Monday, 11 October 2004
[Monday] [Tuesday] [Wednesday]
[Friday] [Saturday] [Sunday]
- If you were among the fifty or so subscribers who signed up
for a review copy of Building the
you should be getting it and a t-shirt very soon, if you haven't
already. The sooner you can write a review of the book and post it to
Amazon, the better. (You can post it to other places such as B&N,
other bookseller sites, user group sites, etc. as well, although
Amazon.com is the really critical place to post it.)
A couple of people have asked me for guidelines about writing a
review. I don't want to influence anyone, so all I'll say is that you
should write however much or however little you wish about the book,
focusing on (a) what you thought of the book generally, (b) anything
you particularly liked or disliked, or thought was useful, or thought
could have been done better. Personal observations and anecdotes are
helpful. When I'm writing a review, I often find it helpful to read
through the reviews that have already been posted to help me remember
points I wanted to make, to take issue with comments in other reviews
with which I disagree, and so on.
But the really important thing is to get your review posted to
Amazon as quickly as you can. Initial momentum counts for a lot.
I spent some time yesterday playing with Nvu,
which is an enhanced version of Mozilla Composer, sponsored by
Linspire. Nvu isn't FrontPage, not by a long shot, but does add many
nice features that are not present in Composer, not least of which is
an ftp site management function.
Barbara has become quite frustrated with the primitive publishing
functions available with Composer, as have I. Perhaps the worst aspect
of publishing with Composer is that it focuses on one page and two
directories--the directory that contains the page itself and a single
directory that contains images for that page. If, for example, a page
contains images that are stored in two different directories, there is
no convenient way to publish the page and both images.
There is also a problem with publishing defaults. It seems to me
that it would be reasonable to store the publishing parameters with the
page, so that if you published the page repeatedly (as we do with our
journal pages) the publishing parameters would be stored the first time
the page was published and then used subsequently until they were
It doesn't work that way with Composer. It insists, for example, in
trying to publish Barbara's journal page to her researchsolutions.net
site rather than to her fritchman.com site. Each time she publishes her
journal page, she has to remember to change the site to which she is
publishing from the default researchsolutions.net to fritchman.com. If
she forgets, the result is not a pretty sight (or site). I spent some
time yesterday using the gFTP client to delete extraneous files
manually from Barbara's site (and mine).
There are other problems with the Composer Publish function. For
example, incredibly, it is incapable of creating a directory on the
server. I found this out by chance. Barbara was attempting to put
together a new page that documented her recent trip to Vermont. After
struggling mightily with Composer's problems with directories and so
on, I decided to create a new root-level folder on her site, named
/vermont-2004. I put the page itself, vermont-2004.html, into that
directory, along with all of her images, both the full-size versions
and the thumbnails I'd had to generate manually with Irfanview. When I
attempted to publish the page and associated images, Composer's Publish
function blew up with an unhelpful error message. As it turned out,
that was because the /vermont-2004 directory did not exist on the
server. I had to go in with Xandros File Manager and create that
directory manually before she could publish to it.
Speaking of which, when I mentioned that I was looking for a good
GUI ftp client for Xandros, several people mentioned that I could
simply enter the ftp address in the URL bar of Xandros File Manager.
That's true enough, but the problem is that XFM isn't much of an ftp
client. For example, when I attempted to use it to publish Barbara's
/vermont-2004 directory by copy/paste from the local drive to the ftp
site, XFM blew up repeatedly with a message about too many connections.
From the little I've used it, gFTP seems to be a very competent GUI
ftp client, so I'll probably use it in combination with Composer's or
Nvu's Publish function--the latter when I'm simply publishing my
journal page routinely, and the former when I have multiple files to
publish, particularly if they're located in other directories.
Incidentally, one other nice thing about Nvu is that it produces
nicely formatted HTML source. Composer produces hideous-looking source,
with excess carriage returns and <br> tags all over the place,
lines that scroll
horizontally for several screen widths, and so on. Nvu provides nicely
formatted, compact HTML code.
Nvu is available for download from Xandros Networks, but only for
subscribers, which I'm not. Because I only wanted to experiment with
Nvu, I simply downloaded the tarball for Linspire (which is also
Debian-based), stuck it in my junk directory, extracted it, and ran Nvu
by running the executable script in the Nvu directory. So, it's not
installed, but I can use it. If I decide to use it permanently, I'll
get a Xandros-packaged install file.
For those of you who don't read German, these articles report that,
beginning next April, Germans who have a home PC will be required to
pay about $20 per month to subsidize public broadcasting. The fee is
waived if they already pay a television license, so in essence this
means that those who have a PC but do not watch public TV will be
forced to pay for public TV anyway. Note that this is not a tax in the
usual sense. The German government will collect it on behalf of public
TV. I hope this doesn't give PBS any ideas...
I've railed on about such subsidies at length in the past. They do no
good, and in fact do much harm. They favor entrenched interests at the
expense of innovation. Your phone bill is a good example. For most of
us, only a small fraction of what we pay each month is actually paying
for our telephone service. Most of it is subsidies, both visible and
invisible, direct and indirect.
Jerry Pournelle, for example, recently noticed that he was paying
something like $8/month to AT&T on a phone line that is never used
to place any long-distance calls, indeed one that places no outgoing
calls at all. This "pay us whether or not you use our service" is a
relatively new phenomenon, but is growing fast. That's not the worst of
The worst is all the subsidies we pay for so-called "universal access".
As I recall, a basic phone line in Winston-Salem is something like
$14/month (nominally, that is; the actual cost is much, much higher).
Of that $14, probably less than $5 (perhaps much less) goes to covering
the cost of providing a phone line within Winston-Salem. The rest is a
concealed subsidy that is spent on other things, such as providing
$14/month phone lines to people who live out in the sticks.
Out there, the actual cost of providing a phone line may be $50, $100,
or even $200/month, but rural residents pay only the standard $14,
leaving the rest of us to make up the difference. And what is really,
truly annoying is that if they want a second phone line, they pay only
$14/month for it as well. You'd think the first subsidized line would
have taken care of whatever supposed obligation there is to provide
"universal access" and that the phone company would charge them the
true cost for second and subsequent lines, but no.
The result of these subsidies, of course, is that there is no reason to
look for better methods. Why consider doing things a new way when it
costs only $14/month to have a phone line out in the middle of nowhere?
If there were no subsidies, people would be looking more cost-effective
ways of providing telephone service to rural users, such as building a
cellular or Wi-Fi network. Maintaining mile upon mile of copper wire is
insane when there are better, cheaper methods readily available. The
problem is, they aren't cheaper because the subsidies have artificially
reduced the apparent costs of doing things the old way.
And people are tiring of those subsidies, which is one of the main
reasons that Vonage and similar VoIP (Voice over IP) companies are
growing like crazy. There's a price war breaking out now. AT&T
reduced their all-you-can-eat VoIP plan to $29.95/month, so Vonage
dropped theirs to $24.95. That's still greatly in excess of the actual
costs involved, so there's a lot of room for them to drop prices
further still. And it's not just the telephone companies. Time-Warner
cable constantly runs ads soliciting our VoIP business. They put a
circular in every bill. It's a goldrush now, because their incremental
costs to provide the service are very small relative to the current
monthly rates for service. Eventually, over the next few years, the
cost of VoIP service will stabilize at something closer to the actual
costs involved. At that point, I expect we'll see VoIP service selling
for perhaps $5/month or less over the cost of our broadband
All you can eat, indeed. There's no real point to billing by the
minute, because the billing costs would make up the bulk of the bill.
VoIP providers can make more money by setting the charges for the
service slightly higher and not tracking usage. Nor is there any point
to billing by distance. IP doesn't care whether I'm talking to someone
across town or across the ocean. The real costs are the local loop,
whether that loop happens to be copper wire to the CO, a coax or
optical cable to the cable modem company, or a Wi-Fi wireless network.
Once your packets are outside the local loop and on the backbone, the
incremental cost to move one more packet is trivially small. When
Pournelle picks up the phone and calls me, for example, the vast
majority of the costs are in the few miles of local loop on his end and
the few more miles of local loop on my end. The thousands of miles
between Los Angeles and Winston-Salem account for a tiny, tiny fraction
of the cost.
What's interesting to me is that neither I nor most of my
technically-competent friends use VoIP. When Pournelle or Bilbrey wants
to call me, he picks up the phone and dials. At three cents a minute
(or whatever), who cares? The obvious moral here is that none of us
much cares about the cost of long distance. What may eventually
motivate all of us to move to VoIP is the monthly cost of local
service. That and the inexpensive or free features available with VoIP
service--automated attendant, voice mail, conferencing, caller ID, and
so on. Some of those services are also available with traditional
telephone service, but typically only if you're willing to pay a stiff
My guess is that most of us will begin our VoIP experience by using
Skype or another free VoIP application that's designed primarily to
connect to others using the same software. That provides free long
distance, but does nothing about the monthly bill for local service.
It'll be interesting to see where things go from there. Video
conferencing, shared whiteboards, and so on would be logical
extensions. It's going to be an interesting next few years.
And some Linux apps do have some bugs...
Don't look for much around here over the next week or two. I have some
- I just sent the following message to Jerry Pournelle:
Honors colleges appear to be all the
I wasn't aware of this phenomenon, but it appears to fit right into
what we've been discussing. Could this be the first step in the coming
bifurcation of higher education into "real" colleges, attended by real
students, and "pretend" colleges which are in business simply to take
money from so-called students and issue them credentials they haven't
earned? And indeed it may be one of the first steps toward a
bifurcation of society in general, into a tiny group who are truly
educated and a huge group who have only worthless, unearned credentials.
I haven't believed for years that any degree, including a Ph.D., from
any but a very few US colleges and universities is evidence of
anything. Oh, a hard-science or engineering degree from MIT, Cal Tech,
Duke, or a few other schools still means something, but that's about
it. Even the MD has been watered down by reduced admission requirements
intended to encourage diversity, whatever that means.
I wonder if schools are finally beginning to realize that they've lost
all credibility and are attempting to establish new, meaningful brand
names for themselves. I hope so.
Which is all true, but not Politically Correct. If someone tells me he
has a degree in Chemistry, Physics, Engineering, or another rigorous
discipline from, say, Winston-Salem State University, that means
absolutely nothing to me. I don't even take it as evidence that the
person can read, literally. And if someone tells me he has a degree in
a non-rigorous discipline from any school, I take that as an indication
that he has no education at all. Unfortunately, I'll be right more
often than I'm wrong.
Tuesday, 12 October 2004
[Monday] [Tuesday] [Wednesday]
[Thursday] [Friday] [Saturday]
[Sunday] [Next Week]
Pournelle sometimes asks his readers what additional capabilities they
would like computers to have. Invariably, his readers say they want
things that would require processors and other components several
orders of magnitude more powerful than we have now. Let's face it, the
fastest AMD or Intel desktop computer doesn't have the IQ of a
But I thought about something yesterday I'd really like to see, and
it's within the capabilities of current systems. I want intelligent
agents based on Operations Research methods to do my bidding. For
example, say I've just discovered an
author I really like. I want to buy all of her books, but most are out
of print. That means going to someplace like ABE Books and searching
down titles one by one. If she's published a dozen books, that means
finding all twelve individually. If I want a reading copy and a
collection copy of each, that's 24 separate searches, and probably 24
With my OR-based agent, I could simply enter various parameters and
have the agent do all the work. For example, I may want
hardback reading copies of all her books, but I may also want signed
first state (first edition/first printing) copies in Fine/Fine
or better condition, signed by the author. I want all 24 books at the
lowest overall cost, including shipping.
If I try to do that manually at ABE
Books, I have to do 24 separate searches, two for each title, because
the books listed in the ABE
Books database are actually in inventory at about 12,500
bookstores throughout the world. So, I can go through book-by-book and
find the best price, counting the
price of the book and shipping costs, for each book individually, but
not for the two dozen I want to buy as a group.
Because shipping can add
significantly to the cost, buying each individual book from the seller
with the apparent lowest price for it is often not optimal. My OR-based
agent would optimize for lowest cost, assuming that solution also met
the other constraints. For example, I might really want one particular
reading copy quickly. I may have already read the first four books in
the series, for example, and decided on that basis that I want reading
and collection copies of all of them. But my library doesn't have books
five, seven, and eleven in the series, so if I want to continue reading
the series in sequence I need book five (and possibly seven) quickly.
So I tell my OR-based agent that I want priority on delivering the
reading copy of book five (and possibly seven). It goes back and checks
the additional costs for expediting shipping from all the sources (not
just the ones it decided on initially) and tells me that it can have
books five and seven to me in two days, but at an extra cost of $12. It
also suggests that if I'm willing to wait three days, I can have books
five and seven in that time for only $3 more than the optimum solution
it calculated based on price. Alternatively, I can have book five in
two days for only $6 additional if I'm willing to wait four to five
days for book seven. And so on.
So I accept the second alternative proposed solution and tell my agent
to print me out a list of the proposed orders. But wait. In scanning
over the proposed solution, I see that my OR-based agent proposes to
order one of the collection copies from a bookseller that I've done
business with in the past and found to be unreliable. So I tell the
agent to remove that bookseller from consideration and recalculate.
The new solution increases the total price by only a few bucks, so I
tell my OR-based agent to go ahead and generate the orders. The final
solution orders the 24 books from 14 different bookstores. My agent
goes to the inventory database of each of the 14 stores and verifies
that the books are still available. It finds that 23 of them are in
fact in stock, but one has been sold and the database not yet updated.
So my agent re-runs against this new information and pops up a message
to tell me the solution I accepted is unworkable. The new proposed
solution is acceptable to me, although it involves 15 orders rather
than 14, so I tell my agent to go ahead and place the orders. It again
verifies in-stock status for each of the 24 books (this time at some
different stores) and finds that all are in stock. As it verifies each
book, it places a hold on it, but does not yet complete the order.
Once it has holds on all 24 books, it generates actual orders for all
24 books, sending the orders directly to the bookstores and notifying
me as each order is placed. My agent then uses the UPS and FedEx
tracking numbers provided to it by the bookstores to monitor delivery
status and to keep me informed of what's showing up when.
This is one of the "new" things I'd like my computers to do for me. And
it can all be done using existing technology.
A few years ago, I'd written Novell off. Although I was one of the
first to earn an Enterprise CNE (and, later, Master CNE) in North
Carolina, I'd let my certifications expire. There seemed to be no point
to keeping them. Novell seemed to be a stodgy old technology firm that
was stuck in the past, depending more every year on their installed
NetWare base to generate their revenues, which were declining year by
year. Almost no new NetWare installations were occurring, other than
those into established NetWare shops.
But, with their move from Utah to Cambridge and under the
stewardship of Jack Messman, Novell has succeeded in redefining itself
to become one of the most innovative and lively technology firms out
there. They've rebuilt their core business around Linux and open source
software, and in the process they've gone from being a ho-hum also-ran
technology company to one of the most important companies in the field.
Unlike some companies that talk the open-source talk but don't walk the
walk, Novell (along with IBM) really "gets" open source.
And, like IBM, Novell is willing to put its money where its mouth
is. After the failure of several previous attempts to kill Linux,
including the SCO debacle, Microsoft appears to shifting their strategy
towards using patents to strangle Linux. On the face of it, this
appears to be a good strategy. After all, Linux source code is freely
available to anyone, including Microsoft, so it's easy enough to search
Linux for patent violations.
But I think Microsoft's patent offensive against Linux is doomed to
fail, for the simple reason that they're bringing a knife to a
gunfight. Microsoft's patent portfolio is truly pathetic when compared
against IBM's patent portfolio. IBM, don't forget, is a staunch
defender of Linux and open source software. IBM has undertaken not to
attempt to enforce any of its patents against Linux, and it is a small
step from there to actually using its patents to defend Linux against
Microsoft and its stooges. If Microsoft goes up against IBM in a patent
war, Microsoft loses. It's as simple as that.
And now Novell
has weighed in,
promising to use its own substantial patent portfolio to defend Linux
against all comers, and that most specifically includes Microsoft.
People who worry that Microsoft will succeed in using its massive
resources to kill Linux and OSS invariably look upon Microsoft as the
big bully and Linux as the skinny little kid who wears glasses. What
they forget is that Godzilla and now King Kong are in Linux's corner.
If the big bully isn't careful, he may end up as a grease spot on the
Wednesday, 13 October 2004
[Wednesday] [Thursday] [Friday]
[Sunday] [Next Week]
Years Ago Today]
- Heh. I got email from one of my editors at O'Reilly last
night. He was
doing the copy edit on the second part of the Building the Perfect Bleeding-Edge PC
article, and he had a few questions. One of them had to do with the
name of a tool I mentioned in the article.
I called the tool "dykes", and the editor nearly had a heart attack. I
explained that this is what the tool is universally called by techs and
that, although I was sure there was a proper name for the tool, I
didn't remember what that might be. So I checked Sears and a couple
of other sources. The formal name is probably just as bad. They're
nippers" or, worse still, "butt nippers".
I suspect "dykes" originated as shorthand for "diagonal cutters", but
that's no longer what most people use it to mean. The techs I've known
cutters "diagonal cutters" or simply "cutters". When they say "dyke"
they invariably mean the flat-nosed plier that is used to cut
So, "dykes" they remain, despite the fact that my editor was concerned
about using a word that is also slang for lesbians. Which reminds me of
something that happened when I was at RIT, lo these many years ago.
Three of us were meeting up to go to some event or other--Dave, Ian
(who was British), and me. I spotted Ian in the lobby, but Dave was
nowhere to be seen. The event was about to start, so I asked Ian, "Have
you seen Dave?" His answer was classic, "He's in the gents, sucking a
- One thing I really need is a good crystal ball. I'm struggling
right now to write the Quick Reference that will appear on the inside
covers of the Pocket Guide. Here's a message I just sent to my editor
I've decided just to stub out the video
adapter section. There's no point to me wasting time on it now, because
everything is up in the air. ATi and nVIDIA are both in the midst of a
huge transition to new-generation chipsets. I have no idea what the
pricing will be on nearly all of the new cards, which makes it
impossible to suggest head-to-head competitors. It's not even clear
which adapters will be PCI-Express only versus PCI-X plus AGP.
The number of competing chipsets (and therefore adapter models) is also
staggering. Just nVIDIA now has the 6200, 6600, 6600 GT, 6600 Ultra,
6800, 6800 GT, and 6800 Ultra. nVIDIA just announced the 6200 Monday,
on no notice, and I expect they'll also announce 6200 GT and 6200 Ultra
versions at some point. ATi is even worse. And that's only their new
chipsets. Their older chipsets will remain in use, at least for a
while. And some of those older chipsets aren't very old at all. For
example, the ATi X600, X600 Pro, and X600 XT are now dead meat, in the
process of being replaced by the X700 series.
Even the tech people at ATi and nVIDIA don't really know what's going
on. Talk about a Chinese fire drill. Tom's Hardware is so disgusted
with the whole thing that they just posted a
, with which I agree completely.
Anand, reviewing the nVIDIA 6200, concludes:
"If all of the cards in this review
actually stick to their MSRPs, then the clear suggestion would be the
$149 ATI Radeon X700. In every single game outside of Doom 3, the X700
does extremely well, putting even the GeForce 6600 to shame; and in
Doom 3, the card holds its own with the 6600. Unfortunately, with the
X700 still not out on the streets, it's tough to say what sort of
prices it will command. For example, the GeForce 6600 is supposed to
have a street price of $149, but currently, it's selling for closer to
$170. So, as the pricing changes, so does our recommendation.
In most cases, the GeForce 6200 does significantly outperform the X300
and X600 Pro, its target competitors from ATI. The X300 is priced
significantly lower than the 6200's $129 - $149 range, so it should be
outperformed by the 6200 and it is. The X600 Pro is a bit more
price-competitive with the GeForce 6200, despite offering equal and
even greater performance in certain cases.
However, we end up back at square one. In order for the 6200 to be
truly successful, it needs to either hit well below its $129 - $149
price range, or ATI's X700 needs to be much more expensive than $149.
In the latter case, if the card is out of your budget, then the 6200 is
a reasonable option, but in the former case, you can't beat the X700.
Given that neither one of the cards we're debating about right now are
even out, anything said right now would be pure speculation. But keep
an eye on the retailers. When these cards do hit the streets, you
should know what the right decision should be."
Of course, the whole ATi X600 series is now dead meat. This is simply
14 October 2004
[Wednesday] [Thursday] [Friday]
[Sunday] [Next Week]
- FedEx showed up yesterday with a package from AMD with a
couple of Sempron processors (Socket A and Socket 754) and a Socket 939
Athlon 64 3500+. I hadn't been able to recommend the Athlon 64
previously because of the dearth of good motherboards available for it,
but that may change now that nVIDIA-based Athlon 64 motherboards are
beginning to appear in quantity.
One of the nice things about the AMD processors is that they consume
less power than Intel processors. That translates to less heat, quieter
fans, and a quieter system overall. I plan to build a Sempron system
with the Antec Phantom fanless power supply. With a Zalman CPU cooler,
that system should be nearly silent.
I really dislike Adobe Acrobat. I suppose it has its place for printed
documents, but for web presentation it is an abomination. Nothing like
having to scroll down one column, up, down the next column, up, down
the next, and so on ad infinitum. But, although I dislike Acrobat on
general principles, I really dislike the Linux implementation of
Acrobat Reader. Once it's fired up, it bonds itself to the Mozilla
Browser, and the only way to get rid of it is to close all instances of
Mozilla. Often, it's not convenient for me to do that, but there's
little choice. Acrobat, you see, also messes with XFree86. This screen
shot shows XFree86 at over 40% utilization, which is enough to make the
system slow to a crawl.
During a typical workday, I might have ten or more instances of Mozilla
open, with an average of perhaps half a dozen tabs each. If I
accidentally click on a PDF link, I'm screwed. As soon as Acrobat
Reader fires up, I'm doomed to have my system slow to a crawl. Not
immediately, but within a few minutes or perhaps an hour. It's bad
enough that I've seriously considered removing Acrobat Reader from my
systems and just using the built-in ability of Xandros File Manager to
display PDF files.
Which brings up another Xandros issue. The screensaver is flaky. It
stops working for no apparent reason. My main office desktop, which I'm
typing this on, runs Xandros 2.0. It's been installed for more than
three months now, and the screensaver continues to work as expected.
But this is the only Xandros system in the house that has a functioning
Xandros screensaver. Until a week ago, Barbara's Xandros 2.5 system had
a functioning screensaver, but it stopped working for no apparent
There is an easy solution. I open a run dialog, type xscreensaver-demo, and press
Enter. That screensaver works reliably and in fact has many more
features than the KDE screensaver that Xandros uses by default. I don't
know why Xandros doesn't enable it by default. As I have it set up on
Barbara's system right now, it will only work until the system is
rebooted. But I seem to remember down near the bottom of the man page
for xscreensaver-demo it tells me what I need to do to load it at boot
I'm using N|vu to write this page.
It's available from Xandros Networks, but only if you're a premium
subscriber. I'd played around with N|vu a bit a couple of days ago
before I got the official Xandros version. I simply visited the N|vu
site and downloaded the package for SuSE. Double-clicking on the file
fired up Xandros Networks and extracted the package. It didn't install
it in the sense of adding it to the apt database or putting it on the
menu, but I was able to run N|vu by executing a script in the directory
I'd extracted it to. That was enough to convince me it was worth a
closer look, so I emailed my contact at Xandros to ask for a code to
access the premium site. They sent that to me within minutes, and I had
N|vu up and running minutes after that.
On the web site, N|vu claims
"Finally! A complete Web
Authoring System for Linux Desktop users as well as Microsoft Windows
users to rival programs like FrontPage and Dreamweaver."
which is, to put it kindly, a bit optimistic. I've never done more with
DreamWeaver than load it and spend a few minutes playing with it, but I
have used FrontPage for years, at first FrontPage 98 and later
FrontPage 2000. As far as I can see, albeit with admittedly little
experience as yet with N|vu, N|vu rivals FrontPage in the same way that
a Yugo rivals the Queen Elizabeth II.
Well, perhaps that's a bit harsh, but, as far as I can see, N|vu is
still just an HTML page editor. They have what they call the "Site
Manager" but unless I'm missing a lot it's not a site manager in
anything like the same way that FrontPage is a site manager. I'll play
with N|vu a lot more. Perhaps my initial impressions are wrong. And, of
course, like a lot of OSS software, N|vu is being actively developed.
[Saturday] [Sunday] [Next
- Was John Kerry dishonorably discharged? This guy
asks some questions that need to be answered. Although Mr. Kerry
introduced military service as a part of the campaign and has made much
of contrasting his service in Viet Nam with Mr. Bush's service in the
National Guard, Mr. Kerry has so far refused to allow all of his own
military records to be released.
Why has he not allowed those records to be released, if he has nothing
to hide? If there is no truth to the accusation, it would be easy
enough to refute. The fact that Mr. Kerry refuses to refute the
accusation is good reason to believe that he cannot refute it.
- I read two articles in The Inquirer today (Part 1 and Part 2) that
discuss the problems Intel faces over the next couple of years. The
author nailed the situation exactly. Intel is in deep trouble, as
evidenced by their recent cancellation of the 4 GHz Pentium 4. They
expected the Prescott core and its follow-ons to scale to 10 GHz or
more. They were wrong. At only 4 GHz, the Prescott is out of headroom,
leaving Intel with nothing to counter the onslaught of new AMD
AMD, which not all that long ago looked to be down for the count, is
now in the catbird seat. They made some huge gambles, notably their SOI
(silicon on insulator) process. Those gambles have paid off, and now
AMD can easily trump the best Intel has to offer. The pendulum has once
again swung, and it's AMD's turn to deal from a position of strength.
I am reminded of the Who.
"The parting on the left is now a
parting on the right,
and the beards have all grown longer overnight..."
If I created my own dictionary, the AMD and Intel logos would appear
next to the word "competition", because these two companies embody the
meaning of the word. They are ferocious competitors, but both compete
honestly and above-board. They let their products speak for themselves,
unlike a large company I could name that thinks nothing of using even
the most underhanded tactics.
Although I'm sure AMD and Intel both wish the other company would
simply disappear, the fact is that their competition has been good for
everyone. Obviously, it has been good for consumers. If it weren't for
AMD, the fastest processors Intel sells would probably be stuck at well
under a gigahertz, and we'd pay $1,000 for the "fast" models. It's been
good for Microsoft, which can continue to introduce ever more bloated
operating systems and applications. Less obviously, it's been good for
AMD and Intel themselves, who might have otherwise rested on their
The current situation is perilous for Intel, obviously, but never count
Intel out. They have massive resources, and they employ a lot of very
smart engineers. Less obviously, the current situation is perilous for
AMD. AMD has numerous weaknesses, not least their limited fab capacity.
If Intel literally disappeared tomorrow, leaving the entire market to
AMD, AMD could meet only 30% of the current demand for processors. AMD
is also famous for failing to execute, and they face several
challenges, including their move to a 65 nanometer process. If AMD is
not very careful, they could stumble. And a small stumble might suffice
to allow Intel to come roaring back.
The major problem AMD has always had is their apparent lack of a killer
instinct. Right now, when they have Intel down, is the time to pour it
on. AMD should be shipping faster and faster processors, as quickly as
they can get them out the door. Instead, when it has the advantage, AMD
has historically been satisfied to beat Intel only by a bit. I'm sure
AMD's MBAs are telling them that they leave money on the table by
shipping faster processors than they need to. That argument is
superficially attractive, but the fact is that AMD is Avis to Intel's
Hertz, and AMD has to not just beat Intel, but slaughter them.
If Intel struggles mightily to deliver a 3.8 GHz Pentium 4 that is
available only in small quantities, AMD shouldn't match Intel's best
effort with an Athlon 64 3800+ or even 4000+ of their own, priced at
about the same level as Intel's flagship processor. AMD should instead
flood the market with cheap Athlon 64 4000+, 4500+, and even 5000+
Some might argue that AMD's limited fab capacity means that strategy
would indeed leave large amounts of money on the table, but I think
they're missing the point. AMD's goal should be to develop and nurture
the perception in consumers' minds that AMD processors are first-rate
and that Intel is an also-ran. The way to do that is to kick Intel when
AMD has problems of its own, of course, many of which can be traced to
its small size and limited resources compared to Intel and its
consequent lack of vertical integration. In concentrating all of its
efforts on producing a competitive processor, AMD has failed to devote
resources to match Intel's twin trump cards, chipsets and motherboards.
For a long time, that hampered the success of AMD processors. No matter
how good the processor, no one with sense will use it if that means
using a motherboard built around a third-rate chipset from VIA, SiS, or
Fortunately for AMD, nVIDIA came to their rescue with the superb
nForce3-series chipsets for AMD processors. Also fortunately for AMD,
motherboards from quality manufacturers like ASUS and built on recent
nForce3 chipsets are becoming widely available. In terms of stability
and compatibility, there is now little to choose between a Pentium 4
running on an Intel motherboard and an Athlon 64 running in an nForce3
My next project systems will be built around AMD processors. I'll be
building three systems, one with a Socket A Sempron 2800+, one with a
Socket 754 Sempron 3100+, and one with a Socket 939 Athlon 64 3500+.
These are long-term project systems that I will use heavily and
evaluate based on that real-world use. You'll hear lots more from me
about them over the coming months. But my initial belief is that these
current-generation AMD systems will present a very strong challenge to
Intel's Pentium 4 hegemony.
16 October 2004
[Saturday] [Sunday] [Next
17 October 2004
[Saturday] [Sunday] [Next
© 1998, 1999, 2000, 2001, 2002, 2003, 2004 by Robert Bruce
Thompson. All Rights Reserved.