PII vs PIII

[Gregory L. Hansen]

snip

The athlon tests also show the problems with a dual system. Comparing
the athlon dual 2200 with the athlon 3000 and you see somewhat mixed results
but mostly the dual 2200 is faster.

And the 3D rendering tests on Tom's web page shows the potential.

http://www6.tomshardware.com/cpu/20030422/opteron-23.html

To summarize, for a speedup factor defined as the time required for a
single processor to complete a task divided by the time for dual
processors we have

Xeon 3.06 Xeon 2.8 Opteron 1.8
Lightwave 7.5 1.7 1.6 1.6
Cinema 4D XL R8 2.3 2.3 1.97
3D Studio Max 5.1 1.9 1.9 1.7

My finger didn't slip in the Cinema results, that's a 2.3x speedup for the
dual Xeons! There's nothing "mixed" or "mostly" about these results, the
single processor machines got their asses handed to them.

Obviously it depends on what you're doing and which OS you're running, so
results may vary.

Heres the problem, if we compared an
athlon 3000 system with a dual 1500 system there just would be no comparison
of the two. The 3000 would win hands down, so whats the point of a dual
system if it loses in every benchmark ? If one of the advantages is running

I have a better idea. Let's compare an Athlon 3000 system with a dual
3000 system. You can use the dual 1500 system if you like, but as long as
we're comparing hypothetical systems mine is going to be tricked out.

 

[Steve Wolfe]

If you posted lots of facts, then post the 'best one' right

here and now. A single irrefutable fact. One simple
paragraph at the very top of your reply will do just fine
stating why dual processors are so superior and numbers that
demonstrate that superiority.

Because interrupts and context switches can make uniprocesser machines
extremely unresponsive, and the problem is much less severe with
dual-processer machines. That's pretty much the only point I've tried to
make.

Where is that best irrefutable fact and
supporting numbers? Simply state the summary of your claim and
provide supporting digits.

I haven't used a stopwatch, so I can't say "On my uniprocesser machine
under (whatever) load, it would take ten seconds for the "start" menu to
come up, or (x) seconds for a login terminal to come up."

However, from personal experience, I can say that the difference under
equivalent loads can literally be the difference between waiting 10-20
seconds from entering your password and the # prompt vs. waiting half a
second to get the prompt.

Everyone who's chimed in who has used a dual-processer machine has agreed
with me. If you and Lane (or are you the same person?) don't believe that,
then while I can certainly get out the stopwatch, I have a feeling that you
still won't believe it.

And again, as for berating me for not having the stopwatch numbers handy,
you and Lane are both being completely hypocritical, as neither of you have
produced a single benchmark. You're also being ignorant liers when you
state that I haven't shown any benchmarks, because I've given links to a
number of benchmarks.

steve

 

[Steve Wolfe]

It's been explained to me that games tend not to be written for multiple

processors because they tend to be written for home versions of Windows,
which doesn't support multiprocessing at all. So naturally you get little
advantage from the second processor, aside from handling interruptions
from the OS. But that has nothing to do with "overhead", rather with the
software not even trying to use the second processor. I suspect that's
the sort of computing that Lane Lewis is drawing on for his opinion, and
then just generalizing it to anyone doing anything on anything.

It's exactly the sort. When you look at the raw time to completion of
most home tasks, Lane is entirely correct - a fast single CPU machine will
indeed provide the fastest time. And if that's all that Lane cares about,
well, then he should certainly stick with that sort of machine.

For me, that's not how I work. I don't pop in a CD to burn, then go wait
around for five minutes for it to complete. And if I'm trying to move tens
of gigs around over the network, I really don't care to start the
transaction, then wait. I'd rather be doing something productive on the
machine.

Let's say that I want to start moving a few tens of gigs around the
network, burn a CD, *and* do something useful on my computer. With a
multiprocesser computer, I know that it will be responsive enough to do
that. With a single-processer computer, it simply won't be.

One of the marketers at work has to process web logs every day. We
generate gigs of logs every day, so between the disk I/O and the CPU cycles
it needed, the machine would be completely unusable. We're literally
talking about a thirty second wait between clicking on "start" and having
the menu pop up - and this was under Windows NT (not 9x), with more than
enough RAM that swap was never used.

I put in a dual-CPU motherboard for him with processers that were
*slightly* faster than the ones he had. The next day he came to me and said
"Geez, now when I process logs, my machine acts like nothing was out of the
ordinary."

That's the sort of thing that really makes dual CPU's work well for a
desktop. Without fail, every single person I've let try my dual CPU
machines has remarked that the machine was much more useful to them. *Every
single one*. W_tom and Lane can tell me I'm a junk scientist all they want,
but they're not going to convince me of the error of what I, and so many
other people have experienced first-hand.

But it seems multiple processors *could* speed up a
game if the software supports it.

There is a lot of room. John Carmack has stated that the new Doom will be
multi-threaded. However, because of limitations in Windows, it still might
not really take advantage of dual processers, we'll see how it goes. He's
talking about putting the AI, the sound, and various other sorts into their
own threads. We'll see how it goes.

steve

 

[Jeroen Geilman]

But it will get you bombarded with more and higher quality information
on Usenet than you could ever get by asking politely.

Hear, hear !

Bet you didn't expect all this about an old HP Kayak, did you now ?

 

[~misfit~]

Do you happen to know what the function or signal name for
that pin was? Rather interesting experiment.

Sory man, I don't. I got the info and some links from the guys at
alt.comp.hardware.overclocking and had them bookmarked. However, I've since
re-installed XP on this machine (new HDD) and no longer have them. On the
CPU there are two corners that have a pin missing, it was on one of those
corners, the middle one of the row of three. I broke the wrong one off
initially as I'm used to looking at socket 7 CPUs that only have one corner
with a pin missing, turned the CPU over, saw a corner like that and
proceeded to wiggle the pin 'til it broke of. *Then* I noticed the other
corner and checked and realised I'd removed the wrong one.

As I said earlier, I also had to link two pins together as part of the mod
and did this by soldering a jumper wire onto the back of the slocket. This
allowed me to use a coppermine CPU in a slocket that wasn't designed for it,
previously it was running a mendicino celeron 400Mhz (at 545Mhz, LOL) and
the coppermine wouldn't run in it.

The guys at a.c.h.o gave me all the info, signal names etc. but I don't
remember them. The motherboard was coppermine-compatible already, it was
just the slocket that was the problem.

As I said, I managed to solder the pin back on. However, that CPU is *never*
coming out of that slocket again, I don't know how well the soldering would
hold up after having the tightening arm on the ZIF socket moved again. I'm
just really pleased it worked, and is still working. I also upped the vcore
form 1.75v to 1.80v in the BIOS. It's a cC0 stepping CPU. I have another
coppermine celeron 600 that is a cB0 stepping (1.5v vcore) that won't do
900, even on 1.85vcore. The best I can get out of that one is 855MHz at 1.8v
and it's noticably slower. I didn't think it would be much slower but it
takes an extra 2 hours or more to do a SETI WU, 12+hours as opposed to 10
hours with the cC0 @ 900.

I have another coppermine celeron machine running here as well. It's a 900,
same mutiplier (9x) but designed to run on a 100MHz FSB. I can't overclock
that one at all, although the guy I got it off said he had it up to nearly
1.2GHz after increasing the vcore. He was using a later chip-set though
(815?) and running at 133 FSB, something the BX was never designed for. I'm
running them all on identical BX boards (MS-6163 Pro) and although the BIOS
has settings right up to 155Mhz FSB it won't even boot at 110 regardless of
increasing the vcore to nearly 2v.

The other two slockets *are* coppermine-compatible, I didn't have to modify
those.

 

[Gregory L. Hansen]

You would think that a Dual P2 450 or a dual celeron O/C 550 could keep
up with a single celeron O/C 900 but they couldn't. I always attributed it
to overhead of the OS but I think most of the smp programs were not capable
of taking full advantage of the dual CPUs. So anyway I parted out the dual
machines and have recommended single processors for the desktop ever since.

Depending on which version of Windows you were using, the SMP programs may
not have been taking any advantage of the dual CPUs at all, the OS
wouldn't support it! A friend of mine even had two processors, but he
never actually put them in a machine because he was using whatever the
current "home" version of Windows was which doesn't support
multiprocessing, therefore video games aren't typically written for
multiprocessing because it's assumed they'll run on home machines. So he
decided for the money he could get from selling those two processors and
the price saved from a dual motherboard, he'd get better performance from
a single, faster CPU. Even if you're running something like XP Pro, if
you're using software written for the "home" version you're not going to
get SMP out of it, anyway. That's not "overhead", that's crappy software.
But if that's the situation you're in, you have little to gain from a
second processor.

But that's not the situation everyone is in, so your advice just doesn't
generalize.

I hear-tell that the next version of Windows will have SMP support
all-around. See if dual machines get more popular after that.

 

[Lane Lewis]

snip

In terms of a single task, yes, like I've been saying, the 3000 would win.
I've never argued that at all. I've said since that sort of thing since the
first of the discussion, like when I pointed out that if I wanted to play a
3D game, I'd never choose the SMP system. Why do you keep beleagering the
point?

You most certainly did argue that. In both the cases below you never
mentioned the fact that most desktop apps would crawl on the dual systems as
compared to the faster single systems. Not to mention the fact that celeron
system would definately beat the dual 600 in smp apps. And not to mention
that the 233 would probably beat or equal the dual 133 in smp apps also. You
gave some the worst advice I've ever seen and then acted like a child when I
asked you to back it up.

[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.

Lane

 

[Lane Lewis]

snip

I have a better idea. Let's compare an Athlon 3000 system with a dual
3000 system. You can use the dual 1500 system if you like, but as long as
we're comparing hypothetical systems mine is going to be tricked out.
--

Nice try.
Here's what the discussion was about and they were not hypothetical.
233 vs. dual 133
1.4 gig Celeron vs. dual P3 600

The vast majority of desktop apps would crawl on the dual systems compared
to the single systems and some of the smp apps wouldn't do much better.

Lane

 

[kony]

I haven't used a stopwatch, so I can't say "On my uniprocesser machine
under (whatever) load, it would take ten seconds for the "start" menu to
come up, or (x) seconds for a login terminal to come up."

Your system had a severe problem which was unrelated to the number of
processors.

However, from personal experience, I can say that the difference under
equivalent loads can literally be the difference between waiting 10-20
seconds from entering your password and the # prompt vs. waiting half a
second to get the prompt.

Again, a severe problem having nothing to do with the number of CPUs.

Everyone who's chimed in who has used a dual-processer machine has agreed
with me. If you and Lane (or are you the same person?) don't believe that,
then while I can certainly get out the stopwatch, I have a feeling that you
still won't believe it.

Dual CPU systems CAN feel more responsive, but not anywhere near what
you're mentioning above, and this responsiveness is typically only
worthwhile when running a high-priority app and trying to do something
else demanding, simultaneously... not just mucking around in the GUI.

And again, as for berating me for not having the stopwatch numbers handy,
you and Lane are both being completely hypocritical, as neither of you have
produced a single benchmark. You're also being ignorant liers when you
state that I haven't shown any benchmarks, because I've given links to a
number of benchmarks.

Ah, time and time again people want links. Today we have the best
search engine ever known to mankind... bet you a nickel it'll find
benchmarks.


Dave

 

[Lane Lewis]

snip

Who said we're comparing dual CPUs to a single CPU of twice the clock
rate? Go back to the start of this thread. The OP has an HP Kayak
system that'll take CPUs upto 600MHz, and either 1 or 2 of them. His
choice is limited to single 600 vs. dual 600. Dual wins. Whether or
not a single 1200MHz CPU would be better is irrelevant---he'd have to
replace the system to get that and that's not what he was asking about!

BTW, I'm running dual 3.06GHz Xeons. Just what single CPU system is
going to be faster? A single 6.12GHz Xeon???

One of my favourite vendors currently list AthlonXPs upto "3200+" rating
and AthlonMPs upto "2800+". Now I ask you, which is faster: dual 2800
or single 3200? Clearly the dualie wins. Frankly the perfomance of a
dual 1600 has little to no bearing on the discussion at hand.

The following quotes are what the discussion was about. Note that this is
the second time I've told you, go check the thread.

[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.

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.
[end]

Those two quotes are what I objected to. They are of course complete
nonsense as I stated before which got this whole thing started.

If your going to get involved at least have the courtesy to follow the
thread closely.

Note that as far as your system goes it does make sense to buy a dual
system, it does not follow that you should upgrade a dual 600 system. With
the price of a 1.2 gig or faster CPUs going for so little why invest in a
slow outdated dual board.

Lane

 

[kony]

IIRC, Intel moved the RESET pin in the move to Coppermine CPUs, this
is likely the one.


Dave

 

[SIOL]

It's exactly the sort. When you look at the raw time to completion of
most home tasks, Lane is entirely correct - a fast single CPU machine will
indeed provide the fastest time. And if that's all that Lane cares about,
well, then he should certainly stick with that sort of machine.

For me, that's not how I work. I don't pop in a CD to burn, then go wait
around for five minutes for it to complete. And if I'm trying to move tens
of gigs around over the network, I really don't care to start the
transaction, then wait. I'd rather be doing something productive on the
machine.

Let's say that I want to start moving a few tens of gigs around the
network, burn a CD, *and* do something useful on my computer. With a
multiprocesser computer, I know that it will be responsive enough to do
that. With a single-processer computer, it simply won't be.

Bingo ! That is what I was talking about. When i do some compiling on my
Gentoo machine, I really couldn't care less if the kernel gets compiled in
two or five minutes, just as long as I can continue to use my machine for
other purposes in the meantime. And dual P3 is perfectly capable for that
sort of jobs, despite small bandwidth of P3 bus, let alone Athlons, Xeons or
Opterons.

Same goes for other real-time tasks, like video watching/editing, music
playing etc

Branko

 

[Philip Armstrong]

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?

Given that none of these have been double blind tests it sounds
fishily like the Dual CPU types suffering delusions of value to me

(It cost me more, so it *must* be better!)

Phil (cynic at large)

 

[Gregory L. Hansen]

snip

Nice try.
Here's what the discussion was about and they were not hypothetical.
233 vs. dual 133
1.4 gig Celeron vs. dual P3 600

The vast majority of desktop apps would crawl on the dual systems compared
to the single systems and some of the smp apps wouldn't do much better.

You seem immune to the point that was made in the original claim that the
dual 133 was never said to be faster or as fast as the 233, rather that
it runs smoother and with fewer interruptions when you're trying to do
more than one thing at the same time.

The discussion was also about

"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 that's what I was addressing above.

 

[John-Paul Stewart]

The following quotes are what the discussion was about. Note that this is
the second time I've told you, go check the thread.

[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.

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.
[end]

Those two quotes are what I objected to. They are of course complete
nonsense as I stated before which got this whole thing started.

I'd agree with both statements, personally. I'd prefer a dual 600 to a
single 1.4GHz celeron. I like to be able to do number crunching in the
background while still leaving a CPU free for interactive applications.
But that's based on my personal preferences and work habits. AFAICT the
person who originally made the statements was also expressing personal
preference. If it doesn't suit you, say so. But don't make sweeping
generalizations in an attempt to negate the first person's personal
preferences.

If your going to get involved at least have the courtesy to follow the
thread closely.

I have been following the thread closely. The first thing to which I

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

That's a very broad blanket statement which has been proven false.
There most certainly are things that show much more than a 10%
improvement on a dual CPU system. That statement may well be true in
some circumstances but you've stated it as universally true when it
isn't. If you wish to qualify it and limit its scope, do so!

 

[Steve Wolfe]

I haven't used a stopwatch, so I can't say "On my uniprocesser machine

Your system had a severe problem which was unrelated to the number of
processors.

How do you know?

Again, a severe problem having nothing to do with the number of CPUs.

How do you know?

If "heavy load" to you means "a problem", I suppose.

Dual CPU systems CAN feel more responsive, but not anywhere near what
you're mentioning above, and this responsiveness is typically only
worthwhile when running a high-priority app and trying to do something
else demanding, simultaneously... not just mucking around in the GUI.

Then you're saying I'm a liar?

Ah, time and time again people want links. Today we have the best
search engine ever known to mankind... bet you a nickel it'll find
benchmarks.

I'll bet you TWO nickels it already did.

steve

 

[Steve Wolfe]

Let's say that I want to start moving a few tens of gigs around the

Bingo ! That is what I was talking about. When i do some compiling on my
Gentoo machine, I really couldn't care less if the kernel gets compiled in
two or five minutes, just as long as I can continue to use my machine for
other purposes in the meantime. And dual P3 is perfectly capable for that
sort of jobs, despite small bandwidth of P3 bus, let alone Athlons, Xeons or
Opterons.

Yeah, but I'm obviously a "junk scientist" because I don't have numbers.
Or if the machine is more responsive, it's all psychological because I paid
more for the machine. Or, it's because there was something WRONG with my
machine, and the number of CPU's doesn't matter.

It's amazing how many different things they'll come up with to try and
tell me that the things I've seen first-hand for years aren't true. ; )

Same goes for other real-time tasks, like video watching/editing, music
playing etc

If they had even half of a brain and had been watching the linux kernel
mailing list, they're realize that improving the responsiveness and
interactivity of a system is one of the highest priorities in the 2.6 kernel
series. But, to them, it doesn't matter, it's all psychological,
junk-science, or just nonsense. : )

steve

 

[Steve Wolfe]

[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.

For the last time, Lane, "usable" does not mean "will finish first".

If you haven't figured that out already, you're a blind idiot. If you
HAVE figured it out already, you're a weak-minded, dishonest idiot for
pretending not to.

steve

 

[Steve Wolfe]

You most certainly did argue that. In both the cases below you never

mentioned the fact that most desktop apps would crawl on the dual systems as
compared to the faster single systems. Not to mention the fact that celeron
system would definately beat the dual 600 in smp apps. And not to mention
that the 233 would probably beat or equal the dual 133 in smp apps also. You
gave some the worst advice I've ever seen and then acted like a child when I
asked you to back it up.

Again, you're still stuck in your world where you start a single
application, click a stopwatch, and wait for it to be done. You've entirely
missed (or possibly ignored) the points I've been making.

[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.

I know you think those are nonsense. But you know what? I've used
those systems, and unless I'm on a psychotropic or hallucinogenic drug, my
expeience with them was more than enough to back it up to me. However, you
refuse to believe it, presumably, because you haven't seen it and I didn't
time it.

steve

 

[kony]

How do you know?

Because there is no reason for a 10 second delay. It wasn't due to
having a single CPU unless the task that was running was improperly
assigned a high or time-critical priority. Tasks that are actually
time-critical, like video capture or CDRW/DVD burning, don't even come
near the delay you mention. If that specific software is THAT bad,
you need either reassign it's priority, or replace it if possible.

How do you know?

If "heavy load" to you means "a problem", I suppose.

These things you've been mentioning, acting like they cause delays, do
NOT cause delays on a normally-working single-CPU system.

On a single-CPU system I can transfer Gigs of files over a lan, while
processing a video capture, capturing ANOTHER video, listening to an
MP3, and a dozen other background tasks. The single CPU system
(actually I have multiple systems set up for this video capture and
editing work) runs at 100% full load for hours at a time. It isn't
some "test", I do similar activities multiple times per week, for
hours. NEVER is there even the slightest delay, let alone seconds.
Granted this requires at least 3 HDDs to do capture AND
post-processing simultaneously, but NOT 2 CPUs.

The only time having a single CPU is a problem is when there's another
semi-time-critical application running, like a modern game. Depending
on the process priorties either the game or the video work loses
performance... if the game framerate drops too low it IS a problem,

and it would be very beneficial to use 2 CPUs,

but still there'd be a FSB and memory bottleneck, it would be more
effective to use a 2nd computer, and considering the extra expense of
the second CPU and dually motherboard in a 2 CPU system, a second
system isn't that much of an additional expense.

Then you're saying I'm a liar?

I'm giving my version of the " truth". If it conflicts with your
version, so be it. If I were going to call you a liar I would just
come right out and say it.

I think it's most likely that your system usage, probably the
applications, are screwed up, don't work properly, so they totally
consume the entire CPU. Another possibility is HDD problems, or
chipset/PCI bus issues. Don't even begin to think you're the only
person doing more than one demanding task simultaneously, many people
do but never have a 10 second delay for something as simple as the
start menu no matter how hard they try to create one.

I'll bet you TWO nickels it already did.

Yep, but then why ask for links?


Dave