Nov 12 2007

Stu­art Corner at iTWire suc­cumbs to our old nemesis, cor­por­ate marketing.

Intel have for years pushed the line that mega­hertz (MHz) equals speed. Apple used to call this the ‘Mega­hertz Myth’. Intel com­pet­it­ors AMD and Cyrix were for many years forced to resort to using a ‘Per­form­ance Rat­ing’ sys­tem in order to com­pete. The fact is that com­put­ing per­form­ance is far more com­plic­ated than raw clock speed.

As the mar­ket­ing droids at Intel gained polit­ical superi­or­ity within the com­pany in the late 1990s, its archi­tec­tures devolved into mar­ketec­tures. The Pen­tium 4’s Net­Burst is a clas­sic example. Unleashed in 2000, in the wake of Intel’s loss to AMD in the race to release the first 1GHz chip, it was widely panned for being slower than similarly-​​clocked Pen­tium 3s in some tests. While less effi­cient clock-​​for-​​clock, it was designed to ramp-​​up in MHz to beat AMD in sheer mar­ket­ing power.

In recent years, Intel have been hit­ting the lim­its of their own fal­lacy. Higher clock fre­quen­cies gen­er­ate more heat and con­sume more power, and start push­ing the phys­ical lim­its of the media. You may have noticed the shift in Intel mar­ket­ing from mega­hertz to com­pos­ite met­rics like ‘per­form­ance per watt’. What they are try­ing to indic­ate is that they are innov­at­ing in all parts of the CPU — not just the clock speed — to deliver greater over­all per­form­ance. Through greater effi­cien­cies, they are able to improve per­form­ance per clock cycle, whilst also address­ing heat and power usage (which is espe­cially import­ant in port­able devices and datacentres).

You should also notice Intel’s sud­den emphasis in recent years on model num­bers (e.g. ‘Core 2 Duo T7200’) rather than just MHz (e.g. ‘Pen­tium 4 3.0 GHz’). They are try­ing to shift the mar­ket away from the myth that they so effect­ively per­petu­ated over a series of dec­ades. My laptop’s Core 2 Duo T7200 (2.0 GHz) is clearly faster than my Pen­tium 4 desktop run­ning at the same clock speed. Reas­ons for this include (but are not lim­ited to) the pres­ence of two cores (each run­ning at 2GHz), faster RAM and a much lar­ger cache.

It is inter­est­ing to note that the design of the cur­rent Core line of CPUs (and its Pen­tium M pre­de­cessor) owes far more to the Pen­tium 3 than to the marketing-​​driven Pen­tium 4.

Now, Stu­art makes the mis­take of pre­sum­ing that Intel’s CPUs are not get­ting any faster since they have not increased in mega­hertz. Instead of berat­ing Intel for finally being hon­est, why can’t we praise them? Address­ing real per­form­ance (not some ‘MHz’ decep­tion), includ­ing the previously-​​ignored factors of power con­sump­tion and heat gen­er­a­tion, is of bene­fit to us all.

If there is any­one to cri­ti­cise, it is the hard­ware vendors. They have suc­cess­fully countered Intel’s mes­sage by con­tinu­ing to mar­ket their sys­tems using MHz as a key selling point. The gen­eral pub­lic (and evid­ently most of the press) are left to believe that com­puters aren’t get­ting any faster. Given the con­veni­ence of a single num­ber as an indic­ator of per­form­ance, who can blame them?

When end-​​user exper­i­ence is taken into account, soft­ware developers fall under the micro­scope. Win­dows Vista is the obvi­ous poster­child — I’ve seen dual-​​core 2GB sys­tems that once flew with GNU/​Linux and (even) Win­dows XP, now crippled to the speed of con­tintental drift after being sub­jec­ted to the Vista torture.

Update: The article’s con­tent seems to have been edited to remove any cri­ti­cism of Intel, but the scep­tical title (‘Intel’s new chips extend Moore’s Law, or do they?’) remains.

Update 2: Now that I have explained that mega­hertz on its own is only of minor con­sequence to CPU per­form­ance (leave alone over­all sys­tem per­form­ance), we can see that it is often not even a con­clus­ive way to com­pare dif­fer­ent CPUs. A Pen­tium 4 can be slower than a sim­il­arly clocked Pen­tium 3. This inab­il­ity to com­pare becomes even more stark when scru­tin­ising com­pletely dif­fer­ent pro­cessor fam­il­ies. Apple had a point when they trum­peted the “Mega­hertz Myth’ back when they were using PPC CPUs. Clock-​​for-​​clock, a PPC CPU of that era was faster than the cor­res­pond­ing (by MHz) Intel chip, often by a con­sid­er­able mar­gin. Apple countered Intel with bench­marks demon­strat­ing the speed of their CPU versus Intel’s. Bench­mark qual­ity aside, their intent was to show that a seem­ingly ‘slower’ PPC chip could out­per­form its Intel com­pet­i­tion. It is a shame that the pro­mo­tion didn’t con­vince more of the gen­eral populace.

LotD: Real Amber vs Pho­toshopped Amber

2 Responses

  1. Chris Samuel Says:

    Another advance that Intel did in Core 2 was to increase the max­imum pos­sible num­ber of float­ing point oper­a­tions per cycle from 2 to 4, which helped them catch up with the AMD Opteron for us HPC folks (though Bar­celona has addressed this for AMD, it’s not just a quad core Opteron). Memory band­width for Intel still sucks though..

  2. speed myths Says:

    […] were for many years forced to resort to using a ???Per­form­ance Rat­ing??? sys­tem in order to com­pete.http://​www​.dhanap​a​lan​.com/​b​l​o​g​/​2007​/​11​/​12​/​m​e​g​a​h​e​r​t​z​-​m​a​r​k​e​t​i​n​g​/​M​ore com­mon speed read­ing myth­sCom­mon speed read­ing myths, solu­tions of com­pre­hen­sion prob­lems in […]

Leave a Comment

Please note: Comment moderation is enabled and may delay your comment. There is no need to resubmit your comment.

Megahertz marketing / 'Til All Are One by Sridhar Dhanapalan is licensed under a Creative Commons Attribution-ShareAlike Australia CC BY-SA AU licence.