Hyperthreading Налаштування

[hunter]

Замерял ли кто, дает ли запуск двух экземпляров ФАХа на процессоре с НТ заметное увелечине ППД? Насколько я понимаю, в таких процах вычислительная часть все равно одна.

[ReMMeR]

Your multiplication looks fine to me, but it looks there's a misunderstanding with how FAH works.

After GEN0, CLONE0-4 is completed, we would not assign GEN0 CLONE5-9. Instead, we'd assign GEN1 CLONE0-4 with those 5 non-HT CPUs.

Thus, to get to GEN200, the non-HT machines would take (34 hours/WU) x (200 WUs) / (24hours/day) = 283 days whereas the HT machines would take (52 hours/WU) x (200 WUs) / (24hours/day) = 434 days: a difference of 150 days ~ 5 months.

Of course, the HT machines would have 2x as many simulations. However, if the minimum number of generations is high (as it is in many of our newer, more complex simulations), having 2x as many useless simulations is still useless. This is what I explained in previous posts (perhaps it might make sense to go over those arguments again).

I hope this clears this up. Thanks for your time and suggestions. Also, please keep in mind that my team has run these types of numbers and we've considered this carefully. If you don't see why we're doing something, please consider the possibility that you might not be familiar with how FAH works and that the answers lie there.

http://forum.folding-community.org/viewtop...p?p=82366#82366

Одним словом :
Виджей Панда настоятельно не рекомнедует включать 2 клиента на компе с гипертрейдингом, так как модель обсчитывается последовательно, на НТ замедляется просчёт отдельно взятого ГЕН-а, и при годовом масштабе потеря будет 5 месяцев.
Получается что набирая немного больше очков вы можете довольно сильно замедлить проект.