HyperThreading

If you are wondering what HyperThreading is actually all about, the simple answer is that it creates two ‘virtual’ CPUs out of one.

If we think about a multi-processor machine, it will have two physical CPU's that can handle multiple tasks or multiple threads of a single application, spreading the load between the two processors. Many of the benefits from multi-processing can be squeezed into one CPU though.

A processor has a fixed number of instructions it can execute per pipeline, and under ideal circumstances it would always be calculating something in each and every part of the pipeline, every clock cycle. The reality is very far from that ideal case, though, and much of the available processing time goes unused because the processor is often waiting for other things, such as an unforeseen branch in the program requiring the CPU to fetch different instructions. If there was a way of utilising the wasted processing time on something else then, in theory, more could be done on a single CPU in the same space of time - this is what HyperThreading attempts to achieve.

HyperThreading

The first of the two diagrams above shows the difference between normal CPUs and those that are HyperThreading enabled. Normally a CPU will have an Architectural State engine that controls what it being executed on the pipeline - with a multiprocessor system each CPU has its own Architectural State engine which means that each one can be working on different applications, or different thread of the same application. With a HyperThreaded CPU there are two Architectural State engines controlling what is being executed on the CPU, by which the state of two applications or threads are controlled. The theory is that the two tasks are controlled effectively enough so that the 'gaps' in processing that normally occur in a single threaded CPU are 'filled' with processes from the other thread with as little contention between the two as possible.

Windows XP Device Manager - Two CPU's or One?

The above image shows that despite there actually only being one physical CPU the Windows XP OS (SP1) recognises it as being two separate processors. The Windows NT Kernel has been multiprocessor aware from its Workstation and Server background and is capable of some task and thread allocation. Despite there actually only being a single physical CPU present it still needs to be detected as two such that it will be 'aware' that it is able to treat this as a multi-processor machine and therefore share tasks between the Architectural State engines.

HyperThreading is not actually new, since this has been in server CPUs for a while, but this is the first time that its been brought to the desktop space -- possibly partly because it wasn't until Windows XP that the mass desktop/consumer OS was multi-processor aware. However, this can operate in multi-processor environments as well, such that if you are utilising two HyperThreaded CPUs you'll have two physical CPUs, but the system will recognise four virtual CPUs.