Is it possible that Overclocked puters will return more errors that non Overclocked puters?

---

short answer: yes

---

Yes they can, I have a P4 3.06 that simply won't run any of the Projects with out returning a lot of computation errors, so I just run it at stock speeds.

But on the other hand I also have a P4 3.4 thats running at 3.850 Ghz that hardly ever returns any computation errors or invaild WU's, if fact it's the leading RAC Computer over at the LHC site so it can't be returning very many invaild WU's IMO...

Thanks Poorboy

O.K. then why does it happend?
I would like to put on line a AMD 1500 but it is slow and might want to overclock it.

---

> Thanks Poorboy
>
> O.K. then why does it happend?
> I would like to put on line a AMD 1500 but it is slow and might want to
> overclock it.

An AthlonXP 1500+ too slow ?

Well, there are alot of Users crunching along with equipment alot slower than that; so that machine should do just fine :)

---

If you overclock your computer and participate in distributed computing projects, you should at the least perform a stability test of some sort on your machine.

I prefer running PRIME95's torture test(not sure which mode at the moment) for 24hrs, with a priority setting of 10. If you get an error, your computer is not stable and performing imperfect calculations that you would not want to return to a project.

For further information on stable overclocking, Visit Overclockers.com's forums at http://www.ocforums.com/
Feel free also to visit our team's SETI/BOINC section of the forums as well.

---

BOINC WIKI
Overclockers.com's Forum

> Is it possible that Overclocked puters will return more errors that non
> Overclocked puters?
>
I have a PII 350@392, and i haven't return any bad wu (that i've realized).

---

<img src="http://boinc.mundayweb.com/one/stats.php?userID=1331">

> Is it possible that Overclocked puters will return more errors that non
> Overclocked puters?

Hopefully the image will come through, and this will be more than just words:



Time runs left-to-right, voltage is up-and-down.

The black line shows a simplified waveform of a signal transitioning from "0" to "1" -- the exact voltage for "1" might be 1.8v or 3.3v or 5v, just depends on where we have our probe.

There is actually a little bit of a ripple at the top after the voltage reaches the "1" state, and the vertical part isn't as straight as I've shown.

The green vertical line represents normal clocking. This is when the next bit of logic looks to see if the signal is 1 or 0. The distance between the "corner" at the top and the green line is called "margin" -- just a little extra time to make sure the signal is stable.

When you overclock, you are moving the green line to the left toward the edge of the signal.

The yellow line is a system that probably runs mostly reliably. The sample is more than half-way up the slope, so it'd __probably__ be seen as a "1".

The red line is what happens when you overclock too much. This system probably won't even run, because the signal that should be a "1" will most likely be read as zero.

Why have margin? The slope of the line will change a little with variations in voltage and temperature. This is why you can often super-cool a CPU and get it to run well above the clock rate -- the sloped line is much more vertical (much faster) if the CPU is down around 0 degrees C.

So, basically, if you know what you're doing and test carefully, you can squeeze a little extra speed out of a machine without inducing errors, but you're trading speed for margin, and you can definitely "fall off" the signal.

---

push the pc slowly and do benchmarking of the pc if it successfully boot up. And do more intensive benchmarking of the pc system once you are happy with that value you have set.

The Ideal you are looking for is difference from pc to pc. What type of cpu, motherboard, ram, graphic card, psu, heatsink, fans in the casing and mod you done to your pc will influence this Ideal line.

---

Very well put. A very informative post, indeed.



> > Is it possible that Overclocked puters will return more errors that non
> > Overclocked puters?
>
> Hopefully the image will come through, and this will be more than just words:
>
>
>
> Time runs left-to-right, voltage is up-and-down.
>
> The black line shows a simplified waveform of a signal transitioning from "0"
> to "1" -- the exact voltage for "1" might be 1.8v or 3.3v or 5v, just depends
> on where we have our probe.
>
> There is actually a little bit of a ripple at the top after the voltage
> reaches the "1" state, and the vertical part isn't as straight as I've shown.
>
> The green vertical line represents normal clocking. This is when the next bit
> of logic looks to see if the signal is 1 or 0. The distance between the
> "corner" at the top and the green line is called "margin" -- just a little
> extra time to make sure the signal is stable.
>
> When you overclock, you are moving the green line to the left toward the edge
> of the signal.
>
> The yellow line is a system that probably runs mostly reliably. The sample is
> more than half-way up the slope, so it'd __probably__ be seen as a "1".
>
> The red line is what happens when you overclock too much. This system
> probably won't even run, because the signal that should be a "1" will most
> likely be read as zero.
>
> Why have margin? The slope of the line will change a little with variations
> in voltage and temperature. This is why you can often super-cool a CPU and
> get it to run well above the clock rate -- the sloped line is much more
> vertical (much faster) if the CPU is down around 0 degrees C.
>
> So, basically, if you know what you're doing and test carefully, you can
> squeeze a little extra speed out of a machine without inducing errors, but
> you're trading speed for margin, and you can definitely "fall off" the signal.
>

---

> > Hopefully the image will come through, and this will be more than just
> words:
>
> Thanks for the post.
>
> So would 'ideal' then be to overclock so that the green line is shifted to
> where the black line just changes to horizontal (i.e. the corner)?

If you're going for maximum speed, yes, but reliability suffers.

The trouble is that the slope of the line can move with several factors, like temperature. There is always a little bit of "jitter" that just randomly bounces back and forth.

So, in an ideal world, you figure out the worst case position of the corner, and then go just a little bit to the right of that just to make sure.

... and believe me, Intel, AMD and everybody else wants to be as close to the edge as possible (because they get more money at higher clock rates) but not so close that it ruins their reputation.

If you want to overclock, you should keep increasing the clock rate until the system becomes unreliable, then run at 95% of the fastest clock rate that appeared stable.

---

> Also interesting. Actually, my technique has always been to overclock to where
> it crashes and then drop down by 5 Mhz, 'til I find 'stability'. Not very
> sophisticated, and not getting the last clock cycle that I can, but once I
> find a point; very stable.

5 MHz isn't bad when we're talking about a 500 MHz clock, but when we're talking about 3 or 4 GHz it's probably not quite enough. I'd want at least 200 MHz of margin on a 4000 MHz (4 GHz) machine -- that's 5%.

Then again, my emphasis is stability.

---

> The red line is what happens when you overclock too much. This system
> probably won't even run, because the signal that should be a "1" will most
> likely be read as zero.

Depends on which part of the CPU is being pushed the hardest. It is a known fact that the FPU (where many distributed projects do their calculations) will fail long before the rest of the CPU so you might never see a BSOD or have lockups but your FPU will be returning invalid numbers while crunching. My view on overclocking: I play it pretty safe. If you spend 5 days fine tuing the system and making sure it runs right with prime95 or whatever, it will take you several weeks to make up the lost crunching time you lost in testing. I would rather not waste my time like that :) At the moment I'm not overclocking at all. I used to have my Athlon Tbird 1.4 running at 1.6 but sadly that chip (and the motherboard it was on) have passed into transistor heaven.

---

A member of The Knights Who Say NI!
For rankings, history graphs and more, check out:
My BOINC stats site

If you spend 5 days fine tuing
> the system and making sure it runs right with prime95 or whatever, it will
> take you several weeks to make up the lost crunching time you lost in testing.
> I would rather not waste my time like that :)

It's not all about just speeding it up, there's some enjoyment in the process : )
Plus, as most ppl have their pc on for more than a few weeks straight, even if you lose a few days, the gain over 1+ years would make it worth it.

Plus, sometimes you can get significant overclocks.... I have the fairly standard 2500+ @ 3200+ overclock which represents a 20% overclock on raw ghz... so if it takes 5 days to tweak (usually only 1-3), after 25 days you're crunching 20% more WU (depending on the mhz scaling).

However, I see all these ppl going "I got this overclock, for 20 mins before it crashed...." absolutely pointless, unless it's prime95 stable for at least 24hr and you're not returning invalid results in a distributed project, it's not a _real_ overclock.

> I would like to put on line a AMD 1500 but it is slow and might want to
> overclock it.

LOL -- You're kidding, right? I've got a dual P-II 300 and a P-II 366 here that I might be inclined to call slow. None of my systems match an AMD 1500 for performance, as sad as that may seem to many. :-)

trane

---

> There is actually a little bit of a ripple at the top after the voltage
> reaches the "1" state, and the vertical part isn't as straight as I've shown.

The "ripple" occurs at both sides of the transitions. It is called "over-shoot" and "ringing". As the voltage goes up )or down) it cannot stop when it reaches the level and so will pop-up a little in the direction of the voltage movement. Then the voltage swings back and forth at the new level until it settles down. With tight enough timing, the voltage will not settle before you hit it with the down transition ...

The other sources of problems with over-clocking have to do with the fact that the speed of light is one of the issues we are dealing with. And at the frequencies we are working with, this means that the distance across the surface of the CPU chip becomes significant.

Ned,

Can I talk you into e-mailing me a copy of that diagram? and the explanation ... I would like to use it for the FAQ and Glossary ... p.d.buck@comcast.net ... Thanks!

---

> The "ripple" occurs at both sides of the transitions. It is called
> "over-shoot" and "ringing". As the voltage goes up )or down) it cannot stop
> when it reaches the level and so will pop-up a little in the direction of the
> voltage movement. Then the voltage swings back and forth at the new level
> until it settles down. With tight enough timing, the voltage will not settle
> before you hit it with the down transition ...
>
> The other sources of problems with over-clocking have to do with the fact that
> the speed of light is one of the issues we are dealing with. And at the
> frequencies we are working with, this means that the distance across the
> surface of the CPU chip becomes significant.

The speed of light is one of the factors that causes the delay....

Admiral Grace Hopper used to hand out 11.7" pieces of wire, saying "here, have a nanosecond."

I was really looking hard for a better diagram, one that showed how the rise started, the overshoot and damped-sine-wave "ring" at the top, and sadly couldn't find anything that I could refer to or scan.

You're certainly welcome to the drawing and explanation, I'll E-Mail shortly.

---