PII vs PIII

[Steve Wolfe]

Nice dodge

now you can show me some "usability" tests or retract your statements.

The dodging was done on your part. You ignored every statement I made,
and refused to answer the question I posed.

steve

 

[Steve Wolfe]

Great even more claims and nothing to back them up.

You haven't done any backing up, either. All you've done is say "I was at
someone's site and saw some benchmarks, dood!"

We have experience with SMP systems, and you don't. You'll understand, of
course, if we find it hard to take you seriously.

Are you starting to get the picture yet. Why are there all these claims
about dual CPU systems and nothing to confirm them with. The fact is their
slow and not worth the money spent for the desktop regardless of what is
said about them.

Translation: "I'm too poor to buy an SMP system, I've never used one, and
I think I know what I'm talking about."

steve

 

[Gregory L. Hansen]

snip

Why do you keep asking about benchmarks? Who said anything about
benchmarks? Dual processor machines are about twice as fast on Photoshop
filters than a single processor of the same type, those sorts of stats
get thrown around comp.sys.mac.advocacy now and then. But we're not
talking about number crunching speed.


Everyone I've talked to (including people in real life) seem to think that
dual processor machines make wonderful desktop machines -- smoother and
more responsive. And faster. Certainly faster even for a single
application if it's multithreaded, but also just because it doesn't have
the operating systme interrupting it all the time.

Two processors versus a single processor of the same type, of course. Two
processors of half the clock speed as the single processor may or may not
be as fast depending on things like how well factored the code is and how
important it is to have twice the cache. It depends on what you're doing
with it.

I happen to have a machine sitting beside me that I can plug a second
processor into. Can't think of a reason not to.

Great even more claims and nothing to back them up.

[qt]
"Dual processor machines are about twice as fast on Photoshop filters
than a single processor of the same type."
[end]

It's something I saw from benchmark tests quoted on comp.sys.mac.advocacy.
They're always arguing about PowerPC versus x86 performance over there.
But I can't think of a reason to track down the reference for you. Feel
free to follow up on it yourself, though.

This one here is blatantly false, nothing runs twice as fast on a dual
machine due to overhead. At best you might see a 10 percent improvement and
a 1.2 gig single will smoke a dual 600 in loading and processing Photoshop
filters.

You just made that up. If there were *that* much overhead, nobody would
be making 4000 processor supercomputers! There's overhead when a single
processor switches from one task to another, too, you know.

I could believe a mere 10 percent improvement or less if you're running
only one, single-threaded task. And I think it was only recently that
Windows got decent multithreading, so it would be only recently that
Windows could take real advantage of more than one processor. But that
has nothing to do with overhead. Maybe for what you do (video games?)
there's little to be gained. But when you're doing more than one thing
at a time (e.g. background tasks, multiple users), or have a problem that
can be factored and multithreaded, there can be a substantial advantage
to having as many processors as you can get.

There's also the matter that if your data sets are so large or your cache
so small that you have to hit main memory all the time, you're going to be
slowed down by how fast you can push data through. If you're data-starved
at 850 MHz, then it won't do you any good to push the clock speed higher
than that, will it? You must know that, all else being equal, performance
just doesn't scale linearly with clock speed.

And all performance arguments aside, if everyone that uses multiprocessor
machines say they're smoother and more responsive in user interface
issues, why would you think it isn't so?

Are you starting to get the picture yet. Why are there all these claims
about dual CPU systems and nothing to confirm them with. The fact is their
slow and not worth the money spent for the desktop regardless of what is
said about them.

The picture I see is that dual processor machines bug you for some reason,
but you know little about them and have no references of your own to
support your claims, so you're trying to assume an air of faux scholarship
by demanding references from everyone else and hoping they don't notice
you're making shit up.

I can't help thinking your knowledge of multiprocessing comes from
Quake frame rates, and that you're missing the subtle point that it
depends on what you're trying to do.

 

[Steve Wolfe]

You're telling me that when you run X, and you have the X server trying

I'm telling you for the second time here to back up your statements.

I've made plenty of assertions, you've ignored every one. I've asked you
questions, you've refused to answer them.

Why did
you ignore the first request. Are you one of those people who thinks they
can make any claim they want and never have to support those claims.
Theories are nice, facts are better.

I've stated plenty of facts. You've ignored every one of them. You've
ignored the increased responsiveness of the APIC circuitry vs. the PIC
circuitry. You've ignored my facts about interrupt flooding. I have no
reason to believe that you are going to accept any other facts I've stated.
You just keep ignoring them.

I'll bet that you read Tom's Hardware a lot, don't you?

steve

 

[Steve Wolfe]

SMP takes some load of the foirst CPU and makes it possible for one CPU
to

And even more nonsense, but not even one website, datasheet, review,
anything to back up any of their claims.

There is absolutely nothing there that is nonsense. You make a lot of
noise about not backing up our statements, but you have not *once* given a
single bit of evidence to back up your position, you simply say "nonsense".

Does anyone else start to see a pattern here. The reason they can't provide
any proof is because it doesn't exist.

I'm seeing a pattern. You've never used an SMP desktop system, and you
ignore all of the facts we provide.

steve

 

[Lane Lewis]

If you're so adamant that others post evidence, it is high time for you
to do likewise. You keep making claims with *nothing* to back them up.
Prove your case (thereby disproving ours) by posting *your* evidence.

As to my own evidence that dual CPU's are faster than a single one, let
us take a look at kernel compile times on a dual 3.06GHz Xeon system.
Using only one of the two CPUs it took 2mins 42sec to compile my test
kernel. Using both CPUs, the compile time dropped to 1min 29sec.
That's about an 80% speedup on that task. Get it?

Sorry, but they made the assertion its their job to back it up. I'll
post any contradictory evidence as soon as they post the evidence to back up
their claims which were made first.

Do you really want to defend the nonsense quoted below. I think people
should take some kind of responsibility for what they post here. People come
here looking for honest advice to build a new system or upgrade or repair an
old one. Do you believe its OK to give false or misleading information. I
didn't request his advice by any means, he took it upon himself to educate
me on the wonders of dual CPU system with a bunch of nonsense that he in no
way could back up. Then instead of admitting his mistake he tried to cover
it up with even more nonsense. If you want to help your friend it might be
best to explain to him what he did that was wrong rather than jump on me.

[qt]
I have a dual Pentium 133 that is still *very* useable as a desktop, be
it under Linux or NT. A Pentium 233 would not be as usable.
[qt]
In fact, given the choice for a desktop system where I wasn't going to play
3D games, I'm much prefer a dual P3/600 over a 1.4 celeron.
[qt]
You can take a single-CPU system and flood the processor with
interrupts, and while the processer does very little actual work, the
machine will slow to a crawl.


Lane

 

[Steve Wolfe]

But since Opterons are around corner, I'll better wait for one decent

board

Still no benchmarks
Still no real world tests

I told you about real-world performance that I've encountered. You
ignored it. Why should I expect anything different?

Still no stats
Still no reviews

Just because it isn't on your favorite fan-boy benchmark site doesn't mean
that there aren't any reviews. When you said that nothing took advantage of
the second CPU, and I pointed out just how many processes are actually
trying to get CPU time, you ignored it.

Please, oh great one who knows everything. Expound to us the cost of
context-switching on x86 hardware, the effects of high levels of interrupts,
and the difference between PIC and APIC interrupt handling.

steve

 

[Lane Lewis]

You haven't done any backing up, either. All you've done is say "I was at
someone's site and saw some benchmarks, dood!"

We have experience with SMP systems, and you don't. You'll understand, of
course, if we find it hard to take you seriously.


Translation: "I'm too poor to buy an SMP system, I've never used one, and
I think I know what I'm talking about."

steve

Actually I've owned two.

And where are the stats to back up what you said?

Lane

 

[Lane Lewis]

snip

Why do you keep asking about benchmarks? Who said anything about
benchmarks? Dual processor machines are about twice as fast on Photoshop
filters than a single processor of the same type, those sorts of stats
get thrown around comp.sys.mac.advocacy now and then. But we're not
talking about number crunching speed.

Dual machines make good servers and
ok single app workstations but they are terrible desktop machines.

Everyone I've talked to (including people in real life) seem to think that
dual processor machines make wonderful desktop machines -- smoother and
more responsive. And faster. Certainly faster even for a single
application if it's multithreaded, but also just because it doesn't have
the operating systme interrupting it all the time.

Two processors versus a single processor of the same type, of course. Two
processors of half the clock speed as the single processor may or may not
be as fast depending on things like how well factored the code is and how
important it is to have twice the cache. It depends on what you're doing
with it.

I happen to have a machine sitting beside me that I can plug a second
processor into. Can't think of a reason not to.

Great even more claims and nothing to back them up.

[qt]
"Dual processor machines are about twice as fast on Photoshop filters
than a single processor of the same type."
[end]

It's something I saw from benchmark tests quoted on comp.sys.mac.advocacy.
They're always arguing about PowerPC versus x86 performance over there.
But I can't think of a reason to track down the reference for you. Feel
free to follow up on it yourself, though.

This one here is blatantly false, nothing runs twice as fast on a dual
machine due to overhead. At best you might see a 10 percent improvement and
a 1.2 gig single will smoke a dual 600 in loading and processing Photoshop
filters.

You just made that up. If there were *that* much overhead, nobody would
be making 4000 processor supercomputers! There's overhead when a single
processor switches from one task to another, too, you know.

I could believe a mere 10 percent improvement or less if you're running
only one, single-threaded task. And I think it was only recently that
Windows got decent multithreading, so it would be only recently that
Windows could take real advantage of more than one processor. But that
has nothing to do with overhead. Maybe for what you do (video games?)
there's little to be gained. But when you're doing more than one thing
at a time (e.g. background tasks, multiple users), or have a problem that
can be factored and multithreaded, there can be a substantial advantage
to having as many processors as you can get.

There's also the matter that if your data sets are so large or your cache
so small that you have to hit main memory all the time, you're going to be
slowed down by how fast you can push data through. If you're data-starved
at 850 MHz, then it won't do you any good to push the clock speed higher
than that, will it? You must know that, all else being equal, performance
just doesn't scale linearly with clock speed.

And all performance arguments aside, if everyone that uses multiprocessor
machines say they're smoother and more responsive in user interface
issues, why would you think it isn't so?

Are you starting to get the picture yet. Why are there all these claims
about dual CPU systems and nothing to confirm them with. The fact is their
slow and not worth the money spent for the desktop regardless of what is
said about them.

The picture I see is that dual processor machines bug you for some reason,
but you know little about them and have no references of your own to
support your claims, so you're trying to assume an air of faux scholarship
by demanding references from everyone else and hoping they don't notice
you're making shit up.

I can't help thinking your knowledge of multiprocessing comes from
Quake frame rates, and that you're missing the subtle point that it
depends on what you're trying to do.

All that and still nothing to back it up.

The problem with Photoshop is the software, though its smp capable you will
never see a big improvement in processing. Dual machines have no advantage
to a single machine that has an equal amount of processing power and with up
to 20 percent of overhead will show a definite disadvantage.

Lane

 

[John-Paul Stewart]

Sorry, but they made the assertion its their job to back it up.

*You* have also made assertions. It is your job to back up *those*
assertions. For example, in Message ID

At best you might see a 10 percent improvement

If you're going to be critical of people for not backing up their
assertions *you* *must* do likewise. Cite sources, post benchmark
figures. Otherwise you're just trolling.

 

[Lane Lewis]

Please follow your own advice, and tell me how many dual-CPU desktops you
use.

None, right?

steve

[qt]
I have a dual Pentium 133 that is still *very* useable as a desktop, be
it under Linux or NT. A Pentium 233 would not be as usable.
[end]

Do you really believe this nonsense.

Lane

 

[Lane Lewis]

The dodging was done on your part. You ignored every statement I made,
and refused to answer the question I posed.

steve

I'll answer any question you pose as soon as you back up this nonsense.

[qt]
I'm much prefer a dual P3/600 over a 1.4 celeron.

Lane

 

[Steve Wolfe]

Still no benchmarks

Start out by reading this, it'll give you some good background and some
good numbers:

http://www.ussg.iu.edu/hypermail/linux/kernel/0201.0/0810.html

Now, move on to some more....

http://people.freebsd.org/~fsmp/SMP/akgraph-a/graph1.html

Shows old SMP code in BSD, when used correctly, giving a near 100% speedup
on a dual-CPU system.

http://www.arstechnica.com/cpu/3q99/smp/smp-5.html

Shows speedups of 50% in PhotoShop and Visual C++.

http://www.xtremepccentral.com/articles/ps6-vs-ps7/

Photoshop speedups of between 10% to 120%.

Also, here's a test for you. Just for fun, I compiled the Linux kernel
four times on my P3-Xeon quad-Xeon machine. As someone who has a job,
compiling and interpretting are part of my normal desktop usage. Now this
is a machine with only 100 MHz of memory bandwidth, so it will run into
I/O-based limitations FAR sooner than more modern architectures. If
anything, that will make is scale more POORLY. Times?

1 process: 5m16.473s
2 processes: 2m29.469s

Now, in the end, as I pointed out, there's also something much more than
just what the elapsed time indicates. During a heavy load, a uniprocessor
system will slow down MUCH more than a multiprocesser system. There are
concepts there such as IRQ balancing and context switching that you simply
don't appear to understand. Even more, since you don't appear to have ever
USED an SMP system, you don't seem to have any experience, either.

Now, for the 133t fan-boys that sit at home with their modded cases
playing video games all day, a multi-processing system is going to be a
waste. You're right. For people who actually have things to get done
during the day, I have not yet met a single person *who has used a dual-CPU
machine* that doesn't greatly prefer it. If you haven't used one, go try it
out some time.

Here's a tip for you in life: Becoming a vehement, rabid dog over
something with which you have no experience is usually just going to make
you look like a fool.

steve

 

[Steve Wolfe]

Even the single processor on my home PC is inhibited by the relatively
slow memory speed on memory intensive applications, e.g. image editing, so I
imagine it would be much worse with dual processors.

I think you're probably misunderstanding. If you have an relatively
ancient PC with a 66 MHz memory subsystem, you can theoretically move 512
megabytes/second to the CPU. In real-world, most 66 MHz-bus systems can get
200 MB/second without breaking a sweat.

So, in telling me that memory is the limitting factor, are you *really*
asserting that your CPU can perform an image manipulation on 200
megabytes/second? If it's an insanely simple operation, I'd believe it.
For anything even remotely complex, I'd have a very hard time believing it.

In more recent architectures, where memory throughput is as high as three
*gigabytes* per second, I don't think that anyone's going to claim that
their processer could actually *perform* any non-trivial manipulation that
quickly.

steve

 

[Steve Wolfe]

All that and still nothing to back it up.

... and since you've provided nothing to back up YOUR position, then at
*best* you're just as unbelivable as us.

Wait! You've never used an SMP system, and we have. That means that at
*best*, your position is weaker than ours. So sorry, sucker.

steve

 

[Steve Wolfe]

I'll answer any question you pose as soon as you back up this nonsense.

No picking and choosing your battles. Put up or shut up.

steve

 

[Steve Wolfe]

I have a dual Pentium 133 that is still *very* useable as a desktop, be

it under Linux or NT. A Pentium 233 would not be as usable.
[end]

Do you really believe this nonsense.

I don't have to believe it, I have used both systems, and know it for a
fact. You have never used such systems, have you? You're an idiot.

You can either call me a liar, or admit that you're wrong.

steve

 

[Steve Wolfe]

I have a dual Pentium 133 that is still *very* useable as a desktop, be

it under Linux or NT. A Pentium 233 would not be as usable.
[end]

Do you really believe this nonsense.

Now that I have another minute, I'll expound.

Until a few months ago, when I gave it away, I had a Pentium 233 with 128
megs, and NT4 installed on it. I still have the 2xPentium 133 with 96 megs,
and NT4.

Yes, the dual Pentium 133 is more usable. Computationally-bound tasks are
slow, but you can continue working on other things while they run. With the
233, practically anything will bring the system to an unusable state until
it's finished.

Like I said, the real reason here isn't CPU cycles. It's a combination of
things that you show no understanding of, and conveniently ignore in every
post.

There, I've told you my experience. You can call me a liar, if you wish.
In the end, I've used a LOT of SMP systems first hand, and know their
strengths and weaknesses. You, on the other hand, have never used one in
your life. You'll forgive me if it's hard for your ideas to disprove my
direct, first-hand experience.

steve

 

[Steve Wolfe]

Still no benchmarks
Still no real world tests
Still no stats
Still no reviews

He asked you what benchmarks you wanted, and what tests you wanted. You
ignored that.

People offer to give you the proof you ask for, yet you won't even let
them know what you want. Wow. Talk about being closed-minded.

steve

 

[Steve Wolfe]

My experience with a home machine based on the SuperMicro P6DGH MB
with dual PIII/850's compared to my office machine with PIV 1.4GZ is the
same. The home system also uses a SCSI disk system, which was much
cheaper when using surplus disks when it was put together two years ago.

The P6DGH is an example of a motherboard built for Katmai CPUs but which
can, in some versions, run Coppermines. The differences aren't very

obvious.

Lane, are you starting to see a trend?

So far, every person who has *tried* a dual-CPU machine has agreed that
they make much better desktops. The only person who is claiming that they
don't is a person who has NOT tried one before.

Doesn't that sound a little fishy to you?

steve