Intel defeats Risc

  • Thread starter Thread starter Yousuf Khan
  • Start date Start date
Yousuf Khan said:
Well, the article makes it look like Intel's architecture has defeated
RISC. That may be true, but Intel's architecture (x86) looks like its on
the verge of defeating Intel itself too.


Guardian Unlimited Technology | Technology | Jack Schofield: Risc
assessment
http://technology.guardian.co.uk/online/insideit/story/0,13270,1506989,00.html

Yousuf Khan

Yes, the article certainly suggests that Intel is in trouble:
"... Intel now owns the workstation market ..."

--

... Hank

http://home.earthlink.net/~horedson
http://home.earthlink.net/~w0rli
 
Yes, the article certainly suggests that Intel is in trouble:
"... Intel now owns the workstation market ..."

Don't be silly - if Intel had its way, x86 would forever stay 32 bit
and by now relegated to the low end consumer PCs, with IA64 ruling the
world. Didn't happen...

NNN
 
Well, the article makes it look like Intel's architecture has defeated
RISC. That may be true, but Intel's architecture (x86) looks like its on
the verge of defeating Intel itself too.


Guardian Unlimited Technology | Technology | Jack Schofield: Risc assessment
http://technology.guardian.co.uk/online/insideit/story/0,13270,1506989,00.html

Aw hell, it's only The Grauniad... utterly clueless about anything &
everything and this guy is "out to lunch"... one of those journo wannabe
computer experts. Someone should ask him what's running under the dash of
his car - very likely it's an embedded PowerPC if it's fairly new.
 
George said:
Aw hell, it's only The Grauniad... utterly clueless about anything &
everything and this guy is "out to lunch"... one of those journo wannabe
computer experts. Someone should ask him what's running under the dash of
his car - very likely it's an embedded PowerPC if it's fairly new.

Yeah but for a mainstream journo, he got pretty close in a few points.
Much better than the usual wrong or laughably wrong on every point
mainstreamers. Even the "expert" press on computers are often laughably
wrong.

Yousuf Khan
 
Aw hell, it's only The Grauniad... utterly clueless about anything &
everything and this guy is "out to lunch"... one of those journo wannabe
computer experts. Someone should ask him what's running under the dash of
his car - very likely it's an embedded PowerPC if it's fairly new.

The guy actually wasn't all that far off, and he did mention that ARM
has done very well in embedded applications while PowerPC has done
quite well in the gaming console market (and some others).

For a generally fluff article without much of any real content, the
guy actually got most of the points right. Now if he tried to do a
more in-depth article, well then things might have been different.
But for this very superficial article, most points were pretty much
bang-on.
 
The guy actually wasn't all that far off, and he did mention that ARM
has done very well in embedded applications while PowerPC has done
quite well in the gaming console market (and some others).

For a generally fluff article without much of any real content, the
guy actually got most of the points right. Now if he tried to do a
more in-depth article, well then things might have been different.
But for this very superficial article, most points were pretty much
bang-on.

Sorry but I think his ignorance is so profound he should not be presented
as an IT expert - typical Grauniad... ethically barren. Mentioning PowerPC
as game console material is glaring in its omission of the embedded and
other apps like especially Cell. Are companies really "busy replacing
expensive Risc servers with cheap and cheerful x86-based models"? I don't
think so.... x86-64 maybe some but even if it is, "cheap & cheerful" does
not cover it. If a company had an "expensive Risc server" they're going to
need a helluva lot more than what he contends.
 
Sorry but I think his ignorance is so profound he should not be presented
as an IT expert - typical Grauniad... ethically barren. Mentioning PowerPC
as game console material is glaring in its omission of the embedded and
other apps like especially Cell.

Come on! It's a 1-page very superficial overview of things. He
didn't say that PowerPC was ONLY being used in gaming consoles, he
just mentioned that as one *example* of areas where this chip is doing
very well.

Remember that The Cell was developed originally for use in the
PlayStation3. It's only recently that IBM et. al. have started
branching out and saying that it will be used elsewhere.

Similarly when talking about ARM he only provides two examples of
where it's used (PDAs and cell phones). Certainly he could have gone
on for pages with the MANY other uses of ARM (which, as I'm sure you
know, outsells x86 and damn near every other ISA out there), but
really most people don't care.
Are companies really "busy replacing
expensive Risc servers with cheap and cheerful x86-based models"? I don't
think so.... x86-64 maybe some but even if it is, "cheap & cheerful" does
not cover it. If a company had an "expensive Risc server" they're going to
need a helluva lot more than what he contends.

Every company I've seen in the past 5+ years has indeed been busy
replacing any old RISC servers with x86 (including x86-64) stuff.
Sure, the big RISC servers still have their place, but in my
experience I've come across rather few *new* projects that companies
are targeting for new RISC servers. The vast majority of cases I've
seen have been companies phasing out old applications on RISC servers
and bringing in newer replacements on x86 stuff.

Keep in mind that only 5 years ago the big RISC and proprietary
servers from Sun, IBM, HP and all the others sold more (dollar wise)
than all x86 servers combined. Today x86 has not only caught and
passed old RISC servers, but they're quickly leaving them behind. For
Q2 of 2005 x86/x86-64 servers sold $5.7B vs. $4.3B for Unix servers.
From the previous quarter that was a 15.1% growth for x86 vs. only 3%
growth for Unix servers.

Source:
http://www.idc.com/getdoc.jsp?pid=23571113&containerId=prUS00223005
 
George said:
Sorry but I think his ignorance is so profound he should not be presented
as an IT expert - typical Grauniad... ethically barren. Mentioning PowerPC
as game console material is glaring in its omission of the embedded and
other apps like especially Cell. Are companies really "busy replacing
expensive Risc servers with cheap and cheerful x86-based models"? I don't
think so.... x86-64 maybe some but even if it is, "cheap & cheerful" does
not cover it. If a company had an "expensive Risc server" they're going to
need a helluva lot more than what he contends.

Actually, I see that all over the place nowadays: RISC being replaced
by x86 for various classes of applications. For example, most shops are
standardizing on Exchange for mail services, replacing whatever Unix
boxes they were previously using for mail. Almost all DNS is done on
Windows now in most places I look at, because it's just too unimportant
to dedicate a Unix box to it anymore. Sure, most people grumble when
they hear that management has decided to replace a perfectly working
Unix mail or dns server with a Windows box with Microsoft software. But
they grumble more about the Microsoft software than about the fact that
they're going to an x86 box.

Yousuf Khan
 
Actually, I see that all over the place nowadays: RISC being replaced
by x86 for various classes of applications.

My point is that they are not "cheap & cheerful" x86 boxes - the author
would appear to be unaware of the difference.
For example, most shops are
standardizing on Exchange for mail services, replacing whatever Unix
boxes they were previously using for mail. Almost all DNS is done on
Windows now in most places I look at, because it's just too unimportant
to dedicate a Unix box to it anymore. Sure, most people grumble when
they hear that management has decided to replace a perfectly working
Unix mail or dns server with a Windows box with Microsoft software. But
they grumble more about the Microsoft software than about the fact that
they're going to an x86 box.

This is a whole different subject: x86 !--> Windows. Mind you, assuming
you are in a position to speak for global deployment of Domain Control
serving, I *do* hope that this is technical management which is making such
decisions... and not some PHB who has been convinced by the M$'s lying
marketspeak that "it's all point 'n' click"... easy stuff which can be
handled by low-level techni-droids.
 
Come on! It's a 1-page very superficial overview of things. He
didn't say that PowerPC was ONLY being used in gaming consoles, he
just mentioned that as one *example* of areas where this chip is doing
very well.

Worse than a CNN sound-bite.:-)
Remember that The Cell was developed originally for use in the
PlayStation3. It's only recently that IBM et. al. have started
branching out and saying that it will be used elsewhere.

It kinda looks like the IBM/Mercury alliance already has come up with some
heavy duty stuff... which a x86 could never cope with.
Similarly when talking about ARM he only provides two examples of
where it's used (PDAs and cell phones). Certainly he could have gone
on for pages with the MANY other uses of ARM (which, as I'm sure you
know, outsells x86 and damn near every other ISA out there), but
really most people don't care.


Every company I've seen in the past 5+ years has indeed been busy
replacing any old RISC servers with x86 (including x86-64) stuff.
Sure, the big RISC servers still have their place, but in my
experience I've come across rather few *new* projects that companies
are targeting for new RISC servers. The vast majority of cases I've
seen have been companies phasing out old applications on RISC servers
and bringing in newer replacements on x86 stuff.

Keep in mind that only 5 years ago the big RISC and proprietary
servers from Sun, IBM, HP and all the others sold more (dollar wise)
than all x86 servers combined. Today x86 has not only caught and
passed old RISC servers, but they're quickly leaving them behind. For
Q2 of 2005 x86/x86-64 servers sold $5.7B vs. $4.3B for Unix servers.
From the previous quarter that was a 15.1% growth for x86 vs. only 3%
growth for Unix servers.

Source:
http://www.idc.com/getdoc.jsp?pid=23571113&containerId=prUS00223005

None of which is "cheap & cheerful". Just the RAS component dwarfs the CPU
as a cost element in the system. IMO the article is misleading.
 
George said:
My point is that they are not "cheap & cheerful" x86 boxes - the author
would appear to be unaware of the difference.

Well, I'm not actually sure how you would classify "cheap" in your
opinion. To me, x86 boxes are definitely cheaper than RISC boxes. Often
an order of magnitude cheaper. As for "chearful", I'm not privy to the
state of mood of these boxes. :-)
This is a whole different subject: x86 !--> Windows. Mind you, assuming
you are in a position to speak for global deployment of Domain Control
serving, I *do* hope that this is technical management which is making such
decisions... and not some PHB who has been convinced by the M$'s lying
marketspeak that "it's all point 'n' click"... easy stuff which can be
handled by low-level techni-droids.

Yeah, well these days there's not a lot of difference between technical
management and PHBs. Everybody is into cost cutting. But yeah, even the
Windows guys get surprised when they're told that they're replacing
some Unix box with Windows-based software.

These days about the only thing left on Unix are Oracle databases.
Everything else has gone Windows or Linux. Even the Oracles will go
mostly Linux pretty soon. Not existing installations of Oracle, but any
new Oracles that are purchased will likely go with Linux.

Yousuf Khan
 
It kinda looks like the IBM/Mercury alliance already has come up with some
heavy duty stuff... which a x86 could never cope with.

The workloads will be different, but I'd hardly say that x86 could
"never cope with" any sort of heavy duty workstation. There are LOTS
of x86 systems doing VERY heavy duty stuff.

Cell is no more of a 'high-end' or 'workstation' processor than x86
is, it's just designed rather differently. There are some workloads
which will be EXTREMELY fast on Cell and there are others that are
going to be VERY slow.
None of which is "cheap & cheerful". Just the RAS component dwarfs the CPU
as a cost element in the system. IMO the article is misleading.

They are definitely cheaper than what RISC used to be, though Sun and
IBM have had to DRAMATICALLY reduce the price of some of their
lower-end RISC boxes so that they can compete.

I guess "cheap" is all relative, but when you compare the average cost
of a RISC server to the average cost of an x86 server you will find
that the RISC ones tend to be about 5-10x more. Different markets and
different applications to be sure, but the price difference is quite
noticeable.

As for the "cheerful" part, I'm not sure that I'm really qualified to
judge that! :>
 
Well, I'm not actually sure how you would classify "cheap" in your
opinion. To me, x86 boxes are definitely cheaper than RISC boxes. Often
an order of magnitude cheaper. As for "chearful", I'm not privy to the
state of mood of these boxes. :-)

Ahh... you've missed the English idiom there - I wasn't sure if that had
migrated to Canada: "cheap & cheerful" is an English term applied to
cheaper, usually tacky, versions of the real thing, e.g. things like a
Tandy branded "Hi-Fi system", a Trabant car or the likes. IMO it was a
gross hyperbole by the author... but typical of what gets past the
incompetent/misguided tomfoolery of the editor(s) at the Grauniad... which
even has blogs dedicated to its "ablution":
http://dailyablution.blogs.com/the_daily_ablution/2005/08/more_humiliatio.html

As for price, is a similarly equipped x86 really much cheaper than a
Risc-based server?... say a low cost Sparc vs. an equivalent power Opteron?
The cost of the CPU as a contributor to final price is pretty small when
you add in the RAS & lights-off management features and a capable chipset
for a real server. When you add in the costs of M$ server offerings the
final price ramps up pretty quickly. Could be that the "Risc-CPU" vendors
have been inflating the prices... overrating the premium due for the unique
CPU??

M$ also seems to have scaled back the number of licenses required to run a
SMB Server/Internet operation fairly recently - previously, "recommended"
server/Internet ops configurations were extravagant.
Yeah, well these days there's not a lot of difference between technical
management and PHBs. Everybody is into cost cutting. But yeah, even the
Windows guys get surprised when they're told that they're replacing
some Unix box with Windows-based software.

These days about the only thing left on Unix are Oracle databases.
Everything else has gone Windows or Linux. Even the Oracles will go
mostly Linux pretty soon. Not existing installations of Oracle, but any
new Oracles that are purchased will likely go with Linux.

My work doesn't bring me into direct contact with many others doing
server/Internet ops but I follow the industry fairly closely and I haven't
seen any indication of a noticeable widespread trend towards Windows there
in any of the published material, e.g. at Network World and similar
industry watchers... and I certainly hope that we're not going to see a
growth in ActiveX polluted Web sites.:-)

Recent surveys still seem to indicate a huge lead for Apache vs. IIS;
whether it's Unix or Linux.... what's the difference? With $4.6Billion of
factory revenue for Unix servers in 2Q05
http://www.idc.com/getdoc.jsp?containerId=prUS00223005 it doesn't exactly
appear to be fading away. Ranking in terms of "factory" $$ is not the only
way to look at this of course but even there, the strongest growth is in
Linux it would appear.
 
The workloads will be different, but I'd hardly say that x86 could
"never cope with" any sort of heavy duty workstation. There are LOTS
of x86 systems doing VERY heavy duty stuff.

Cell is no more of a 'high-end' or 'workstation' processor than x86
is, it's just designed rather differently. There are some workloads
which will be EXTREMELY fast on Cell and there are others that are
going to be VERY slow.

Well yes, precisely that: where Cell is expected to score well, e.g.
Mecury's medical imaging app, x86 couldn't do the same job.
They are definitely cheaper than what RISC used to be, though Sun and
IBM have had to DRAMATICALLY reduce the price of some of their
lower-end RISC boxes so that they can compete.

Makes ya wonder... about about cost vs. price and overvalued premiums
applied by marketing for their "high cachet" products.
I guess "cheap" is all relative, but when you compare the average cost
of a RISC server to the average cost of an x86 server you will find
that the RISC ones tend to be about 5-10x more. Different markets and
different applications to be sure, but the price difference is quite
noticeable.

As for the "cheerful" part, I'm not sure that I'm really qualified to
judge that! :>

In England "cheap & cheerful" go together - it's quite an indictment
against any product. Packard Bell was "cheap & cheerful".:-)
 
George said:
As for price, is a similarly equipped x86 really much cheaper than a
Risc-based server?... say a low cost Sparc vs. an equivalent power Opteron?
The cost of the CPU as a contributor to final price is pretty small when
you add in the RAS & lights-off management features and a capable chipset
for a real server. When you add in the costs of M$ server offerings the
final price ramps up pretty quickly. Could be that the "Risc-CPU" vendors
have been inflating the prices... overrating the premium due for the unique
CPU??

The answer is a resounding yes. Where I worked in 2002 we used HP
workstations (J6000). These were 2 cpu, 4GB systems, the top of the
HPUX workstation line at the time. These machines cost $25,000 each and
every employee had one for technical simulation, and then there was the
server farm. We'd been using them for a decade or more. Later that
year we spent a few months porting the software to x86/Linux and bought
Xeon systems. These systems were 2 cpu, 4GB systems also, and the top
of the workstation line for x86. They were IBM machines and so we
surely overpaid to get the top level of service and we still only paid
$5,000 per machine. Not only did we pay 1/5th per machine, but the
simulation speed immediately jumped up 2x-4x. So the price/performance
advantage we saw was 10x-20x. We replaced several hundred HPUX machines
after that.

Alex
 
Tony said:
Remember that The Cell was developed originally for use in the
PlayStation3. It's only recently that IBM et. al. have started
branching out and saying that it will be used elsewhere.

The question becomes "where elsewhere?" Other embedded applications?
That sounds right, although they need to be BIG applications unless I
miss the price points of PPC and ARM, as well as Intel's RISC CPUs.

As general scientific workstations the question would be software. My
guess would be that if this were to be done at all it would be using
Linux, because I can't imagine ports of AIX, Solaris, or Windows
happening, and I am unconvinced that any other O/S is viable in this
market, just looking at the tools available.
 
George said:
Ahh... you've missed the English idiom there - I wasn't sure if that had
migrated to Canada: "cheap & cheerful" is an English term applied to
cheaper, usually tacky, versions of the real thing, e.g. things like a
Tandy branded "Hi-Fi system", a Trabant car or the likes. IMO it was a
gross hyperbole by the author... but typical of what gets past the
incompetent/misguided tomfoolery of the editor(s) at the Grauniad... which
even has blogs dedicated to its "ablution":
http://dailyablution.blogs.com/the_daily_ablution/2005/08/more_humiliatio.html

Actually, it's the first time I've even gotten a hint that you're from
out UK way. You may have mentioned it to other people in some other
threads, but I missed those completely.

But of course, you're from there, you got a colorful derogaratory term
already available for the Guardian newspaper, which would indicate
you're locally familiar with it.
As for price, is a similarly equipped x86 really much cheaper than a
Risc-based server?... say a low cost Sparc vs. an equivalent power Opteron?
The cost of the CPU as a contributor to final price is pretty small when
you add in the RAS & lights-off management features and a capable chipset
for a real server. When you add in the costs of M$ server offerings the
final price ramps up pretty quickly. Could be that the "Risc-CPU" vendors
have been inflating the prices... overrating the premium due for the unique
CPU??

The new Sun Fire (Galaxy) x2100 server costs about US$750 in its most
basic configuration. It has LOM and RAS features included. Sun tried to
sell some crappy UltraSparc boxes a few years ago (Sun Ultra 5 & 10)
for cheap but they never went below $1000. Usually they never go below
$10,000.
My work doesn't bring me into direct contact with many others doing
server/Internet ops but I follow the industry fairly closely and I haven't
seen any indication of a noticeable widespread trend towards Windows there
in any of the published material, e.g. at Network World and similar
industry watchers... and I certainly hope that we're not going to see a
growth in ActiveX polluted Web sites.:-)

Not just Windows, but x86 in general. For example, websites seem to be
going overwhelmingly Linux and Apache, replacing Sparc/Solaris and
Apache. Microsoft IIS is almost dead now, killed by its own reputation
for being an unsecure POS.

However, in the database side of things, you'll see equal amounts of
Oracle and MS-SQLserver. SQLserver on smaller projects, Oracle on the
bigger ones. Oracle you'll see running on either Windows or Linux quite
often these days too, again replacing most of the Unixes (including
AIX, Solaris, and HPUX). Whether they choose Windows or Linux largely
depends on whether there is any local Linux/Unix expertise, which
usually there isn't.

Other things you might see running on x86 more often than not these
days are Windows-based Active Directory (which could be used to replace
DNS, since it includes the DNS protocol); in many ways, AD is superior
to old fashioned BIND, simply because of its automation -- it adds
servers automatically to its list without needing to change a config
file. Network backup services, such as Veritas Netbackup, Tivoli
Storage Manager, Legato, etc., are often being run off of Windows
servers rather than Unix ones; they'll even be backing up Unix servers.
Mail servers, namely Exchange, is the undisputed corporate email
standard, and runs only on Windows.
Recent surveys still seem to indicate a huge lead for Apache vs. IIS;
whether it's Unix or Linux.... what's the difference? With $4.6Billion of
factory revenue for Unix servers in 2Q05
http://www.idc.com/getdoc.jsp?containerId=prUS00223005 it doesn't exactly
appear to be fading away. Ranking in terms of "factory" $$ is not the only
way to look at this of course but even there, the strongest growth is in
Linux it would appear.

Misleading to compare revenues rather than actual numbers. The reason
Unix revenue is so high is precisely because it is too expensive.
People are simply not willing to play that much for servers anymore
these days, unless they have no choice. In other words if they already
have a preexisting RISC Unix installation, and they want to upgrade the
hardware they will go with the exact same RISC/Unix combo they already
have. You won't even see them move from one RISC/Unix to another
RISC/Unix, because if they could migrate in that direction, they might
as well have migrated all of the way to x86/Windows. So RISC/Unix is
being supported mainly on repeat sales these days.

Yousuf Khan
 
The answer is a resounding yes. Where I worked in 2002 we used HP
workstations (J6000). These were 2 cpu, 4GB systems, the top of the
HPUX workstation line at the time. These machines cost $25,000 each and
every employee had one for technical simulation, and then there was the
server farm. We'd been using them for a decade or more. Later that
year we spent a few months porting the software to x86/Linux and bought
Xeon systems. These systems were 2 cpu, 4GB systems also, and the top
of the workstation line for x86. They were IBM machines and so we
surely overpaid to get the top level of service and we still only paid
$5,000 per machine. Not only did we pay 1/5th per machine, but the
simulation speed immediately jumped up 2x-4x. So the price/performance
advantage we saw was 10x-20x. We replaced several hundred HPUX machines
after that.

Hmm, you'd been using the HP-UX systems for a decade? I'm sure there are
better, contemporary examples for comparison.
 
Back
Top