these things, it is many times a case of which moves are selected first and in what order for the time it takes to reach a certain depth, more than anything else.

The only real way to test an effective increase in strength (if there is indeed one) is the old one: playing a lot of games with and without HT; the problem is that it is very time consuming especially because in this case you cannot test at very short TCs.

[shrapnel]

The only real way to test an effective increase in strength (if there is indeed one) is the old one: playing a lot of games with and without HT;

Exactly what I have done !

However autotune is not the best way to measure real HT impact since the testing of the single positions is too short for HT to really build up fully.

Exactly my sentiments ! The performance gain I get by enabling HT on my overclocked 3930 K is much higher than what can be explained by "autotune" results.
I will post my Fritz Chess Benchmark results when I get home. I don't understand why the Fritz benchmark is being disparaged by some people here. After all, it is used by even respected chess engine testing experts like Sedat Canbaz and many other chess websites.
Later...

[Laskos]

As tactical suite is not very accurate to estimate the strength of an engine, I've another system to evaluate the speed : In the starting position, with 8 best lines, measure time to reach depth 22 (it's about 1 min on your PC), make this operation 10 times with and without hyper-threading, make the average for both and report the time here ...

That doesn't work so well. Depending on the pruning, SMP randomization and all these things, it is many times a case of which moves are selected first and in what order for the time it takes to reach a certain depth, more than anything else.

The only real way to test an effective increase in strength (if there is indeed one) is the old one: playing a lot of games with and without HT; the problem is that it is very time consuming especially because in this case you cannot test at very short TCs.

Agree, and the time taken to have a conclusion is very long, ultra-fast games don't work. For 95% confidence, 20 Elo points are determined in some 500 games. The games have to be at least 10s/move, therefore 20 min/game. Totally 7 days of computer time, without touching the computer, as with 8 threads it runs at 100% CPU. I am unable to do this. I think the second best thing to do is analyzing (hard) test suites, as the time to correct move is the the goal in chess, not depth and other things. The tree is wider with 8 threads than with 4, so the result might be a bit distorted, but I think this is the best I can do without actually playing games.

Kai