<![CDATA[
<body bgcolor="#DDE4EC">
<h3><a href="../">WhoPhD</a> | <a href="./">Frivolous Diversions</a> | Megahertz Overview <font size=1><!--#exec cgi="/cgi-bin/counter-nl" --></font></h3>
<h1 align=center><small>(Almost!)</small><br>Totally Useless Megahertz Overview</h1>
<center>
<img src=mhz2004-1-noise.png><br><i>Updated March 2004 — with special <a href="http://www.angryflower.com/timelo.gif">Time-Looker-Forward Capability™</a><br><b>Now with the Added Goodness of Itanium!</b></i>
</center>
<table border=0>
<tr><td align=center><big><i>What This Shows</i><br></big></td></tr>
<tr><td>
<ul>
<li>Moore's Law, strong and steady
<li>Some historical figures for fuzzy memories, rough trends for manufacturers and industry
<li>What that latest speed jump is <b><i>really</i></b> worth (compared to previous jumps)
<li>What kind of a boost a dual-CPU system might really provide (use the grid)
<li>Who's got bragging rights (in <b><i>real</i></b> megahertz, not just in a name!)
<li>How bad it really is to fall behind, in terms of development time or raw MHz
</ul>
</td></tr>
<tr><td align=center><big><i>What This Does <b>Not</b> Show</i><br></big></td></tr>
<tr><td>
<ul>
<li>What's cheap and good value — only the latest & greatest are shown here; none of the low-power models or overall product range
<li>The effect of any non-CPU performance, eg motherboard, RAM, disks, other subsystems, multiple CPUs, system manufacturer
<li>The effect of any non-CPU-clock performance, eg MMX, SSE, AltiVec, 64-bit, VLSI, CISC, RISC, Hyperthreading™, dual cores
</ul>
<center><b>Some of the above factors have been known as part of the <a target=_new href="http://www.google.com/search?q=megahertz+myth&ie=UTF-8&oe=UTF-8">"Megahertz Myth"</a></b></center>
<br>
</td></tr>
<tr><td align=center><big><br><br><i>Observations</i><br></big></td></tr>
<tr><td><br>
<b>March 2004</b>
<div align=left><i>
Holy Moore's Law, Batman! Where's it gone!??
</i></div>
<ul>
<li>I go away for a bit, and what happens!? Everybody's having a siesta! 2003 was a really snoozy year, after the mad scramble to 3-and-a-bit gigahertz a year ago. Let's ignore the 2004-5 predictions for a minute (though I do include them with the best of intentions).<br><br>
<li>AMD: Still bloody snoozing. Yes their x86-64 is a real coup, but they've had no increase in core clock speed since 2.2GHz. But people are speaking with renewed confidence about AMD this year, with the Q2 2004 seeing some 200MHz speed boosts and the same again 6 months later. Meantime, in marketing fairyland, the nomenclatures have reached 3200, 3400 and soon 4000. Just in time for ...<br><br>
<li>Intel: Taking a breather over 2003, after that huge 30% increase in 2002. You'll see the curve is a lot flatter in the last year. There's only very token increases going on since the "milestone" 3GHz was reached, and the marketing machine may have been turned down too, so I was surprised to find when I did my research I could only come up with 3.4GHz (and by George, you'll pay for it). Most people will be buying, and only finding, 3.2GHz (see more of this magic number below). There has been a plan to get out to 4GHz by 2005, but despite this, isn't it a bit of a worry to see Moore's Law take a flattening trend for such a long 2-year-period? (I shouldn't really be delving into corporate roadmaps though, but I couldn't help myself, sorry).<br><br>
<li>PowerPC: Against many rumours, no speed update was made to the G5 in this semester. This was also against a lot of track record at Apple. It's possibly a combination of residual G5 newness, stock levels, and rumoured glitches. Anyway that doesn't change the big fat (and out-of-character) prediction they made (well, IBM made) last year that they would reach 3GHz later this year. So I have confidently plotted it, and ...<br><br>
<li>AMD has been overtaken by PowerPC! (Real Soon Now). Yes, and what does this mean? A conservative assessment is that nothing changes in relative architectural performance, but even that means the Megahertz Myth is now flowing in a different direction. (Oops, sorry, forgot it's 3GHz G5 vs Athlon "3700"!)<br><br>
<li>Argh, my track record for predictions has stuffed up. Pentium has turned off the afterburner and won't reach 5GHz until ... a while from now. Meantime, my PowerPC prediction is screwed but I was smart enough to say that if it changed from G4 to G5 (which it did), all bets were off. It seems the MHz-slip that occurred from G2 to G3 has now been reversed, and IBM has provided a nice marketing boost for free (hey it was the same folks who made the original "copper" G3! And Steve cleaned his own mess now. Finally). I was spot-on for the G5 case design, bwahaha. I want one.<br><br>
<li>"Moreover," (just reading the old logs below), yes it seems there's an end to Intel's IA32. Well, its meteoric speed rise, at least. Just after I asked, too :-) Geez I can't even remember what "crossover" project I was referring to, getting killed off, because a year later there's a new announcement about putting 64-bitness into IA32. That's x86-64, not IA64. Which reminds me ...<br><br>
<li>Almost forgot: We'd like to welcome Itanium to the chart. We'll see how this goes. I'm very tempted to add an Alpha and even a Sparc line, because they're both about to expire (and that means less work!) It was with "Itanium 2" that Intel gave it a very respectable speed boost. There is, sadly, only one small speed boost on the way. If it wasn't for everyone else treading water, Itanium would really stand out for the last 6 months.<br><br>
<li>Oh yes, I'm sure you understand that the big speed jumps in real life are represented by nice curves on the graph. They look better (I'm sure you understand that too), and after all, this can roughly represent your likelihood of getting your hands on the high-end models, re their distribution, your funds and equipment lifecycle.<br><br>
<li>One little note, as we seem to be at an alignment of the stars here: For those who are interested, there's an <a href="http://www.anandtech.com/cpu/showdoc.html?i=1965">article at AnandTech that details three different Intel 3.2GHz CPUs.</a> This is going to be really handy, to compare the variations of CPU we are going to see where the differences are much less in the clock speed and much more in price, power usage, features and so on. This is all amidst the scary background of Intel changing its naming structure to -- shock, horror -- remove references to megahertz.<br><br>
<li>Predictions? PowerPC 3GHz, easy. Pentium 4GHz, probably to retain at least a 1GHz bragging buffer (remember, that means less than 1GHz used to!) Athlon "4000", which really means 2.6GHz according to published plans. But only if they can get it out in time -- in fact I doubt you will be able to buy an Athlon 4000 by the time you can get a PowerPC 3GHz, so I'll just stick with "3800" == 2.4GHz. I've put as much in the graph already. And Itanium, pffft, nothing much really.<br><br>
</ul>
<br>
<b>Late 2003</b> (not really, but this is by way of apology)
<div align=left><i>
I updated the graph but didn't bother uploading it. I've noticed a lot more hits on this site recently, so I'm sorry if you've been waiting for an update. If it's any consolation I've decided to do more forward-future-looking stuff, because, well, I'm impatient for the next site update too ;-) Errors will of course be fixed when we reach each semester, and figures are of course error-free only up to the current semester.<br><br><br>
</i></div>
<b>17 January 2003</b>
<div align=left><i>
Name change, to reflect how I really feel about this webpage :-)<br>
Frames Are Evil™ but I think I may have discovered a use for them.
</i></div>
<ul>
<li>I've done something different today, and placed Pentium and Athlon into a hypothetical shootout, so we can see where things are headed in anticipation of another "flash" announcement from Apple outside the scheduled keynote opportunities, to speed bump the G4 or quite possibly even roll out the G5 (PowerPC 970?). I'll increase the Athlon and Pentium speeds if they make another revision before Apple releases a new PowerMac. As you can see, the G4 is not "stuck" in the low gigahertz region — whatever relative strengths and weaknesses its clockrate had years ago, it is still tracking.<br><br>
<li>In fact, I've done this before on the Excel spreadsheet, and it's always been very successful at indicating the next speed bump (with the exception of, it seems only, the G3 introduction). So I'll stick my neck out and say 1.5GHz for the next desktop — if it is just a speedbump — but if it isn't a G4, I'm too chicken to speculate, and will just leave it at a prediction of seeing more metallic materials in the new desktop case!<br><br>
<li>So, has AMD really dropped the ball? I guess what I've been hearing from PC / Linux users may have something to it. What's interesting, and difficult to see without a logarithmic scale like on this chart, is how the numbers mean such different things at different scales. The Athlon XP 2800+ is 116MHz more than the Athlon XP 2600+, but it's really a bit of a shocker (unless the front-side bus clockspeed increase makes up for it), especially when you discover the new XP 2700+ is only 33MHz faster than the XP 2600+. That's peanuts in the gigahertz arena.<br><br>
<li>Moreover, where's AMD's controversial naming scheme going now? A "2800"-equivalent MHz chip is all well and good, but what does this mean when Intel is at (a peek over) 3GHz? Oh yeah, that's what the "+" sign is on the name for.<br><br>
<li>Moreover moreover, is there no end to Intel's IA32? It's just picked up a bit of speed again. And we're <i>still</i> waiting to see how IA64 (Itanium and friends) will catch up and join the Intel party (their crossover CPU project just got killed the other day). I'm quite willing to include Itanium as a separate colour on this graph. Does anyone know where I can go online to buy an Itanium CPU? I'd be very interested to hear from anyone who can supply a history of Itanium CPU speeds dating back to 2000.<br><br>
<li>Scary throwaway prediction 2003: Pentium @ 5GHz this time next year.<br><br>
</ul>
<b>28 August 2002</b>
<div align=left><i>
One last roll of the G4 dice, it seems, and time for a third mini-redesign of the case. Apple will probably go with a lot more metal (as opposed to "mercury"-coloured plastic) with the next major model (the G5, for instance); but for now an extra piece of metal will do. More interesting is all the air outlets on this one, and people have been nicknaming it "G4 Windtunnel".
</i><br></div>
<ul>
<li>The 0.25GHz (250MHz) increase is probably more than it sounds — the trendline for PowerPC, far from slouching, is even picking up a tiny bit. Well done to Motorola it seems. And to me — another prediction, from 29 January, comes true! Why am I patting myself on the back about this? Well, think about what the rumour sites have been going on about. 1.5GHz ... 2GHz ... my point, on this page, is to prove it's all a psychological gap, or at least there's been no new gap created in the last three years or more.<br><br>
<li>The Pentium 2.8GHz was announced already, I am very aware, but I cannot find it on sale anywhere today. You can't even pre-order it. And yes, I've checked that you can get a 1.25GHz G4 delivered today.<br><br>
<li>Meantime, I need to find an easier online reference for buying Athlons, but <a href="http://www17.tomshardware.com/cpu/20020826/p4_2800-02.html#amd_responds_athlon_xp_2800_coming_up_in_october">Tom's Hardware Guide says</a> AMD is coming up with the Athlon XP 2800+ in October (actual clock speed == 2.25GHz), so I am placing their XP 2600+ (2.133GHz) speed on here now.<br><br>
</ul>
<b>29 January 2002</b>
<div align=left><i>
With nothing more than a press release timed to maximise anticlimax, Apple and Motorola have released dual-1GHz systems. A dual-CPU configuration hasn't been done on the high-end since the G4 was stuck on 500MHz and became 2-for-1 to sweeten the bad taste, so the 3-week delay could easily be explained by typical low yields of the top speed. Or, it could just be shyness over a sore point about the gigahertz barrier.</i><br></div>
<ul>
<li>The trendline is now remarkably straight for the last 12 months. As stated below, a 1GHz PowerPC was neither unreasonable nor unexpected, and keeps the status quo in the pure megahertz stakes to date (dual-CPUs aside!)<br><br>
<li>So that vindicates the prediction on 02 January! This page has a 100% track record so far :-)<br><br>
<li>There will be no further predictions other than to say the PowerPC G4 should be at 1200MHz in the second half of 2002. The G5 is an unknown quantity, and if the trendline stays this pristine, the Pentium should hit 2.7GHz at the same time; the Athlon will actually have 2GHz, but its name will probably be the "Athlon XP 2400+", increasingly below Intel's clockrate. These forecasts are worth as much as you paid for them, and I'll let another website debate the upcoming dilemma for AMD as its new naming scheme causes the same marketing difficulties it was designed to avoid.<br><br>
<li>The G4 confirms it has made up for losses it incurred when it was stalled at 500MHz.<br><br>
<li>The original 500MHz G4 exceeded three gigaflops, while the dual-1GHz G4 has 15 gigaflops. Apple claims its dual-1GHz system will compress MPEG-2 video at a rate 300% faster than a 2GHz Pentium IV. This would translate to the performance of a theoretical 3GHz Pentium IV on a per-chip basis. There are other claims of a 70% speed advantage for Photoshop tasks, and so on, and there are bound to be some tasks where the Pentium has an advantage. Apple's tests are likely to be based on converting DV streams to MPEG-2 streams, but I am generally concerned with video compression speeds overall because this represents the last raw-speed barrier faced by home computer users. If anyone has any info on PC speeds of converting DV to MPEG-2 or uncompressed video to MPEG-2 then please email me at the account for this website.<br><br>
</ul>
<b>23 January 2002</b>
<div align=left><i>
Still no update. Maybe Steve Jobs is stalling to spite MacUser UK magazine. :-)<br>
</i><br><br></div>
<b>07 January 2002</b>
<div align=left><i>
Oh well, surprise surprise, no updates at all! That breaks almost everyone's predictions, leaving most people twice-bitten by the hype at the last two MacWorlds. (The only theory of mine left standing is that Steve Jobs has decided to change tack in his fight on the rumour sites and discredit them by feeding them more rope).<br></i></div>
<ul>
<li>Apple has left very little gap between the G4 iMacs and the G4 PowerMacs. There is consensus that some kind of update will still happen this quarter. Speculative dates include MacWorld Tokyo, Seybold San Francisco, and a low-key Apple Event in late January.<br><br>
</ul>
<b>02 January 2002</b>
<div align=left><i>
The rumour mills are again in overdrive on MacWorld Eve, almost as rampant as the previous MacWorld where a lot of the web-savvy user base were left feeling unappeased. Aside from the grandest hoax ever subjected on the Mac community, speculation of flat-panel iMacs and 1.6GHz G5 PowerMacs remains.<br></i></div>
<ul>
<li>Interestingly, the logarithmic scale very much assuages the "ever-widening" gap that the Pentium IV is leading with. In fact, it isn't widening at all. The logarithmic scale is the only way to accurately measure and show Moore's Law, as well as how "fast" a computer feels — an upgrade from 1000MHz to 1800MHz is nowhere near as dramatic as the jump from 200MHz to 1000Mhz.<br><br>
<li>In fact, Moore's Law is holding up as well as ever. For most architectures, for most periods, the trendline is very straight.<br><br>
<li>Where the trend is broken, for better or for worse, a legacy is created that lasts for years afterward. The trailing architecture is always trying to catch up, but as new developments are made to try to leapfrog into the lead, the competition also gains more speed.<br><br>
<li>For MacWorld San Francisco 2002, PowerPC is tracking towards pipping the gigahertz barrier for its high-end chip, while still having sub-1GHz chips on the low-end. Any more than this, and the Mac community should not only be grateful, but surprised at a break in the trend.<br><br>
<li>Surprisingly, especially following the reaction at MacWorld New York mid-2001, the PowerPC seems to have regained whatever pace it lost as a result of the inability to yield 500MHz chips in 1999. If this were a baton relay race, the G4 certainly wouldn't deserve the poor megahertz legacy it seems it will now earn.<br><br>
<li>In fact, the gap between the PowerPC and the Pentium seems purely down to two other factors: The G3 and the P4. The G3 was a lean, efficient processor that Apple "wisely" chose to use in its professional as well as consumer lines after they observed how much grunt it surprisingly cranked out of each megahertz. Despite the drop in megahertz, a performance parallel with the Pentium (or arguably, performance lead) was maintained; and as long as the vertical distance in this graph is maintained, that parallel will also remain.<br><br>
<li>The antithesis to the G3 was the alternative that Apple rejected in 1997, the "Exponential" chip that ran at much higher clockrates than G2 chips at the time, and promised to break 500MHz long before the PowerPC or Pentium. However, the chip consumed too much power and expelled high levels of heat, and Apple observed that its actual performance was close to or below the G3 chips it was evaluating. Apple's decision led to Exponential Technology's downfall, and its BiCMOS technology was auctioned to graphics chip maker S3. This company merged with Via Technologies, who manufacturers chipsets for motherboards of Intel-based PCs. Via and Intel have had a close but competitive marriage of convenience, and in resolution to one of their past lovers' tiffs they cross-licenced some technologies. The BiCMOS technology was one of the technologies that Intel gained access to, during development of the Pentium IV. The management at Intel must have laughed out loud, at being handed such pure marketing technology that had once been rejected by a rival architecture.<br><br>
<li>So while the move to the G3 may have been wise for all the right reasons, like engineering and power consumption, it is questionable right up to today whether it was a wise marketing manoevure. We would not be sweating over a jump to the gigahertz barrier in 2002, rather than 2000, if this were not the case. Steve Jobs was present and active at Apple during the days of Exponential, and years later found himself standing up at a keynote address having to explain something called "The Megahertz Myth" to Mac users.<br><br>
<li>Meantime at Intel, the engineers are living on borrowed time. The Pentium architecture was cranked for MHz at the beginning of 2000, when AMD and Intel were fighting to beat the gigahertz barrier. (AMD was displaying refrigerated 1GHz units at CeBIT in February 2000). The business environment has changed since then, plus Intel's upcoming VLIW 64-bit architecture, the Merced/Itanium range of chips, does not carry the same marketable megahertz benefits that the Pentium does. The Itanium is not represented on this chart, and in fact lies at a lower MHz than even the PowerPC in late 2001.<br><br>
<li>The AMD, especially in dual-CPU configurations, has remained the darling choice among PC / Linux users. However some Mac users are worried that AMD speeds never get mentioned when Apple makes comparisons between PowerPC and Pentium. If you're worried about the ability to configure any G4 or G5 in dual-CPU as easily as an AMD or Intel, consider the logarithmic graph and plot where the speed gain would lead if you assume a 100% improvement. The gain is a fixed vertical distance every time. Then remember that vertical distance equates to apparent speed!<br><br>
<li>Also remember the Megahertz Myth, if you think it applies to you, when working out dual-CPU advantages.
</ul>
<br>
</td></tr>
<tr><td align=center><br><b><i>Important Stupid Little Things</i></b><br></td></tr>
<tr><td>
<b><small><small>
<ul>
<li>Absolutely no concession is made to the relative performance of architectures / sub-architectures, SPECints, gigaflops, Megahertz Myth, etc (all symptoms of the same thing) hence the PowerPC G3's release represented a drop of 33MHz (12%) but had a performance increase of 10% over the previous G2 (604e) architecture.
<li>Only the top-speed chips of the time are represented. All architectural brands released by each manufacturer have been consolidated. Dual CPUs are ignored.
<li>This graph is biased towards PowerPC, only in that the data is quantized where the PowerPC is upgraded (twice a year). If Apple don't release a CPU update in a semester, then the bias is actually the other way. ... Basically, I'm lazy, so I update this page whenever Apple releases something, and check out Dell for Pentium, HP for Itanium / Alpha / PA-RISC, a few other big sites, and my local yum-cha dealer. (Lately I've been using Froogle to look up the latest CPUs reviewed on AnandTech and Tom's Hardware, to see if it really is shipping). ... Hence the 2GHz Pentium isn't represented in Q3 2001, whereas the 2.2GHz Pentium and "XP 2000+" Athlon were in systems by February 2002, when the dual-1GHz PowerPC was in Apple's systems. ... In the end it's really okay because the line curve really sorts things out.
<li>With AMD's MHz, I've gone to the trouble of finding out the true (raw) numbers! :-) ... In case you didn't know, the numbers used in their Athlons' names do not reflect their true number of cycles per second. ... What the names <i>do</i> "reflect", is closer to "true", anyway (if you're comparing to Intel chips, which you probably are).
<li>All lines are rounded by MS Excel.
<ul>
</small></small></b>
</td></tr>
<tr><td align=center><big><br><br><i>What Is This For?</i><br></big></td></tr>
<tr><td>
<ul>
<li>Seeing if one architecture is lagging behind or excelling another, more than it did before, or less than it did before
<li>Put another way, answering the question, "Are they catching up? Are they falling behind? Are they streaking ahead?"
<li>Predicting how many megahertz an architecture will have later in the year, or when it will pass the next milestone
<li>Showing how the increase in CPU speeds is <b>(a)</b> straight, not skyrocketing <b>(b)</b> very smooth, as Moore's Law said
<li>To push my conspiracy theory on BiCMOS technology and Exponential Technology Inc. :-)
</ul>
</td></tr>
<tr><td align=center><big><br><i>What Is This <b>Not</b> For?</i><br></big></td></tr>
<tr><td>
<ul>
<li>Seeing if one architecture family is faster, better, whatever, than another, at <i>anything</i>
<li>Comparing typical or average-speed ("<i>normal</i>") systems from each family, let alone factoring in best-value pricing
<li>Accounting for whose systems have the latest übergeek technology within their CPUs, for multiple CPUs, for RAM, or the system bus
</ul>
<center><b>We're basically comparing pure numbers with numbers, to show you how predictable and meaningless they are!!</b></center>
</td></tr>
<tr><td>
<br><br>
<i>Feedback and reactions gladly accepted by email to the unix account of this user.</i>
<br>
<i><small>In other words, "whophd" at "progsoc.uts.edu.au". I think the traditional unix conventions have been lost by most people these days. Never mind — I've written this in small text so the spam search engines won't find it.</small></i>
</td></tr>
</table></center>
</body>
]]>