©2000 Joe Carson

"And this too shall pass away."

(Ancient proverb)

Cassandra's Curse

"Cassandra had been gifted with the power of prophecy, but Apollo, whose advances she had refused, brought it to pass that no one believed her predictions, although they were invariably correct."

Bullfinch's Mythology

In Twilight of Empire, parts one and two I explored growing weaknesses in the two legs of the Wintel hegemony that can eventually end what I characterized as a "Water Empire", one based on total control of critical commodities. Needless to say this generated a certain amount of "discussion" on some forum sites from some individuals dedicated to the care and maintenance of the Wintel Hegemony. Of course, this discussion from the Wintel partisans was less than complimentary towards the articles as expected, but then again, it does indicate that I hit my targets.

I should mention that I did recognize the tell-tale footprints of some who have responded to me. One is an old Unix curmudgeon who once wanted to know how to make Mac OS X run X-Windows as the interface. Another is an "expert" who over time at a forum has revealed himself by bits and pieces to be a high school kid who does not own his own computer, has had a part time job assembling the pieces for a local computer store, and gets his misinformation from a certain Latin-named PC-centric technical source that often promulgates serious misinformation, and from well-known dedicated anti-Apple site. As an example of how he parrots the Party Line a site declared that Mac OS X wasn't Unix because Apple was not paying license fees to AT&T who owns the rights to the original Unix. I won't go into the pure absurdity of this silly claim since it has already been laughed down at Slash-Dot. This self-appointed "expert" then went onto a forum the next day and parroted this claim literally word for word. After that, I can hardly take him seriously.

One of Ars Technica's bits of misinformation is the claim they have made that somehow the newer AMD and Intel processors were really RISC processors. I only need to point you to two articles by David K. Every in his column, Design Matters. One,"Is the x86 a RISC?" soundly debunks the claim of x86 being RISC. In the opening paragraphs of the article he expresses his own informed opinion of Ars Technica that mirrors my own. He also explores what really is RISC in his article "What is RISC?". Oh yes... what does David K. Every do to pay his mortgage? He is a software engineer, not some kid with wishful fantasies posting his delusions at a PC-centric site.

The biggest "disagreements" were in regard to my pointing out the weaknesses in the Wintel hardware. The Wintel theologians disputed my assertions, using technobabble to prove they were right and I was wrong, at least to their own satisfaction. In reality, tech specs simply don't tell the whole story and often glowing tech sheet numbers will tell a bald-faced lie straight to your face. Case in point: the recent introduction of Intel's 1.5 GHz Pentium 4.

For months the Wintel types had been praising the marvels of the new high clock speed wonder to come from Intel, and the technical specs did look marvelous... on paper. Imagine, twenty pipeline stages in place of the Pentium III's ten and given a new technobabble name, "NetBurst Micro-Architecture". Imagine, a new SSE2 register with 144 new instructions! Imagine, 14 million new transistors that yield a grand total of 42 million! WOW!

When all is said and done many of the PC-er's claims are full of sound a fury, signifying nothing. To illustrate my point, let us examine one of the Pentium 4's critical flaws more closely.

There is this little matter of deep pipelines to consider. In order to wow the Wintel partisans, Intel dreamed up a new name for the longer pipeline to make it seem to be a bigger deal than it really is. They call it "NetBurst Micro-Architecture". By giving an ordinary feature a new name it suddenly becomes something more important. Kind of like changing the job title to "Maintenance Engineer" from the old "Janitor".

Deeper data pipelines do allow higher operating frequencies, hence the Pentium III's current theoretical limit of 1.266 GHz at ten stages and the projected ability of the Pentium 4 to reach 2 Ghz, someday. Unfortunately, deeper pipelines have a serious downside as well. To quote David K. Every on the issue of deep pipelines:

"The pipes are deep but expensive -- they had to go very deep just to improve anything. The deeper you go in a pipe, the simpler each stage is, but the more complexity you have to create overall. Also the deeper you go, the faster in MHz your processor runs, but the more penalties you have when you stall and the less actually gets done in each cycle. So Intel has to run the processor faster (MHz), just to keep up with slower RISC chips. Overall -- they put in a very powerful pipeline, just to make up for the inefficiencies of the instruction set."

If Mr. Every's comments aren't enough to tell you that there are problems involved with the deeper pipelines in the new Pentium 4's twenty pipelines, then how about a couple of quotes clipped from a largely PC-centric site, Sharkey Extreme? Here are a couple of quotes from the "Sharkey Extreme Pentium 4 Guide".

"The first important point to consider is that processor performance is not determined solely by frequency (raw MHz). Rather, it is a function of frequency multiplied by IPC, or instructions per clock cycle. "

This is one of those little facts that invariable escape the notice of PC techies. In short, it means there is a lot more to processor performance than clock speed. Let's read on...

"Not all things are peachy in the land of the 20-stage pipeline, however. By doubling the depth of the branch prediction pipe, the penalty associated with mis-predictions is greatly increased - rather than flushing 10 speculatively executed instructions, the Pentium 4 has to flush 20, and start the execution over again in the correct program branch. The recovery time on the 20-stage pipe is much longer than the 10-stage pipe, resulting in a lower average number of instructions successfully executed per clock cycle."

This tells us that there is a trade-off to a long pipeline. The longer the pipeline, the higher a frequency at which the processor can run. However, since much of the way a modern processor works is to guess what speculative instruction (the processor is "guessing" what instruction will be used by reading a cache and shoving it down the pipeline) will be used and the unused ones have to be flushed each cycle. It takes time for the processor to recover from the bad guesses and the longer the pipepline, the longer the recovery time. In short, longer pipelines may run faster, but will run fewer valid instructions per clock cycle. RISC processors also can suffer from the same problems if the pipelines are long enough, but since RISC processing is inherently far more efficient than CISC they can run at higher frequencies with shorter pipelines and therefore will tend to process more instructions per clock cycle. Intel has added some new technologies to the new Pentium 4 in order to attempt to counter these problems, but they may not be enough to counter the inevitable flattening of an x86's performance per clock cycle as the clock speed increases.

As you read into the Sharkey Extreme article on the Pentium 4 you will discover something very interesting: it doesn't actually process data much better than the previous Pentium III, and in some ways it is actually slower than a GHz Pentium III or a 1.2 GHz AMD Athlon.

In fact, this unexpected performance problem has been noted by several news articles when the Pentium 4 was announced. Some articles merely regurgitated press releases from Intel or IBM in one form or another, such as InfoWorld's "IBM launches trio of Pentium 4-based desktops", or CNET News.com's "IBM taps new chip for gaming, movie fanatics", or the Sacramento Bee's "In speed war, Intel answers with Pentium 4" or even Semiconductor Business News' "Intel's future rides with Pentium 4 chip", all of which chose to ignore the problems with the Pentium 4.

What problems you ask? Well, where shall I start?

There is that niggling little problem with the RAMBUS memory chip not being quite what its paper specs implied and now Intel has a problem with unnecessarily high prices. This was mentioned in Tom Foremski's article "Intel introduces Pentium 4" at ft.com, where he mentions RAMBUS and the huge size of the Pentium 4 causing problems with the current fabrication process. Another article, "Pentium 4 ships amid doubts over Rambus" by John Leyden at vnu.net echoed the worry about RAMBUS.

Also, although Ken Popovich at eWeek tried to gloss over a few of the Pentium 4's problems and gave it a good try to make a happy report in his article "Intel sends Pentium 4 out the door", he did let slip that the Pentium 4 uses 50 watts of power to drive it, making it unlikely that you will see it in a consumer level home PC any time soon. Also mentioned was the fact that standard non-graphics business applications simply do not run faster, making the Pentium 4 a poor price/performance choice for business.

The truth about the Pentium 4's poor performance was explored in a long list of articles, such as Michael Kanellos' article "Pentium 4 fails to close gap on Athlon, say testers" of CNET, who opened the article with the following sentence:

"Intel's Pentium 4 chip released Monday doesn't provide a real performance advantage and is often slower when compared with the fastest Athlon chip from Advanced Micro Devices, benchmark testers and analysts said."

In other words, clock speed is simply not the deciding factor in attempting to assess the performance of different processors... exactly what we Mac types have been trying say!

Another article exposing the weaknesses of the Pentium 4 are "Under the Hood: An Inside Tour of the Pentium 4" by David Essex of PC World. Here is a bit clipped from the article:

"In PCWorld.com tests, the new chip barely keeps pace with the 1-GHz PIIIs used for comparison, and it even fell behind these older systems on some measures. Matched against a 1.2-GHz Athlon PC with Double Data Rate RAM, the P4 fares worse."

"They've come up with an architecture that clearly allows them to scale the clock rate," says Brookwood (Nathan Brookwood, Principal Analyst at Insight 64). "The challenge for Intel is to demonstrate that the clock-rate superiority translates into performance superiority."

In other words, as we have been saying all along, clock speed isn't the be-all and end-all for judging a processor's performance. It's interesting to see Wintel types finally admitting it!

In another article by David Essex at PC World, "Pentium 4 Ships: A Disappointment at 1.5 GHz", he says:

"Intel releases its long awaited Pentium 4 today to a surprise: PC WorldBench tests suggest you'll be initially served every bit as well, if not better, by older Pentium IIIs and Advanced Micro Devices Athlons."

For all of the PC loyalists telling us how "fast" a Pentium 4 was going to be, the reality is quite different. Mr. Essex points out what others have noted... a Pentium 4 simply isn't that fast in real world performance.

In fact, as Michelle Delio's article "Behold the new Pentium 4" at Wired news pointed out, the Pentium 4 only runs business applications faster than a GHz Pentium III by less than 10%. In fact, some of the articles I read quoted a 4% to 8% boost at best. In other words, the Pentium 4 just ain't quite the big Gee Whiz the PC Weenies have been telling us.

It does speed up games and some graphics instructions, but that is more due to the special new SIMD instructions in the SSE2 register (assuming the program is SSE2 enabled) rather than inherently faster capabilities of the main processor. They did add extra parallelism on the processor which should have boosted the performance noticeably for non-SSE2 enabled applications.

It didn't.

And they were poo-pooing us who were saying that x86 performance is flattening out, but the figures tell the tale: an estimated 10% general performance (I am being generous. The actual quoted figure was 4-8%) increase on a 50% clock speed increase translates into (*gasp*) about a 2% true performance increase per 10% of speed increase. Wait! just to be generous, let's assume he was using the cheaper and slower 1.4 GHz version. This then translates into a blazing 2.5% increase per 10% clock speed increase... DUH!

All I can see is that the performance curve is still flattening out for the x86. They can pump up the clock speed all they want, but at the price of fewer instructions processed per clock cycle, larger and therefore more expensive chips, more wattage (50 watts!) used, more heat produced and in the end, poorer cost/performance ratios.

Intel engineers are well aware of the limitations of their technology. In a Reuters report from October 10, 1999, "Intel scientist sees chip design limits", Paul Packan of Intel was quoted from an article he wrote in the September 24, 1999 issue of Science about the limitations of current computer processors are being reached. Paul Packan said that semiconductor engineers have not found ways around basic physical limits beyond the generation of silicon chips that will begin to appear next year. Since he said this in 1999, and he works for Intel, he means the current Pentium 4 processor and any successors.

Mr. Packan referred to the problem as:

"...the most difficult challenge the semiconductor industry has ever faced."

"These fundamental issues have not previously limited the scaling of transistors. There are currently no known solutions to these problems."

Of course, Intel's own Dilbert-esque Bosses tried to poo-poo the problems away. They cautioned against seeing the problem as insurmountable, adding they were confident answers could be found.

However, Dennis Allison, a Silicon Valley physicist and computer designer, tells us:

"The fact that this warning comes from Intel's process group is really significant. This says that they see actual limits."

The days of ever increasing performance from CISC technology are coming to a close. Of course, RISC technology also faces the same limits imposed by the laws of physics. Since we are rapidly approaching the limits of how small the components can be made, currently at around 100 atoms per transistor, then other means must be found to get around the limits.

Some have proposed that computer processors go to molecular computing, Eventually that may happen (or it may not...), but not in the foreseeable future. The nearer future is more likely to go to more parallelism in processing. Intel is trying to do this by adding parallel capabilities within a single processor. Both AMD and Motorola as well as IBM have opted for multiple cored processors. IBM's Power 4 already uses multiple cores and AMD's "Hammer" project plans dual cored AMD processors. Motorola has been rumored to have quadruple cores on a single chip planned for the G5, to be released sometime in 2002. The only problem here is that all of these projects are some time off and we are still stuck in the here-and-now.

Of course, all of these technical problems in the Wintel camp don't really relate to Mac users... or do they? After all, we are still stuck at a plodding 500 MHz... or are we?

In Twilight of Empire: Part Four I will put a little light on the subject and show how it relates to Macintosh.

Short Takes

In what could be characterized as the The Grass is greener On The Other Side Of The Fence category, I have known for some time from revelations made by a former girlfriend of Bill Gates that he is a Mac user, and so is Steve Ballmer. That may be a reason that Microsoft has decided to actually make good Mac products again, because Steve and Bill have to use them too!

On the other side of the coin... when Steve Jobs came back to Apple he didn't use a Mac or any other Apple product to do his work. His personal choice was a NeXT Cube and two Wintel laptops. Maybe that's where some of the more annoying Windows-like features in Mac OS X and Aqua came from.

Do you get the idea that all of these guys are just a little confused? How does the saying go... "Eat your own dog food."

joecarson@applelust.com