Hyperthreading, another mmx? not really

Commentary: The 3GHz PC, Hyperthreading and You

Now that Intel Corporation is poised to release its first 3GHzprocessor, IT shops everywhere are scrambling to understand just howthis watershed moment in CPU development will impact their PC purchasestrategy. Further compounding the issue is the incorporation, for thefirst time, of Hyperthreading technology in a desktop platform. So notonly will this new CPU be the fastest ever (in terms of raw GHz), itwill also feature some revolutionary new scalability-enhancingcapabilities that will redefine how organizations view processorperformance.

Unfortunately, many of the early reviewers of this new CPU will get thestory wrong. That's because the tools they're using the measureprocessor performance - traditional, linear test scripts (Winstone,SysMark) with minimal concurrency - are incapable of generating the kindof multitasking, multi-process workloads that Hyperthreading wasdesigned to address. In fact, they may even show a Hyperthreadingprocessor to be marginally slower than an equivalent CPU (in terms ofclock speed) running without Hyperthreading.

It would be a shame for customers to walk away from these misguidedreviews thinking that Hyperthreading is another "MMX" - i.e. atechnology of little value to mainstream customers. The truth is,Hyperthreading will provide tremendous performance gains across a widearray of usage scenarios, from general business computing (where thenumerous layers of drivers, services and agents all compete for CPUcycles) to knowledge workers and other power user types who frequentlyrun multiple applications. Virtually any compute model that involvesmultitasking has the potential to benefit from Hyperthreading.

In the interest of countering the inevitable "FUD" about Hyperthreading,we've developed a new "How To" guide on multi-process benchmarking. We're also working with leading hardware vendors to helpthem better understand the complexities of performance testing for thereal-world. It's our hope that, by encouraging customers to look beyondthe antiquated linear benchmark numbers, we can open their eyes to thevery real potential of this exciting new technology.

Randall C. Kennedy
Director of Research
[email protected]