Intel first used Hyper-Threading when it introduced the Pentium 4 “Northwood” processor at 3.06 GHz and the Xeon MP “Foster” series in 2002. The proprietary technology’s main purpose is to improve processor utilization through increased parallelization. With the latest Core i7-980X and its six physical cores, Hyper-Threading yields 12 logical cores on desktop PCs.
This raises the question: how much of the software that you run truly takes advantage of eight or more threads? Is Hyper-Threading good or bad for power efficiency? And wouldn’t it make more sense to stay with six physical cores, rather than risking performance hits caused by less-heavily-threaded applications unnecessarily distributing workloads to logical units?
Intel’s Gulftown implements Hyper-Threading to provide 12 virtual processing cores. Serious performance benefits can only be found in a few, specific applications.
Hyper-Threading was introduced almost out of necessity. Because the Pentium 4 processor employed a rather long instruction pipeline, it was imperative to ramp up operating clocks as quickly as possible and keep the pipeline busy. Therefore, Intel duplicated the units that store the architectural state, allowing a Hyper-Threaded core to appear as two logical processors to the operating system. The scheduler could dispatch two threads or processes simultaneously, and if Intel’s branch prediction worked well, it would ensure that instructions got loaded and executed efficiently.
The benefits for the Pentium 4 were mainly increased system responsiveness on single-core systems and small performance gains on applications. However, this applied to the desktop space. In servers, where parallel processing is key, Hyper-Threading showed more impact. Naturally, this was a reflection on the software industry at the time. Applications written for desktop users weren’t threaded yet, since the hardware enabling this usage wasn’t around. Initially, Hyper-Threading got a bad rap because it failed to improve performance in those titles that ran in a single thread.
… and the Return
With the arrival of the Core 2 processor, Hyper-Threading disappeared. But Intel decided to resurrect it with the Nehalem micro-architecture, which is the basis for all Core i7, i5, and i3 CPUs available today—including the just-released six-core Core i7-980X.
The situation is much different today than when Hyper-Threading made its first rounds. For starters, software developers are much more in tune with the hardware ecosystem, so it’s uncommon to find a popular title that can benefit from parallelism and isn’t threaded. Beyond that, AMD currently can’t apply pressure to Intel in the performance segment, and Hyper-Threading has turned into a value-added feature and series differentiator, rather than a must-have innovation. With six physical cores, does Hyper-Threading really make sense?
We decided to look at the quad-core Core i7-975 Extreme Edition (Bloomfield) alongside the new six-core Core i7-980X (Gulftown) and compare performance, as well as power efficiency, using our updated platform benchmark suite.