Would hyperthreading on Nehalem degrade performance?

[wickedmonster]

Since HT fakes a CPU, would it be possible that the operating system saturates the fake CPU first before saturating the real physical CPUs. In that case, performance would be degraded. I wonder how Intel will handle this scenario. I'm sure they have some way to handle it or Microsoft will anyway.

 

[ajfink]

Since HT fakes a CPU, would it be possible that the operating system saturates the fake CPU first before saturating the real physical CPUs. In that case, performance would be degraded. I wonder how Intel will handle this scenario. I'm sure they have some way to handle it or Microsoft will anyway.

I think you've misunderstood what HyperThreading is. Do a forum search or a Google search, there are some really informative articles out there.

 

[dagonoth]

I don't see how it would decrease performance. The pentium 4's with hyperthreading performed at least as well as the ones without hyperthreading. I would expect that given the extra time they have had to improve hyperthreading it might actually give a good boost. At this point who knows if it will be better but it shouldn't be worse.

 

[wickedmonster]

Let's take this scenario: You have 2 threads. With a dual core CPU, HT gives you 4 cores. Would the OS give the 2 threads to the first CPU, thus it would be less efficient if Core1 and Core2 got one thread each. If one core had to do both threads, HT would cause more harm than good. the P4s were only one core so this problem is not an issue.

 

[gOJDO]

Wicked, as ajfink said, do some research about any subject before you open a new thread. Learn what is hyper threading and how it works, and you'll find all the answers you don't know right now.
Also, Nehaleem will implement new type of hyper threading, which is much more advanced and sophisticated. Search and read about mitosis. You can find details on Intel's web page.

 

[Harrisson]

In certain cases hyperthreading may decrease performance as was seen by P4 HT. Since Intel improved HT for Nehalem, such cases will be rare, while performance boost should be much more common, especially with new wave of multithreaded apps.

 

[robbie1687]

Since HT fakes a CPU, would it be possible that the operating system saturates the fake CPU first before saturating the real physical CPUs.

Wickedmonster, you are correct that this problem would occur if the operating system didn't take steps to avoid it.

I wonder how Intel will handle this scenario. I'm sure they have some way to handle it or Microsoft will anyway.

Yes, you're right, Microsoft already builds a scheduling algorithm into Windows to handle this scenario. This has been the case since Windows 2003.

There's a brief discussion of this issue in the following document on Microsoft's website (see section 5.1):

Windows Support for Hyperthreading