Changing 'default' FSB speed with Tualatin?

David Maynard wrote:

snip, I've been following this interestedly, thanks guys

It does illustrate the significance of too high a Vcore on the
regulator FET. Using 1.7 vs your 1.4 raises the FET temp an estimated
30C.

So, David, how would someone like me g go about unsoldering FETs to
replace them? There's a large surface stuck to the board. And, if I did
manage to get them unstuck would I be better leaving the replacements
standing up with heatsinks on them?

Cheers,
--
~misfit~

~misfit~ wrote:

David Maynard wrote:

snip, I've been following this interestedly, thanks guys

It does illustrate the significance of too high a Vcore on the
regulator FET. Using 1.7 vs your 1.4 raises the FET temp an estimated
30C.


So, David, how would someone like me g go about unsoldering FETs to
replace them? There's a large surface stuck to the board.

Well, they're bad to begin with so damaging the FET itself isn't a concern.
I've used two methods. One is just the plain old use enough heat to get it
loose. For ones where surrounding devices present a problem I 'cut' away
the ceramic body till I can get on the device mounting plate to heat it.

And, if I did
manage to get them unstuck would I be better leaving the replacements
standing up with heatsinks on them?

Kinda problematic as that is the drain connection.

I just use a FET with lower RDSon and solder it back down. I did try
soldering a home made copper heatsink onto one on a BH6 but I ended up
removing it before I realized it wasn't the problem.

Speaking of BH6, I have a real head scratcher on another one pulled from a
dead system. FET had blown, which was pretty obvious from the dead short on
the power rails that went away when I cut the leads off. Just put a new one
on and Vcore is perfect, and I can even see, with my multimeter, it change
to the new proper Vcore when I change processors, but nothing boots and I
can't see anything 'obvious' (like a burned traced, etc).

OOOOOOOOHHhhhh wait, holy toledo, it just dawned on me, this was the one
that needed a 'wedge' to cock the slot-1 card. I'll have to try that again
because I had just loosely tossed some spare parts into it for a quick test
but, by golly, that sounds hopeful.


Cheers,
--
~misfit~

David Maynard wrote:
~misfit~ wrote:

David Maynard wrote:

snip, I've been following this interestedly, thanks guys

It does illustrate the significance of too high a Vcore on the
regulator FET. Using 1.7 vs your 1.4 raises the FET temp an
estimated 30C.


So, David, how would someone like me g go about unsoldering FETs to
replace them? There's a large surface stuck to the board.

Well, they're bad to begin with so damaging the FET itself isn't a
concern. I've used two methods. One is just the plain old use enough
heat to get it loose. For ones where surrounding devices present a
problem I 'cut' away the ceramic body till I can get on the device
mounting plate to heat it.

And, if I did
manage to get them unstuck would I be better leaving the replacements
standing up with heatsinks on them?

Kinda problematic as that is the drain connection.

I just use a FET with lower RDSon and solder it back down. I did try
soldering a home made copper heatsink onto one on a BH6 but I ended up
removing it before I realized it wasn't the problem.

Thanks for the info.

Speaking of BH6, I have a real head scratcher on another one pulled
from a dead system. FET had blown, which was pretty obvious from the
dead short on the power rails that went away when I cut the leads
off. Just put a new one on and Vcore is perfect, and I can even see,
with my multimeter, it change to the new proper Vcore when I change
processors, but nothing boots and I can't see anything 'obvious'
(like a burned traced, etc).

OOOOOOOOHHhhhh wait, holy toledo, it just dawned on me, this was the
one that needed a 'wedge' to cock the slot-1 card. I'll have to try
that again because I had just loosely tossed some spare parts into it
for a quick test but, by golly, that sounds hopeful.

I remember you talking about that board when I had some success cleaning
slots.

Good luck,
--
~misfit~

~misfit~ wrote:

David Maynard wrote:

~misfit~ wrote:


David Maynard wrote:

snip, I've been following this interestedly, thanks guys

It does illustrate the significance of too high a Vcore on the
regulator FET. Using 1.7 vs your 1.4 raises the FET temp an
estimated 30C.


So, David, how would someone like me g go about unsoldering FETs to
replace them? There's a large surface stuck to the board.

Well, they're bad to begin with so damaging the FET itself isn't a
concern. I've used two methods. One is just the plain old use enough
heat to get it loose. For ones where surrounding devices present a
problem I 'cut' away the ceramic body till I can get on the device
mounting plate to heat it.


And, if I did
manage to get them unstuck would I be better leaving the replacements
standing up with heatsinks on them?

Kinda problematic as that is the drain connection.

I just use a FET with lower RDSon and solder it back down. I did try
soldering a home made copper heatsink onto one on a BH6 but I ended up
removing it before I realized it wasn't the problem.


Thanks for the info.

The first BH6, that I tried a heatsink on, was a bit 'distressed' and acted
odd after I repaired it so I removed the heatsink thinking that was
creating a problem. You'd have to have seen it to see why I wondered;
definitely Rube Goldberg. hehe.

I pounded a thick copper ground wire into a thin sheet for the heatsink.
Not all that bad, really, and I learned about cold flow hardening first
hand so if I ever get tossed back into Ancient Greece by some rampaging
time machine I'm all set on the technique for pounding copper into nifty
swords.


Speaking of BH6, I have a real head scratcher on another one pulled
from a dead system. FET had blown, which was pretty obvious from the
dead short on the power rails that went away when I cut the leads
off. Just put a new one on and Vcore is perfect, and I can even see,
with my multimeter, it change to the new proper Vcore when I change
processors, but nothing boots and I can't see anything 'obvious'
(like a burned traced, etc).

OOOOOOOOHHhhhh wait, holy toledo, it just dawned on me, this was the
one that needed a 'wedge' to cock the slot-1 card. I'll have to try
that again because I had just loosely tossed some spare parts into it
for a quick test but, by golly, that sounds hopeful.


I remember you talking about that board when I had some success cleaning
slots.

Yeah. After I posted I did a quick visual into the slot hoping to see
'something', like maybe dirt or a bent pin, but there wasn't anything
obvious. Not that that eliminates the 'cock' issue because that was
definitely the one that needed it.

It'll have to wait though because I'm all agog doing a DSL Linux re-master
right now. Did one with no changes, to test the make ISO procedure, and
that worked so now I'm working on the changes part.


Good luck,
--
~misfit~

On Wed, 28 Jul 2004 23:14:28 -0400, P2B wrote:

I taped a temperature sensor on top one of the FETs, and it peaked at
95C after running a Tualatin 1.0A @ 1.5Ghz on 1.4Vcore full bore for 10
minutes, then eventually dropped to 65C at idle.

Quite high if you measured them on plastic part
/than means that are not enough strong!/
(their metal base soldered on MoBo must be heating 20° more!

I did not tested the temp on them before puting a HS there, but @ 20°
amb. there on a HS was 45° idle (55° full load); now after mounting
few days ago a small fan there running on 5V sucking air, there are
same temps but with amb.temp 5° higher with my setup 1.0A@1,35
1,47Vcore .../the HS is just glued there on plastic top w/ some silver
paste too../
--
Regards, SPAJKY ®
& visit my site @ http://www.spajky.vze.com
"Tualatin OC-ed / BX-Slot1 / inaudible setup!"
E-mail AntiSpam: remove ##

David Maynard wrote:

P2B wrote:

David Maynard wrote:

P2B wrote:

David Maynard wrote:

P2B wrote:

David Maynard wrote:

P2B wrote:

David Maynard wrote:

P2B wrote:


It really depends on the board - my P2B systems have run 7x24
for over 2 years with no sign of problems. The FETs and caps
don't even get hot (no extra cooling or sinks), just don't
touch the clock chip at 150Mhz FSB!

Keep in mind the Tualatin only draws slightly more power than
a Katmai P3 600 - so if the board was *properly* designed to
power a Katmai, it should handle a Tualatin.


It isn't (processor) 'power', per see, that's the problem with
the regulators, it's the *current*.

35 watts at 2 volts is 17.5 amps but at 1.5 volts it's 23.3
amps, or 33% more than it was 'designed for'.


I agree with your calculations, but my understanding is it's
overheating that causes the demise of these components over the
medium term - and I would expect temperature to be proportional
to power dissipated, i.e. 35 watts in both cases.


Well, yes, it's the 'power' but not the processor's power. It's
the current through the FETs, which causes voltage drop across
them (due to their RDSon resistance), which is power.

Say it's RDSon is .02 ('typical' for my BH6 FET with 5V gate).
That would be (P = I^2 x R) 23.2^2 x .02 = 10.8578 watts (average).


Ah, that explains it :-)

Asus used several different Vcore FETs on the various flavours of
P2B I have here, but after trawling for datasheets I've learned
the 'wimpiest' one they used has .015 ohm RDSon (typical @ Vgs=4v,
Id=20A), maximum continuous drain current of 45A (peak 180A), and
maximum channel dissipation of 75W (at room temperature). At a
case temperature of 100C, they are rated for ~30W.

Even using worst-case numbers (RDSon max=.025 ohms, Vcore=1.4V,
CPU=35W) yields an average dissipation of only 15.625W - so it's
little wonder P2B series boards can run overclocked Tualatins
without raising a sweat :-)


You need to be careful reading specs and while you didn't say what
the part number was there's no way that FET can handle 75W free
air, or soldered to a dinky PCB heat tab. I'll bet that rating is
the 'theoretical' maximum with an 'infinite' heatsink, or some such
'in-your-dreams' criteria.


Point taken.

75W is the absolute maximum rating for the surface mount version
with a component case temperature of 25C (only parameter specified).
There's no mention of heatsinking requirements, I assume you are
supposed to figure that out from the handy-dandy case temperature
vs. channel dissipation de-rating graph - which is where I obtained
the 30W @ 100C figure.


Ah. OK, the 'case temp' spec. That one is even more 'subtle' because,
unless you've tried it, there's no clue as to just how dern near
impossible keeping the CASE at 25C is. Note, that's on the 'wrong'
side of any thermal compound (or solder in the case of a PCB mount)
you might employ to sink it to something.

Given there's no reason to think the FETs will be asked to dissipate
more than about 16W average, I'm fairly confident they aren't at
risk - but feel free to peruse the datasheet to see if you agree:

http://www.alldatasheet.co.kr/datash...HI/2SK2885.pdf


Well, it seems to me that you're looking at the 'watts' vs 'temp'
and, whether consciously or not, imagining the case will BE at that
temp when those watts are dissipated, but that isn't what the spec is
saying. It's saying, IF you KEEP the CASE at this temp (good luck,
how?) then it can dissipate those watts. (Specifying a 'fixed',
independent, non varying case temp is equivalent to saying it's on an
infinite heatsink, btw)

That's basically a junction to case thermal spec,
I.E. C/W = [max_junction_temp-case_temp]/Watts; e.g. [150-25]/75=
1.67C/W;
and not an 'as used in the real world' temperature spec, and
certainly not the power handling capacity when simply solder mounted
onto a PCB heat tab.

They don't give 'free air' numbers but, for an example, here's a
similar type device that does give free air and 'minimum footprint'
for surface mount thermal resistance numbers.
http://www.semiconductors.com/acroba...P55N03LT_6.pdf


The TO-220AB version is 60C/W in free air and the surface mount is
50C/W. For it's max junction of 175C, and an ambient of 25C, that
comes to 2.5 Watts free air and a whopping 3 Watts on a minimum
footprint (meaning no extra thermal tab area) surface mount. Now,
before you reel in shock and say that one's no where near yours, it's
junction to case resistance is 1.45C/W vs 1.67C/W for yours. And,
using the 'case temp' methodology, if you keep the *case* at 25C
(good luck, how?), that comes to [175-25]/1.45 = 103.45 watts (@175C
junction temp, 25C more than yours allows). Which is why it lists
"103W" was the "total power dissipation" (at Tmb=25C where 'mb'
stands for "mounting base," which is 'the case'). If, however, we use
the same 150C junction temp yours is limited to it's still 86 watts
(at 25C CASE temp): more than the 75 watts yours is rated at.

It's always a shocker the first time a home hobbyist discovers his
"75 Watt Power FET" is ready to cook eggs with only a couple of watts
being dissipated (woops, guess I need a heatsink =:O)).

Now, your Vcore FET isn't dissipating into free air, and most likely
has a motherboard thermal tab, but those numbers illustrate just how
important the thermal sinking is and how the 'case temp' spec can
lead one to over-estimate the power handling capability.


I think I followed most of that - and thanks for your effort - but I'm
glad I don't design this stuff for a living :-)


Hehe. It's always a contradictory set of conditions and requirements:
like "Free Lunch: $1.00"

I taped a temperature sensor on top one of the FETs, and it peaked at
95C after running a Tualatin 1.0A @ 1.5Ghz on 1.4Vcore full bore for
10 minutes, then eventually dropped to 65C at idle.

Based on the above dissipation calculations, and the temperature
derating graph from the datasheet, it still seems to me the FET is
only dissipating about 50% of it's maximum under load. Agree?


You're still looking at that curve cockeyed, as if knowing the temp
'means' something, but your FET is not mounted on a perfect, infinite,
heatsink, as if required for that curve to exist. And so, seeing 'xC'
doesn't mean a thing about how much of it's 'rating' is being used,
depending on what you 'mean' by 'rating' (current? total power? junction
temp?) You can't even tell how much power is being dissipated because
you don't know the thermal resistance of whatever 'heatsinking' it has.

Be gentle with me, 'cos I'm still missing something....

What does the case temperature vs. channel dissipation curve tell us?

Assuming the case temperature *is* 100C, does it mean the channel *is*
dissipating ~30W, or does it mean anything *under* ~30W is safe at that
temperature? (My assumption). Or something else entirely?

I don't understand why you say we need to know the thermal resistance of
the sink to determine how much power is dissipated. Didn't we previously
agreed it's equal to drain current squared x RDSon? - and we have a
number for RDSon (from FET datasheet) and a reasonable guess as to drain
current (from CPU datasheet).

I can see it now... you will try to explain again, I'll finally get it -
then my FETs will blow up :-)

What you CAN say is it seems to be within it's temperature limit of 150C
with 'whatever' power it's dissipating but you don't really know how
many 'watts' is producing that temp and so not how many 'more' would
push it over. That wouldn't answer the peak current question, either, btw

Well, let's see if we can 'guess' some things. Stop grant is 15.37 amps
(but then you're not at the 'right' Vcore), and maybe we can guess
that's the majority of the 'idle' time. That would be 15.37^2* x .015
(typical RDSon) for 3.54 Watts and you're seeing 65C there. What is
ambient? 30C? That would be (65-30)/3.54 for 9.89C/Watt (must be getting
some PCB heatsinking to lower that down from 50C/Watt).

If we take the tualatin's max current spec and the 'typical' RDSon of
.015 we'd get 8.14 watts. If we 'guess' that our 'idle' C/W number is
right that's 8.14 x 9.89 for 80.50C, plus the assumed 30C ambient, for
110.5C. Well, we're significantly off (16%) but we're in the ball park
(it gets closer if the ambient around the FET is higher) and it suggests
that the tualatin probably isn't pulling the spec sheet 'max' current;
which we'd expect anyway as yours is, hopefully, more 'typical' and
you're also running Vcore a bit low (although clocking it faster than
the max spec'd 1.4 gig. I can account for about 7% from these two). Not
to mention we made some 'guesses' (such as my wild stab in the dark
about your case ambient) and who knows just how precise your temp
measurement is. Which one is closer is debatable but I'd trust the
idle-stop grant number before I'd take a 'max' current spec as 'typical'
(for the reasons given plus it's independent of clock speed).

BUT, if you were a designer you'd have to design for the max because,
sure as shootin, SOME of them WOULD pull that much. (Which is how you
get the manufacturer not willing to say it works while 'hotshots' tell
you theirs works 'just fine')

Anyway, let's say our 'idle' thermal resistance is 'close'. That would
mean your 'max load' test, since you got 95C instead of 110.5, was
closer to 20.9 amps actual current draw (through the .015 RDSon) rather
than the spec'd 'max'. I.E. (95-30)/9.89 = 6.57 watts dissipated and
sqrt(6.57/.015) = 20.9 amps

Soooooo. What amps would it take to hit the 150C? Using the same
equation we get 28.4 amps will heat'er up to 150C.

Which means, at your 'maximum load', you're running at about 74% of what
it would take, measured as current, to exceed the maximum temp rating;
assuming these numbers resemble reality in any way, ambient doesn't
increase, and the creek don't rise g

Btw, if someone else were running 1.7V Vcore to get the same overclock
that's a 21.4% increase, for 25 amps, and that's getting close enough to
the 28.4 'cooking point', considering how much of a 'guess' these number
are, that I can see it possibly failing. 1.8V Vcore would be 26.9 amps.

It does illustrate the significance of too high a Vcore on the regulator
FET. Using 1.7 vs your 1.4 raises the FET temp an estimated 30C.

~misfit~ wrote:

David Maynard wrote:

snip, I've been following this interestedly, thanks guys

It does illustrate the significance of too high a Vcore on the
regulator FET. Using 1.7 vs your 1.4 raises the FET temp an estimated
30C.


So, David, how would someone like me g go about unsoldering FETs to
replace them? There's a large surface stuck to the board. And, if I did
manage to get them unstuck would I be better leaving the replacements
standing up with heatsinks on them?

Cheers,
--
~misfit~

I hope Homie won't mind... the following is his description of FET
replacement techniques from an email discussion we had quite some time
ago ...

The power fets can be removed/replaced using a very wide tip. First lift the
outer legs, then idea is to spread the heat evenly across the entire tab/pad
area. wet the pad & tab with some kester "44" solder, then apply the wide tip
in a manner that gives the most surface area contact, at the same time grasp
one of the legs with needle nose pliers, the fet will come right off.
I keep some alcohol in a small dish to drop the fet in so it doesn't stay hot &
possibly damage it (if I am not sure it's good).
To install a new fet, use the same wide tip & wick the pad very clean, then
wet the pad with a very thin coat of silver based solder, let the pad cool,
align the new fet & tack the legs to hold it as aligned as possible, now heat
the pad & re-wet while using a wooden dowel to apply slight pressure to the new
fet, as soon as you see complete flow & the fet sinks into position, remove the
heat, hold the fet for a few seconds then use an alcohol soaked q-tip to cool
the fet.
I replaced a lot of these fets before I was able to make it look good.

P2B

David Maynard wrote:
[snip]
Which means, at your 'maximum load', you're running at about 74% of what
it would take, measured as current, to exceed the maximum temp rating;
assuming these numbers resemble reality in any way, ambient doesn't
increase, and the creek don't rise g

You forgot "if the Good Lord's willing". Johnny is spinning in his grave :-)

P2B wrote:
~misfit~ wrote:

David Maynard wrote:

snip, I've been following this interestedly, thanks guys

It does illustrate the significance of too high a Vcore on the
regulator FET. Using 1.7 vs your 1.4 raises the FET temp an
estimated 30C.


So, David, how would someone like me g go about unsoldering FETs to
replace them? There's a large surface stuck to the board. And, if I
did
manage to get them unstuck would I be better leaving the replacements
standing up with heatsinks on them?

Cheers,
--
~misfit~

I hope Homie won't mind... the following is his description of FET
replacement techniques from an email discussion we had quite some time
ago ...

The power fets can be removed/replaced using a very wide tip. First
lift the outer legs, then idea is to spread the heat evenly across
the entire tab/pad area. wet the pad & tab with some kester "44"
solder, then apply the wide tip in a manner that gives the most
surface area contact, at the same time grasp one of the legs with
needle nose pliers, the fet will come right off.
I keep some alcohol in a small dish to drop the fet in so it doesn't
stay hot & possibly damage it (if I am not sure it's good).
To install a new fet, use the same wide tip & wick the pad very
clean, then wet the pad with a very thin coat of silver based
solder, let the pad cool, align the new fet & tack the legs to hold
it as aligned as possible, now heat the pad & re-wet while using a
wooden dowel to apply slight pressure to the new fet, as soon as you
see complete flow & the fet sinks into position, remove the heat,
hold the fet for a few seconds then use an alcohol soaked q-tip to
cool the fet.
I replaced a lot of these fets before I was able to make it look
good.

Thanks P2B, and Homie.
--
~misfit~

P2B wrote:



David Maynard wrote:

P2B wrote:

David Maynard wrote:

P2B wrote:

David Maynard wrote:

P2B wrote:

David Maynard wrote:

P2B wrote:

David Maynard wrote:

P2B wrote:


It really depends on the board - my P2B systems have run 7x24
for over 2 years with no sign of problems. The FETs and caps
don't even get hot (no extra cooling or sinks), just don't
touch the clock chip at 150Mhz FSB!

Keep in mind the Tualatin only draws slightly more power than
a Katmai P3 600 - so if the board was *properly* designed to
power a Katmai, it should handle a Tualatin.



It isn't (processor) 'power', per see, that's the problem with
the regulators, it's the *current*.

35 watts at 2 volts is 17.5 amps but at 1.5 volts it's 23.3
amps, or 33% more than it was 'designed for'.



I agree with your calculations, but my understanding is it's
overheating that causes the demise of these components over the
medium term - and I would expect temperature to be proportional
to power dissipated, i.e. 35 watts in both cases.



Well, yes, it's the 'power' but not the processor's power. It's
the current through the FETs, which causes voltage drop across
them (due to their RDSon resistance), which is power.

Say it's RDSon is .02 ('typical' for my BH6 FET with 5V gate).
That would be (P = I^2 x R) 23.2^2 x .02 = 10.8578 watts (average).



Ah, that explains it :-)

Asus used several different Vcore FETs on the various flavours of
P2B I have here, but after trawling for datasheets I've learned
the 'wimpiest' one they used has .015 ohm RDSon (typical @
Vgs=4v, Id=20A), maximum continuous drain current of 45A (peak
180A), and maximum channel dissipation of 75W (at room
temperature). At a case temperature of 100C, they are rated for
~30W.

Even using worst-case numbers (RDSon max=.025 ohms, Vcore=1.4V,
CPU=35W) yields an average dissipation of only 15.625W - so it's
little wonder P2B series boards can run overclocked Tualatins
without raising a sweat :-)


You need to be careful reading specs and while you didn't say what
the part number was there's no way that FET can handle 75W free
air, or soldered to a dinky PCB heat tab. I'll bet that rating is
the 'theoretical' maximum with an 'infinite' heatsink, or some
such 'in-your-dreams' criteria.



Point taken.

75W is the absolute maximum rating for the surface mount version
with a component case temperature of 25C (only parameter
specified). There's no mention of heatsinking requirements, I
assume you are supposed to figure that out from the handy-dandy
case temperature vs. channel dissipation de-rating graph - which is
where I obtained the 30W @ 100C figure.



Ah. OK, the 'case temp' spec. That one is even more 'subtle'
because, unless you've tried it, there's no clue as to just how dern
near impossible keeping the CASE at 25C is. Note, that's on the
'wrong' side of any thermal compound (or solder in the case of a PCB
mount) you might employ to sink it to something.

Given there's no reason to think the FETs will be asked to
dissipate more than about 16W average, I'm fairly confident they
aren't at risk - but feel free to peruse the datasheet to see if
you agree:

http://www.alldatasheet.co.kr/datash...HI/2SK2885.pdf



Well, it seems to me that you're looking at the 'watts' vs 'temp'
and, whether consciously or not, imagining the case will BE at that
temp when those watts are dissipated, but that isn't what the spec
is saying. It's saying, IF you KEEP the CASE at this temp (good
luck, how?) then it can dissipate those watts. (Specifying a
'fixed', independent, non varying case temp is equivalent to saying
it's on an infinite heatsink, btw)

That's basically a junction to case thermal spec,
I.E. C/W = [max_junction_temp-case_temp]/Watts; e.g. [150-25]/75=
1.67C/W;
and not an 'as used in the real world' temperature spec, and
certainly not the power handling capacity when simply solder mounted
onto a PCB heat tab.

They don't give 'free air' numbers but, for an example, here's a
similar type device that does give free air and 'minimum footprint'
for surface mount thermal resistance numbers.
http://www.semiconductors.com/acroba...P55N03LT_6.pdf


The TO-220AB version is 60C/W in free air and the surface mount is
50C/W. For it's max junction of 175C, and an ambient of 25C, that
comes to 2.5 Watts free air and a whopping 3 Watts on a minimum
footprint (meaning no extra thermal tab area) surface mount. Now,
before you reel in shock and say that one's no where near yours,
it's junction to case resistance is 1.45C/W vs 1.67C/W for yours.
And, using the 'case temp' methodology, if you keep the *case* at
25C (good luck, how?), that comes to [175-25]/1.45 = 103.45 watts
(@175C junction temp, 25C more than yours allows). Which is why it
lists "103W" was the "total power dissipation" (at Tmb=25C where
'mb' stands for "mounting base," which is 'the case'). If, however,
we use the same 150C junction temp yours is limited to it's still 86
watts (at 25C CASE temp): more than the 75 watts yours is rated at.

It's always a shocker the first time a home hobbyist discovers his
"75 Watt Power FET" is ready to cook eggs with only a couple of
watts being dissipated (woops, guess I need a heatsink =:O)).

Now, your Vcore FET isn't dissipating into free air, and most likely
has a motherboard thermal tab, but those numbers illustrate just how
important the thermal sinking is and how the 'case temp' spec can
lead one to over-estimate the power handling capability.



I think I followed most of that - and thanks for your effort - but
I'm glad I don't design this stuff for a living :-)



Hehe. It's always a contradictory set of conditions and requirements:
like "Free Lunch: $1.00"

I taped a temperature sensor on top one of the FETs, and it peaked at
95C after running a Tualatin 1.0A @ 1.5Ghz on 1.4Vcore full bore for
10 minutes, then eventually dropped to 65C at idle.

Based on the above dissipation calculations, and the temperature
derating graph from the datasheet, it still seems to me the FET is
only dissipating about 50% of it's maximum under load. Agree?



You're still looking at that curve cockeyed, as if knowing the temp
'means' something, but your FET is not mounted on a perfect, infinite,
heatsink, as if required for that curve to exist. And so, seeing 'xC'
doesn't mean a thing about how much of it's 'rating' is being used,
depending on what you 'mean' by 'rating' (current? total power?
junction temp?) You can't even tell how much power is being dissipated
because you don't know the thermal resistance of whatever
'heatsinking' it has.


Be gentle with me, 'cos I'm still missing something....

What does the case temperature vs. channel dissipation curve tell us?

It tells you how much power it can dissipate --IF-- the case is kept at
that temperature (not that it WILL be at that temperature). It's a
statement of the 'ideal'. It's nothing more than the junction to case
thermal resistance plotted on a power/temp graph.


Assuming the case temperature *is* 100C,

That IS the 'given' (well, in the way you're using it).

does it mean the channel *is*
dissipating ~30W,

No.

or does it mean anything *under* ~30W is safe at that
temperature? (My assumption).

That is correct.

So if you measure your FET case temp, as you did, and see it at 65C then
the power being dissipated could be ANYTHING from the curve all the way
down to 0. Which is why I said measuring case temp doesn't TELL you
anything about how much power is being dissipated or 'what percentage' of
'what?' it's operating at.

Or something else entirely?

I don't understand why you say we need to know the thermal resistance of
the sink to determine how much power is dissipated.

Well, you said it yourself, the power being dissipated can be "anything
*under*" the measured temp/power point. Ok, since it's "anything *under*"
then what is it?

Didn't we previously
agreed it's equal to drain current squared x RDSon?

Except you don't know what the drain current is from the temp/power graph
either.

- and we have a
number for RDSon (from FET datasheet) and

a reasonable guess as to drain
current (from CPU datasheet).

Which has nothing to do with the FET's power/temp graph. The issue here was
whether knowing it's at 65C means anything vs the power/temp graph. It
doesn't, except you can be sure it isn't above that point.

That graph tells you the BEST you could POSSIBLY get; with 'possibly get'
being with a 'perfect' heatsink. But since you know for dern sure you don't
have anything resembling perfection then all you know is it's going to be
something less than that point, but just what is unknown without more data.


I can see it now... you will try to explain again, I'll finally get it -
then my FETs will blow up :-)

Hehe. Let's hope for number 2 without number 3 (number 1 was a given)


What you CAN say is it seems to be within it's temperature limit of
150C with 'whatever' power it's dissipating but you don't really know
how many 'watts' is producing that temp and so not how many 'more'
would push it over. That wouldn't answer the peak current question,
either, btw

Well, let's see if we can 'guess' some things. Stop grant is 15.37
amps (but then you're not at the 'right' Vcore), and maybe we can
guess that's the majority of the 'idle' time. That would be 15.37^2* x
.015 (typical RDSon) for 3.54 Watts and you're seeing 65C there. What
is ambient? 30C? That would be (65-30)/3.54 for 9.89C/Watt (must be
getting some PCB heatsinking to lower that down from 50C/Watt).

If we take the tualatin's max current spec and the 'typical' RDSon of
.015 we'd get 8.14 watts. If we 'guess' that our 'idle' C/W number is
right that's 8.14 x 9.89 for 80.50C, plus the assumed 30C ambient, for
110.5C. Well, we're significantly off (16%) but we're in the ball park
(it gets closer if the ambient around the FET is higher) and it
suggests that the tualatin probably isn't pulling the spec sheet 'max'
current; which we'd expect anyway as yours is, hopefully, more
'typical' and you're also running Vcore a bit low (although clocking
it faster than the max spec'd 1.4 gig. I can account for about 7% from
these two). Not to mention we made some 'guesses' (such as my wild
stab in the dark about your case ambient) and who knows just how
precise your temp measurement is. Which one is closer is debatable but
I'd trust the idle-stop grant number before I'd take a 'max' current
spec as 'typical' (for the reasons given plus it's independent of
clock speed).

BUT, if you were a designer you'd have to design for the max because,
sure as shootin, SOME of them WOULD pull that much. (Which is how you
get the manufacturer not willing to say it works while 'hotshots' tell
you theirs works 'just fine')

Anyway, let's say our 'idle' thermal resistance is 'close'. That would
mean your 'max load' test, since you got 95C instead of 110.5, was
closer to 20.9 amps actual current draw (through the .015 RDSon)
rather than the spec'd 'max'. I.E. (95-30)/9.89 = 6.57 watts
dissipated and sqrt(6.57/.015) = 20.9 amps

Soooooo. What amps would it take to hit the 150C? Using the same
equation we get 28.4 amps will heat'er up to 150C.

Which means, at your 'maximum load', you're running at about 74% of
what it would take, measured as current, to exceed the maximum temp
rating; assuming these numbers resemble reality in any way, ambient
doesn't increase, and the creek don't rise g

Btw, if someone else were running 1.7V Vcore to get the same overclock
that's a 21.4% increase, for 25 amps, and that's getting close enough
to the 28.4 'cooking point', considering how much of a 'guess' these
number are, that I can see it possibly failing. 1.8V Vcore would be
26.9 amps.

It does illustrate the significance of too high a Vcore on the
regulator FET. Using 1.7 vs your 1.4 raises the FET temp an estimated
30C.