What scope freq do I need?

[alpha_uma]

To look at the digital signals/waveforms of a modern CPU, say, f = 1GHz
Pentium to be specific, how "slow" can an oscilloscope be? Is it governed by
Nyquist criteria here? Thanks
Al-U

 

[Robert Baer]

To look at the digital signals/waveforms of a modern CPU, say, f = 1GHz
Pentium to be specific, how "slow" can an oscilloscope be? Is it governed by
Nyquist criteria here? Thanks
Al-U

Absolutely.
If the scope is to be used as a (poor man's?) logic analyzer, the
sample window should be less than the clock rate of the PC you want to
look at.
House-some-ever, if you want to see the rise, fall, and ringing of the
waveforms in the same PC, then the effective bandwidth should be at
least 10 times (raw minimum) that of the PC clock rate.
Mind you, once upon a time, there were *analog* scopes with this
capability...

 

[mike]

Absolutely.
If the scope is to be used as a (poor man's?) logic analyzer, the
sample window should be less than the clock rate of the PC you want to
look at.
House-some-ever, if you want to see the rise, fall, and ringing of the
waveforms in the same PC, then the effective bandwidth should be at
least 10 times (raw minimum) that of the PC clock rate.
Mind you, once upon a time, there were *analog* scopes with this
capability...

True enough...but...
A 1GHZ pentium may not have anything faster than
100MHZ on any pins that you can probe.

Probes and probing technique are VERY important.

What are you gonna do with the info when you get it?
You probably won't be able to find a "good" ground near where you want
it. You probably can't find out what the "exact" timing relationships
should be. Probing will disturb sensitive timing. Triggering will be
EXTREMELY important...you probably won't be able to trigger on what you
want to look at without a logic analyzer to trigger a digital scope
with pretrigger memory.

A scope to do any serious work will cost WAY more than you can afford...
unless you're Bill Gates.
mike

--
Return address is VALID.
Bunch of stuff For Sale and Wanted at the link below.
Compaq Aero floppy,ram,battery.
FT-212RH 2-meter 45W transceiver.
Toshiba & Compaq LiIon Batteries, Test Equipment
30pS pulser, Tektronix Concept Books, spot welding head...
http://www.geocities.com/SiliconValley/Monitor/4710/

 

[Robert Lacoste]

To look at the digital signals/waveforms of a modern CPU, say, f = 1GHz
Pentium to be specific, how "slow" can an oscilloscope be? Is it governed by
Nyquist criteria here? Thanks
Al-U

Except if you are a mother board designer you will very probably not need to
look at these high frequency signals... And I guess that even motherboard
designers don't... Firstly because all external buses are far slower (400MHz
max ?), then because probing such signals is a nightmare (just calculate how
many pF you can afford to have a low impact on 1GHz digital signals that
have bandwidths up to 10GHz...), and lastly because you will very probably
not be able to do anything useful with the scope's outputs...

So my guess is that a standard (say 200-500MHz) scope is working for 99,99%
of the applications (all I/O applications). If you need really to look at
high speed buses then don't buy a scope but a quick logic analyzer, and
especially buy VERY GOOD probes. By the way look at the recent announcements
from Tek or HP : they are all working at least as much on the probes than on
the analyzers...

Friendly,
Robert

 

[John Miles]

Except if you are a mother board designer you will very probably not need to
look at these high frequency signals... And I guess that even motherboard
designers don't... Firstly because all external buses are far slower (400MHz
max ?), then because probing such signals is a nightmare (just calculate how
many pF you can afford to have a low impact on 1GHz digital signals that
have bandwidths up to 10GHz...), and lastly because you will very probably
not be able to do anything useful with the scope's outputs...

Also, because the propagation of these signals is simulatable with an
insane degree of fidelity these days. Chip and motherboard
manufacturers spend millions developing and maintaining simulation
technology to prevent unpleasant surprises when the hardware comes back
from the fab.

There is not much you can do with any physical test equipment on a
modern PC.

-- jm

 

[alpha_uma]

if you want to see the rise, fall, and ringing of the
waveforms in the same PC, then the effective bandwidth should be at
least 10 times (raw minimum) that of the PC clock rate.

Thank you all for your kind explanations. The days of my hands-on EE
laboratory work with now-obsolete CPUs (8086-80486) are long gone. I'm just
trying to get an overall understanding of the relationships between the
system clock frequency, the speed of the front-side-bus and the frequencies
of the waveforms on the pins of a modern CPU. Please pardon me if some of
the following questions are stupid.

Questions:
(1) Mathematically speaking, how is the speed of the FSB derived/calculated
from the system clock frequency?
(2) What is the max frequency of the waveforms going in or coming out of the
CPU pins? How are these related to (1)?
(3) If (for some "stupid" reasons) I want to look at the timing of memory
read/write cycles on a PC (say, 1GHz P4, PC133MHz FSB), what scope speed do
I need in order to clearly discern edge triggering? Is it at least
10x133MHz? Or lower?

Thanks
Al-U

 

[mike]

Thank you all for your kind explanations. The days of my hands-on EE
laboratory work with now-obsolete CPUs (8086-80486) are long gone. I'm just
trying to get an overall understanding of the relationships between the
system clock frequency, the speed of the front-side-bus and the frequencies
of the waveforms on the pins of a modern CPU. Please pardon me if some of
the following questions are stupid.

Questions:
(1) Mathematically speaking, how is the speed of the FSB derived/calculated
from the system clock frequency?

You've already done it in question 3.

(2) What is the max frequency of the waveforms going in or coming out of the
CPU pins? How are these related to (1)?

You're probably more interested in the timing relationships among these
signals. Since we're working in the time domain, you probably want to
be able to easily determine 1nS time relationships. Your probes will
probably affect it more than that...unless you have EXPENSIVE probes and
know how to use 'em and you have a place to connect 'em. Hint: you
won't/can't.

(3) If (for some "stupid" reasons) I want to look at the timing of memory
read/write cycles on a PC (say, 1GHz P4, PC133MHz FSB), what scope speed do
I need in order to clearly discern edge triggering? Is it at least
10x133MHz? Or lower?

One cheap way to look at memory timing is in the memory data book.
The motherboard either meets it or it's broke...in which case you won't
have anything to look at.

Again, PC signals are not simple. Some pins have shared/multiplexed
functions. You have to be able to synchronize to the system in such a
way that you can get a repeatable trace. That's a HARD problem for
most interesting signals. 25 years ago, we were building emulators
to track the internal workings of processors so we could generate the
(externally unavailable) signals needed for triggering. It's gotten
MUCH worse over the last quarter century.

Even if you have a multi-GHz digital scope, you won't be able to make
much sense out of what you capture.

Get yourself the fastest quality scope you can afford. It won't be fast
enough to do what you want, but it might do what you need...wasn't there
a song about that???

TEK 465 is a good cheap scope. 7704A can be had for peanuts.
HP and others make some good stuff too.

One thing I did was to put a PLD on a small board with a bunch of
probes. I decide how I want to trigger, write the equations and program
the PLD as a state machine to sense that and generate the trigger.
It's only practical for HARD problems when you're out of options.
A general purpose system with that complexity would cost a fortune.


mike

Thanks
Al-U

--
Return address is VALID.
Bunch of stuff For Sale and Wanted at the link below.
Compaq Aero floppy,ram,battery.
FT-212RH 2-meter 45W transceiver.
Toshiba & Compaq LiIon Batteries, Test Equipment
30pS pulser, Tektronix Concept Books, spot welding head...
http://www.geocities.com/SiliconValley/Monitor/4710/

 

[alpha_uma]

<snip>
Probes and probing technique are VERY important.

What are you gonna do with the info when you get it?
You probably won't be able to find a "good" ground near where you want
it. You probably can't find out what the "exact" timing relationships
should be. Probing will disturb sensitive timing. Triggering will be
EXTREMELY important...you probably won't be able to trigger on what you
want to look at without a logic analyzer to trigger a digital scope
with pretrigger memory.

A scope to do any serious work will cost WAY more than you can afford...
unless you're Bill Gates.
mike

It is just curiosity. Once upon an ancient time, I was able to "see" the
memory read and write cycles on a 80386-based system on a decent scope in
real time. I'm just curious if I can still see them on a P3 or P4-based
motherboard without expensive equipment costing a great fortune (>$3,000.00,
say). Are EE students taught not to look at these waveforms in real time
anymore and to use something else instead to make the correct inferences?
Have logic analyzers evovled to such a state that scopes are not needed?
Al-U

 

[alpha_uma]

<snip>
Also, because the propagation of these signals is simulatable with an
insane degree of fidelity these days. Chip and motherboard
manufacturers spend millions developing and maintaining simulation
technology to prevent unpleasant surprises when the hardware comes back
from the fab.

There is not much you can do with any physical test equipment on a
modern PC.

From the answers I have gathered so far in this thread, would I be correct,
then, to think that few EEs (not to mention EE students) have seen these
waveforms in real time? However, I'm still curious as to whether there are
any EEs (working at, say Intel) who HAVE seen these in real time. Afterall,
Intel publishes the memory data books and timing diagrams. Are these
diagrams just results of simulation models, or are they measured from "real"
circuits?
Al-U