Intel® Hyper-Threading Technology (Intel® HT Technology) in charge of not being bugged 14 years after its introduction.
You'd think they'd actually test against the webserver hosting 40% of the market.
>>59281960
>of not being bugged for 14 years..
funny you should say that
because a)nehalems and b)first gen of skylakes
>>59282079
It's still bugged even today since it tanks performance on a static Apache serve, and it's clearly not a single thread benchmark.
>>59282097
It decreased going from 5 to 6. If free cores aren't helping, why would you think throwing more threads at at your busy ones would?
>>59282244
Fuck off, this isn't a fanboy thread.
>>59282300
That's not even 1%, within margin of error, more connections would also require more cores obviously, half the internet isn't served on 5 core chips.
But strictly speaking HT, it would usually lower performance when the workload is already using the cores without the logical threads, but by 1-3%, not by 40%
Real cores anytime anyday
>>59282337
What margin of error? It has 20% more cores and lost performance. The test is obviously completely saturated at 5 threads and chokes beyond that. So assigning more threads to the cores you have obviously isn't going to help if assigning them to free cores doesn't.
Test 2 cores vs 2/4.