Motherboard Forums


Reply
Thread Tools Display Modes

What's the point of overclocking?

 
 
Alexi Tekhasski
Guest
Posts: n/a
 
      12-23-2006, 09:01 AM
"mike3" <(E-Mail Removed)> wrote in message news:(E-Mail Removed) ups.com...

> So then why do they sell them as X speed, anyway, if they are "exactly"
> the same?
> But if they are no different, why do they sell them as such?
>


As I see, you are not getting correct answers so far.

The thing is that all processors are different, every
individual processor, even if it is made on the same
wafer. The reason is in inherent statistical variations
in transistor quality across the wafer, and across every
chip. The variations are due to subtle differences in
mask alignment during multi-phase exposures, due to
kinetics of etching and film depositions (especially if
some layers like gate insulation are only several atoms
thick), and god knows due to what else.
As result, every block in a CPU performs
differently, some slower, some faster. To achieve
reliability in computations, processors are tested
at extreme specified conditions - lowest voltage
and highest specified case temperature, when all
transistors perform slower. The tests also cover
extreme data switching conditions, since there is
some internal crosstalk between signals that depends
on data content and sequences. Also, there are
certain industry-acceptable margins (10-15%) above
the guaranteed conditions. If a CPU fails that
suite of tests at certain frequency at that margin,
it gets binned down to a lower frequency, etc.

So, as you can see, every CPU should be
overclockable by 10-15% simply by definition
(of course, that margin definition strongly
depends on manufacturer's financial circumstances
and the level of competition :-)

Then, an overclocker usually uses extensive
cooling to keep Tcase way below the temperature
at which the particular CPU was tested. It gives an
additional (I would say 20-30%) margin in clock rate.
Or, when using extreme water or liquid gas cooling,
the margins are way better. This kind of chip
use does not do any electrical harm to the processor,
except that there might be mechanical fatigue problems
due to extensive thermal cycling.

More, overclockers usually try to rise supply
voltages. Elevated voltages lead to faster
switching gates, and it gives additional speed
margin. However, elevated voltages may cause
rapid deterioration of the chip, and may cause
gate oxide breakdown and other slower effects
from increased electromigration effects in metal
interconnects. As result, the chip might simply
stop functioning one day, or certain inner block
will produce a subtle error during calculations
you never tried. Again, there are built-in margins
into max Vcc based on certain life expectancy
of a processor, so a slight elevation should be
tolerable. However, effects of voltage are
typically exponential, so the safe overvoltage
margins are much lower.

In short, the point of overclocking is to narrow
operating conditions of a CPU and hope to
maintain those conditions all the time.

If a substantial voltage increase is involved, then
it is no different from gambling with limited
information and elevated risk.

I hope this answers your question.

Cheers,

- aap

 
Reply With Quote
 
 
 
 
mike3
Guest
Posts: n/a
 
      12-23-2006, 09:03 AM

Alexi Tekhasski wrote:
> "mike3" <(E-Mail Removed)> wrote in message news:(E-Mail Removed) ups.com...
>
> > So then why do they sell them as X speed, anyway, if they are "exactly"
> > the same?
> > But if they are no different, why do they sell them as such?
> >

>
> As I see, you are not getting correct answers so far.
>
> The thing is that all processors are different, every
> individual processor, even if it is made on the same
> wafer. The reason is in inherent statistical variations
> in transistor quality across the wafer, and across every
> chip. The variations are due to subtle differences in
> mask alignment during multi-phase exposures, due to
> kinetics of etching and film depositions (especially if
> some layers like gate insulation are only several atoms
> thick), and god knows due to what else.
> As result, every block in a CPU performs
> differently, some slower, some faster. To achieve
> reliability in computations, processors are tested
> at extreme specified conditions - lowest voltage
> and highest specified case temperature, when all
> transistors perform slower. The tests also cover
> extreme data switching conditions, since there is
> some internal crosstalk between signals that depends
> on data content and sequences. Also, there are
> certain industry-acceptable margins (10-15%) above
> the guaranteed conditions. If a CPU fails that
> suite of tests at certain frequency at that margin,
> it gets binned down to a lower frequency, etc.
>
> So, as you can see, every CPU should be
> overclockable by 10-15% simply by definition
> (of course, that margin definition strongly
> depends on manufacturer's financial circumstances
> and the level of competition :-)
>
> Then, an overclocker usually uses extensive
> cooling to keep Tcase way below the temperature
> at which the particular CPU was tested. It gives an
> additional (I would say 20-30%) margin in clock rate.
> Or, when using extreme water or liquid gas cooling,
> the margins are way better. This kind of chip
> use does not do any electrical harm to the processor,
> except that there might be mechanical fatigue problems
> due to extensive thermal cycling.
>
> More, overclockers usually try to rise supply
> voltages. Elevated voltages lead to faster
> switching gates, and it gives additional speed
> margin. However, elevated voltages may cause
> rapid deterioration of the chip, and may cause
> gate oxide breakdown and other slower effects
> from increased electromigration effects in metal
> interconnects. As result, the chip might simply
> stop functioning one day, or certain inner block
> will produce a subtle error during calculations
> you never tried. Again, there are built-in margins
> into max Vcc based on certain life expectancy
> of a processor, so a slight elevation should be
> tolerable. However, effects of voltage are
> typically exponential, so the safe overvoltage
> margins are much lower.
>
> In short, the point of overclocking is to narrow
> operating conditions of a CPU and hope to
> maintain those conditions all the time.
>
> If a substantial voltage increase is involved, then
> it is no different from gambling with limited
> information and elevated risk.
>
> I hope this answers your question.
>
> Cheers,
>
> - aap


So are you saying then, that overclocking chips by
a little bit (say 7%) is not as big a risk as I think?

And is there really a vast conspiracy to delude
people into thinking that some chips are built to run
faster than others?

I'm still curious about the life expectancy, as even
if the overclock suceeds without a hitch, it's the
long term effect that could come back and bite.
Would the 7% vanity overclock significantly
reduce the lifespan of the processor? I'm using
the 7% figure because it's in the "safe" margin
you give and because it probably wouldn't give
any useful gains, ie. just to get an idea of what
is going on. If you overclock, say, a 3 GHz
chip by 7% so it's running at 3.21 GHz, is that
really a good speed increase? Can you really
notice it? Ie. is it really WORTH it in the first
place??

 
Reply With Quote
 
 
 
 
John Whitworth
Guest
Posts: n/a
 
      12-23-2006, 09:59 AM

"mike3" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed) ps.com...
> If you overclock, say, a 3 GHz
> chip by 7% so it's running at 3.21 GHz, is that
> really a good speed increase? Can you really
> notice it? Ie. is it really WORTH it in the first
> place??
>


Put it this way...it's worth it enough for some people to pay a few hundred
pounds more for the official chip that clocks at that speed!

JW


 
Reply With Quote
 
Ed Medlin
Guest
Posts: n/a
 
      12-23-2006, 11:30 AM

"John Whitworth" <(E-Mail Removed)> wrote in message
news:458c75b7$0$31236$(E-Mail Removed)...
>
> "Ed Light" <(E-Mail Removed)> wrote in message
> news:mbZih.26406$(E-Mail Removed)...
>>
>> "Ed Medlin" <(E-Mail Removed)> wrote
>>
>>> With today's systems, you can adjust the processor frequencies without
>>> raising anything else that would cause any major issues.

>>
>> Don't forget the memory and the bus, such as hypertransport, go up with
>> the clock, and you can't raise the cpu multipliers on most cpus. If you
>> have value memory you'd hit a wall just a little way up if you didn't
>> change the ratios.

>
> But only if your motherboard didn't allow the FSB and memory clock to be
> adjusted independently. I can raise my CPU FSB and memory FSB completely
> independently.
>
> JW


That is the idea I was trying to get across.....maybe just not too
well.....:-) My present board is the same. I have to run my memory a bit
lower because at the time it was the fastest I could get my hands on, but
now things are a bit faster.
>
>



 
Reply With Quote
 
Ed Medlin
Guest
Posts: n/a
 
      12-23-2006, 11:42 AM

"Ed Light" <(E-Mail Removed)> wrote in message
news:a0%ih.23$(E-Mail Removed)...
>
> "John Whitworth" <(E-Mail Removed)> wrote in message
> news:458c75b7$0$31236$(E-Mail Removed)...
>>
>> "Ed Light" <(E-Mail Removed)> wrote in message
>> news:mbZih.26406$(E-Mail Removed)...
>>>
>>> "Ed Medlin" <(E-Mail Removed)> wrote
>>>
>>>> With today's systems, you can adjust the processor frequencies without
>>>> raising anything else that would cause any major issues.
>>>
>>> Don't forget the memory and the bus, such as hypertransport, go up with
>>> the clock, and you can't raise the cpu multipliers on most cpus. If you
>>> have value memory you'd hit a wall just a little way up if you didn't
>>> change the ratios.

>>
>> But only if your motherboard didn't allow the FSB and memory clock to be
>> adjusted independently. I can raise my CPU FSB and memory FSB completely
>> independently.
>>

>
> Wow, that's neat.
>
> I think it's pretty rare. Most boards change the memory ratio in relation
> to the fsb. So, to raise the fsb, you'd set the memory as a slower type.
> --
> Ed Light
>

My present board will allow me to set my memory bus slower as I raise the
processor frequency too. I think I have 3 choices that are dependent on the
processor speed or can be manually set. One of them always seem to be pretty
close to the stock speed, one higher and another lower. These ratios change
as I change the processor speed. I have 533mhz Samsung DDR2 that runs just
fine without any errors up to the 660-670 range for sure. I have never
pushed it further because my processor just won't go any more anyway. I plan
on a Core 2 Duo build soon and will have a go. It is an EM64T Prescott I-630
(3.0ghz) that runs at I-660 (3.6ghz) speeds. The I-660 was about double the
price of my processor a couple of years ago. Yea, it runs hot, but it runs
the hot at stock too. The same die is used on all of this series. My temps
are exactly the same at 3.6 as they are at 3.0. No increase in core voltage
is needed.

Ed

Ed


 
Reply With Quote
 
Ed Medlin
Guest
Posts: n/a
 
      12-23-2006, 12:12 PM
> Well, I don't want to do it, sorry. It obviously is risky, since you're
> pushing
> the thing over it's designed speed, but I suppose though that if done
> right
> the overclock could work, however I'm not that big a gambler! The
> more
> you overclock, the higher the risk. I don't think a few "vanity" hertz
> would
> do much anyway ("vanity" level = "safe" level.). Not enough to be
> *really*
> useful, and any more would be pushing the envelope too much. I doubt
> that pushing it up from the speed I've got now (2.8 GHz) to 3.0 GHz
> (+200
> MHz), what I call a "vanity" increase, would really do much, and
> anything
> more just doesn't feel safe to me. Like one said, it's best not to get
> greedy,
> and any "non-greedy" overclock does not seem like it would be worth all
>
> the fuss! If I want to play around with overclocking things, I'd better
> go
> and build myself a system that I wouldn't really care about if I lost
> it, ie.
> a pure "experiment bed".
>
> Remember, it takes only ONE mistake to nuke the chip, and if you don't
> watch the temp like a hawk...
>

It is a hobby really. With today's processors they have so much protection
from overheating that you really have to try hard to damage a processor. A
few years ago, AMD chips did not have thermal protection and would fry once
in awhile, but since then all processors have some sort of thermal
protection. Intels throttle back their speed and eventually will just shut
themselves down if temps get in the danger zone. 2.8-3.0ghz is not a major
overclock and should be fine and you should not have to mess with any core
voltage increases that can cause more heat. Whether you will notice the
difference or not is debatable.

Ed


 
Reply With Quote
 
Ed Light
Guest
Posts: n/a
 
      12-23-2006, 10:28 PM

"mike3" <(E-Mail Removed)> wrote

> Well, I don't want to do it, sorry. It obviously is risky, since you're
> pushing
> the thing over it's designed speed


Have you been reading the responses? The chips are identical except that
some need a tiny bit of extra volts to go top speed.

But -- I agree that you shouldn't overclock. Those who are leery can't dig
it.

Also, if you have a car without the mods to the engine of a higher model,
you should never add those mods. It wasn't designed for it, and might blow.
If you have a toothbrush that comes without attachments that a higher model
with the same base comes with, you shouldn't add the attachments. ;-) If a
small appliance comes with a power brick of a certain spec., you should
never use an off-brand of the same spec. Etc. etc.
Sorry! Couldn't help it! :-)
--
Ed Light

Bring the Troops Home:
http://bringthemhomenow.org
http://antiwar.com

Send spam to the FTC at
(E-Mail Removed)
Thanks, robots.


 
Reply With Quote
 
Ed Light
Guest
Posts: n/a
 
      12-23-2006, 10:42 PM

"mike3" <(E-Mail Removed)> wrote

> If you overclock, say, a 3 GHz
> chip by 7% so it's running at 3.21 GHz, is that
> really a good speed increase? Can you really
> notice it? Ie. is it really WORTH it in the first
> place??


Only in certain cicumstances. Let's say you're time-shifting tv and it tops
out your processor and drops frames and is a little jumpy. Now let's say a
small overclock is enough to cure it. Or, to make enough room to do some
other work while time-shifting in a small window.

Or you are rendering a movie, which loads up the cpu 100%. Whatever
overclock you do shortens the rendering time by that much. If it was taking
5 hours, then a modest overclock could take almost an hour off.

In my case, my 20% overclock often means my train simulator gets just out of
the intolerably jumpy zone. From 14.5 to 18 fps, when in a highly detailed
area.

But if you are just going to surf the internet and do word processing, then
overclocking is not necessary. A Duron 700 will do the trick.

Assuming that all chips can only overclock a little before going crucial is
not correct, though. 50% overclocks are common, with the chip still running
cool with stock cooling. Try to remember that it is the same chip as the
fast one, but requiring a small voltage increase. It is not getting near a
destructive temperature, and it has passed a long torture test. Example:
Sempron 64 2800+ overclocked from 1.6 to 2.4 with an increase of .05v. Never
goes over 45C. 60C is the maximum specified for those chips, from what I
know. The chip was about $60 with heatsink. Now, is that not attractive? If
not, I see you are overpowered by your safety precaution instincts. But
trying to convert us is pretty pointless, don't you think? We've had lots of
good overclocking experiences, of course going through some trepidations to
get there. Maybe you're hoping we'll convert you? It seems strange -- sorry.
--
Ed Light

Bring the Troops Home:
http://bringthemhomenow.org
http://antiwar.com

Send spam to the FTC at
(E-Mail Removed)
Thanks, robots.




 
Reply With Quote
 
Ed Light
Guest
Posts: n/a
 
      12-23-2006, 10:44 PM

"Ed Medlin" <(E-Mail Removed)> wrote

> My present board will allow me to set my memory bus slower as I raise the
> processor frequency too.

I'm not very expert at the underlying hardware operations; I've thought of
that as changing the ratio between the bus and the memory. Oh wait -- for
AMD 64, the memory operates right off the chip. Well, anyway.


--
Ed Light

Bring the Troops Home:
http://bringthemhomenow.org
http://antiwar.com

Send spam to the FTC at
(E-Mail Removed)
Thanks, robots.



 
Reply With Quote
 
Ed Light
Guest
Posts: n/a
 
      12-23-2006, 10:46 PM

"Ed Medlin" <(E-Mail Removed)> wrote
>>

> It is a hobby really. With today's processors they have so much protection
> from overheating that you really have to try hard to damage a processor. A
> few years ago, AMD chips did not have thermal protection and would fry
> once in awhile, but since then all processors have some sort of thermal
> protection.


For AMD it's on the motherboard and at least on some can be
enabled/disabled, with the crucial temp and/or a speed reductionpossibly
adjustable. Mine slows the chip down.


--
Ed Light

Bring the Troops Home:
http://bringthemhomenow.org
http://antiwar.com

Send spam to the FTC at
(E-Mail Removed)
Thanks, robots.


 
Reply With Quote
 
 
 
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are Off


Similar Threads
Thread Thread Starter Forum Replies Last Post
How to connect two Ethernet PHY devices on the same PCB point-to-point ? John Babcock Embedded 1 03-18-2007 03:30 PM
TMS320F2812 Floating point vs fixed point speed? cman Embedded 2 03-10-2007 04:19 AM
To assign ip addr to Point to Point interface. khan Embedded 1 05-12-2006 06:58 AM
floating point to fixed point conversion riya Embedded 7 02-24-2006 07:01 AM
point to point with satellite sknich Hardware 2 02-05-2004 04:44 AM


All times are GMT. The time now is 04:01 PM.


Welcome!
Welcome to Motherboard Point
 

Advertisment