Hyperthreading

[handscombmp]

From what i understand hyperthreading allows a core to work on two tasks at the same time. So instead of queing up to do all of task a first and then wait for task b it does bits of task a and b at the same time. so a,b,a,b,a,b.
But the thing is are they really faster. Isn't the total time for the two tasks going to be the same as it has to slow the task down instead of them running at full speed but one waiting for the other to finish.
If its really like this i can't really see how the have a great performance increas espically with software not even making use of quad cores yet.
Also how do the fair in gaming. If i remember correctly in the pentium 4 ht people would turn hyperthreading off as it actually lowered gaming performance.

[matty-hodgson]

they do work like a,b,a,b,a,b.. but if it's just a,a,a,a,a,a,b,b,b,b,b,b then it's more like:

a,recieve result,a,recieve result,a.. etc etc.
but with hyperthreading the b can step in while the a works out what it's wanting to do next, therefore speeding things up.

[Funkstar]

Hyperthreading in all about running two threads or processes at the same tiem using different parts of the chip, so all the hardware there is getting used as efficiently as possible.

For instance if you have two processes with heavy floating point calculations then they won't benefit from Hyperthreading, but if you introduce an integer operation only process then it can be done at the same time as juggling the other two.

*edit* a good primer: http://en.wikipedia.org/wiki/Hyperthreading

[matty-hodgson]

double posted, oops :S

Agent, if you check back, delete this post please?

[Agent]

The super simplistic way of putting it: It's very hard to get 100% core utilization under normal conditions, there will always be some of the processing power being unused (might not be enough work, errors have happened within the CPU...whatever).
By separating the core into 2 separate addressable areas, you are minimising the risk of this. You get the biggest gains in large threaded / multi-core applications.

In some, very rare conditions it can hurt performance. But this is not the norm.
Back in the P4 days, you could get a nice little increase in render speed with HT