Four examples comparing HT

[jpqy]

Try these Hardends2013!

http://chessengine.blogspot.it/p/hardte ... imits.html

JP.

[Laskos]

..., but if you are overclocking, it heats more.

If you are overclocking you get more than 10% extra clock when HT is off and now you have the same problem again.

Besides that, HT when using all core with one engine might(!) be ok, when using more than one engine at a time (worst is just single core engines) you run into problems as the load is not nessesarily distributed equaly.

The best would certainly be if there would be no HT in CPU. Things would be clear and easy!

Bye
Ingo

The best is to use HT, but engines on thread number equal to that of physical cores.
That way you have room for other easy on CPU tasks.

[shrapnel]

30Mn/s!! nice machine.

imo, not enough tests to conclude anything. i cherry-picked depth 27 to examine, & the hyperthreaded version faster. i then looked at d29, and its about even, with the fastest time to depth being w/ HT on. if i split the difference & look at depth 28, similar results: 3min & 5min w/ HToff, 4 min & 2.5 min w/ HT on.

seems possible to me the hyperthreaded runs possibly happened to have 'bad luck' between depths 29 & 30, but of course, the longer & deeper the search, the more accurate the results will be & it's equal possible that the ineffeciences take thaat long to become apparent.

This is EXACTLY what happens, as I mentioned in an earlier post on the same topic.
At first, even I was in favor of leaving HT On. No doubt, the kN/s increases dramatically, with more Depth being attained faster and more Hash usage.
There is even no APPARENT drop in ELO most times.
BUT, I am certain, from my experiences, that SLIGHT errors and inefficiencies, not easily detectable, start creeping in at these very high kN/s speeds and with usage of all the threads, virtual and real.
So, in hindsight, R. Houdart was right after all !