How Hyper-Threading Helps ?

Most Intel CPUs use a feature called Hyper-Threading to double the effective number of cores. For supercomputing purposes, Hyper-Threading is of limited use, but it isn't worthless.

To understand why, consider how Hyper-Threading works. Inside the chip, each core contains one central execution unit and two sets of registers and other resources, between which it can switch very rapidly. By dividing its time between two threads, the execution unit can service both in what appears to be real­time. We sometimes describe this arrangement as representing one physical core and one virtual one. although in reality both cores are half-real and half-virtual.

This approach works well for everyday computing but unfortunately, the same can't be said for supercomputing tasks. Here, the object is ordinarily to complete a large number of calculations as quickly as possible: the execution unit is kept constantly busy, and there's nothing to be gained by splitting the workload across two threads rather than one.

That's the theory, anyway, but it's very difficult to create code that makes full use of your computing resources. So even when a thread is intended to fully tax the CPU it's likely that some cycles will be left spare, which o Hyper-Threaded core can use.

We tested this using the multithreaded 3ds Max rendering application, and found that Hyper-Threading on a dual-core processor improved rendering times by 11%. A quod-core processor had a more drastic effect, though, knocking 36% off the time. Hyper-Threading has value, but it's no substitute for real silicon.