| Date: Mon, 30 Dec 2002 21:49:52 -1000 | From: Warren Togami <[EMAIL PROTECTED]> | Subject: Re: [luau] Hyper technology | | LinuxDan wrote: | > Also, hyperthreading technology started with the 2.0 chip. It increases | > processing speeds up to twice as fast as a dual processor and utilizes the | | Hyperthreading will NEVER have performance twice as fast as a real dual | processor. Most of the time you will have only a few percent | performance increase, and some things actually go SLOWER with | hyperthreading. Only very special cases optimized for hyperthreading | have significant gains. | | > L1 and L2 cache as well so don't buy into any negative comments about | > hyperthreading until you read the facts. Buy the way, the unit described in | > this article was a Dell and came preinstalled with RH7.2. | > | > Dan | | I don't profess to understand completely how L1 and L2 cache works, but | I know enough to say I don't think you understand the true nature of | this situation. Hyperthreading simply doesn't make things twice as | fast. No disrespect intended.
Warren is correct. The hyperthreading technology essentially speeds up context switching for tasks and threads by allowing the processor core to maintain two or more contexts "in situ" and switch rapidly between them rather than having to waste a bunch of time spilling registers to memory on a context switch. It does not provide the equivalent of having a second processor, but is a useful optomization for threading and multitasking. In some cases the overhead of managing this extra state can actually cause a degredation in speed, especially if the program in question does not make effective use of multithreading. The digital video editing PPC vs. P4 article I referenced earlier has a nice comparison of the same tests being done with hyperthreading turned on vs. off (it can be turned off via a BIOS setting). The gains with hyperthreading were very modest, but definitely measurable. --Eric -- Eric Jeschke http://cs.uhh.hawaii.edu/~jeschke
