Was HyperThreading Killed or did it serve any real value to begin with?

Having a BS & MS in Electrical Engineering, I was fascinated by the idea of HyperThreading.  HyperThreading is a feature popularized by the Intel Pentium4 that allowed for unused portions of the CPU to be used in parallel with other portions.  The net effect was that there appeared to be two CPUs to the operating system.  While performance was not double (it would not double for 2 CPUs), there was about a 1.2-1.3x performance increase for most enviornments over a single CPU with no HyperThreading.  Yes, there are benchmarks showing that are situations where HT provided no benefit.

I noticed the recent announcements of new CPUs that HT will not be in the announced chips.  Obviously, there is some complexity in implementing HT, however, I am curious as to why Intel would remove this feature.  I thought it was a pretty cool propellor head feature.  Any thoughts?

6 Comments

  • Dual core obviates the need I guess.

  • Hyperthreading was designed to take advantage of the long deep pipelines of the Pentium 4's architecture and keep those pipelines full by feeding it effectively two threads.



    The new processors are based on the Pentium M architecture and have much simpler pipelines, ones that will not need to be stuffed with two threads worth of code.



    As far as I can tell...



    [)amien

  • Would it be anything to do with the new chips being multi-core?

  • I went from Pentium 4 with HT on my last computer to an AMD processor and I can definitely tell a difference. The AMD is very fast in single operation (like playing a game), but when you start trying to have IE, vs.net, etc.. open WHILE playing a game, there is some competition for the processor.



    When I had my P4 I could do all of these things without any real negative performance problems.



    I think the main reason the new chips that were announced don't have HT is because they are based on the Mobile processors. The centrino chips didn't have HT either I think.



    Intel is on this new measurement kick called Performance per Watt or something.

  • HT is more of a hack. The main issue with HT is that it doesn't make the second thread unit be able to independently grab the buses. (if I'm not mistaken). A more efficient approach is a dual-core cpu, which effectively has the ability to have each core take the buses for a short period of time.



    Though I still wonder if dual core will solve the root of the problem: the big difference between the amount of data the core(s) theoretically can consume and the amount of data the memory bus is able to provide. With the current architectures with a single chipset it still has a weak spot: not enough bandwidth to feed the cpu's, which means they'll wait / waste cycles.

  • Thats odd. The last thing I heard about HT was that I going to get 4 CPU's



    Oh well.

Comments have been disabled for this content.