What is hy­per­thread­ing?

The basic idea behind hy­per­thread­ing is to be able to process several threads sim­ul­tan­eously with one mi­cro­pro­cessor. While pre­vi­ously only one thread could be managed per processor, Intel’s hy­per­thread­ing tech­no­logy now allows a processor to be divided into two logical, virtual cores that process hyper threads sim­ul­tan­eously. You can find out how hy­per­thread­ing can improve CPU per­form­ance here.

What may sound like science fiction is nothing more than the old wisdom that four hands can do more than two: hy­per­thread­ing. Intel’s hy­per­thread­ing method ori­gin­ates from the server sector, where several physical pro­cessors are usually used sim­ul­tan­eously to increase computer per­form­ance. With hy­per­thread­ing, on the other hand, there is only one physical processor, but it behaves like two pro­cessors. This is made possible by the processor dividing itself into two virtual cores – also known as kernel – which process threads in parallel. This allows multiple in­struc­tion queues to be processed sim­ul­tan­eously as a hyper thread, processes to be split between the virtual cores, and CPU util­isa­tion to be improved.

Hy­per­thread­ing tech­no­logy (HT) is owed to the processor and CPU man­u­fac­turer Intel. Private computer users have been enjoying Intel’s hy­per­thread­ing since November 14^th 2002. Intel brought its hy­per­thread­ing tech­no­logy onto the market with the Pentium-4 including the Northwood-B core. The successor models, Pentium D and Core-2-Duo, Intel withdrew hy­per­thread­ing again and used dual-core main pro­cessors instead.

If you’re hearing about hy­per­thread­ing for the first time, then you may think that every processor should have this tech­no­logy. For­tu­nately, hy­per­thread­ing is active by default in CPU cores, provided they support HT. Nev­er­the­less, hy­per­thread­ing can be turned on and off in the BIOS under ‘Hyper threading tech­no­logy’ via ‘Enable’ and ‘Disable’. In pro­cessors that don’t support hy­per­thread­ing, a physical core only processes several logical processes se­quen­tially, i.e. one after the other. You can check whether hy­per­thread­ing is active or supported in the Device Manager. For example, if you have a computer with two CPU cores, but you see four pro­cessors under ‘Pro­cessors’, then hy­per­thread­ing is active.

Probably the biggest advantage of hy­per­thread­ing is that im­ple­ment­ing hy­per­thread­ing tech­no­logy in mi­cro­pro­cessors is less expensive than using two physical mi­cro­pro­cessors. However, since most computers today already have multi-core pro­cessors, this advantage can only be exploited if CPUs with hy­per­thread­ing have the same number of physical pro­cessors as machines that don’t support hy­per­thread­ing.

The actual advantage of hy­per­thread­ing is the efficient util­isa­tion of resources. Two virtual cores in one physical core does not mean that all tasks run at double speed. However, the computing load for several processes can not only be dis­trib­uted se­quen­tially, but also sim­ul­tan­eously between the virtual cores. This avoids un­ne­ces­sary idle times so that processes can be executed without gaps. Threads don’t have to wait until a com­pu­ta­tion­ally intensive thread has been processed, but simply run over the second core.

The advantage of multi-core pro­cessors is that a physical core does not have to share its resources. CPUs with ten physical cores, for example, would basically have an advantage over five physical cores with hy­per­thread­ing, because each core can use all available resources to handle processes. Virtual cores, in turn, use the same computing power of the physical cores to manage parallel register sets. Nowadays, it makes sense to work with CPUs with at least two or more physical cores.

Mul­ti­th­read­ing and hy­per­thread­ing are very similar at first glance: Both are tasked with ef­fi­ciently dis­trib­ut­ing and pro­cessing threads in CPU cores. However, hy­per­thread­ing is a sub­cat­egory of mul­ti­th­read­ing and is also called sim­ul­tan­eous mul­ti­th­read­ing (SMT). To un­der­stand mul­ti­th­read­ing, it’s important to know that threads are in­struc­tion queues that are processed dif­fer­ently depending on the hardware and software.

Mul­ti­th­read­ing means that multiple in­struc­tion queues are managed sim­ul­tan­eously. However, while mul­ti­th­read­ing methods such as switch-on event mul­ti­th­read­ing and time-slice mul­ti­th­read­ing are referred to as pseudo-sim­ul­tan­eity, since threads do not run sim­ul­tan­eously, sim­ul­tan­eous mul­ti­th­read­ing, i.e. hy­per­thread­ing, stands for true sim­ul­tan­eity in pro­cessing. Fur­ther­more, hy­per­thread­ing is a hardware-supported SMT tech­no­logy, while mul­ti­th­read­ing can also be supported ex­clus­ively by programs and software.

When it comes to hy­per­thread­ing, the question often arises whether the function really has ad­vant­ages compared to multi-core pro­cessors without hy­per­thread­ing. The answer is: it depends. For example, it’s crucial what hardware is available and what it’s used for. The fact that hy­per­thread­ing can optimise thread pro­cessing through separate pipelines and parallel register sets is un­deni­able. For example, the Cray MTA su­per­com­puter is able to manage an im­press­ive 128 threads with only one CPU kernel.

Gaming, in par­tic­u­lar, is often noted in con­nec­tion with SMT or hy­per­thread­ing. Gamers in par­tic­u­lar require a lot of computing power when games, music, and maybe even a streamed Twitch recording are running in parallel. Whether hy­per­thread­ing actually improves gaming per­form­ance is debatable. Some gamers say that hy­per­thread­ing actually slows down CPU per­form­ance, while others say that hy­per­thread­ing can be useful when games support four or more cores.