Highest healthy temperature?

Hi. Just wanted to know how high you can push a 7950GT in temperature.
The treshold is at 130 degrees celsius.
Would 100 be ok or would that shorten the life of the card
dramatically??

Thx

Hynix wrote:
Hi. Just wanted to know how high you can push a 7950GT in temperature.
The treshold is at 130 degrees celsius.
Would 100 be ok or would that shorten the life of the card
dramatically??

Thx


100c is ok, but high, for an on die temp reading during full load for a
normal lifespan. However, you must make sure that proper airflow is
occurring around the unit or that radiation will heat up components that
are not as robust. (Drying out capacitors for example).

I have found, with all pipes active and zero airflow, the 7950GT
naturally finds 92-95c in a closed case...75c in an open case.

On 26 Jan, 20:00, "Mr.E Solved!" wrote:
100c is ok, but high, for an on die temp reading during full load for a
normal lifespan. However, you must make sure that proper airflow is
occurring around the unit or that radiation will heat up components that
are not as robust. (Drying out capacitors for example).

I have found, with all pipes active and zero airflow, the 7950GT
naturally finds 92-95c in a closed case...75c in an open case.


Thank you.
I found that the card was between 75 and 99 degrees during an hour of
gameplay.
The card is slightly OC'd but has so far only seen an increase of about
5-10 degrees compared to non OC.
The airflow of the case is sufficient but perhaps I should consider a
better cooler for the GPU.
Well, thanks again.

Hynix wrote:
On 26 Jan, 20:00, "Mr.E Solved!" wrote:
100c is ok, but high, for an on die temp reading during full load for a
normal lifespan. However, you must make sure that proper airflow is
occurring around the unit or that radiation will heat up components that
are not as robust. (Drying out capacitors for example).

I have found, with all pipes active and zero airflow, the 7950GT
naturally finds 92-95c in a closed case...75c in an open case.


Thank you.
I found that the card was between 75 and 99 degrees during an hour of
gameplay.
The card is slightly OC'd but has so far only seen an increase of about
5-10 degrees compared to non OC.
The airflow of the case is sufficient but perhaps I should consider a
better cooler for the GPU.
Well, thanks again.


Maybe you need one of these. Fan is a separate item. It's pretty
big, so I don't know if it'll run into stuff or not.

Thermalright HR-03 VGA Cooling Heatsink $50
http://www.newegg.com/Product/Produc...82E16835109134

Paul

Hynix wrote:
On 26 Jan, 20:00, "Mr.E Solved!" wrote:
100c is ok, but high, for an on die temp reading during full load for a
normal lifespan. However, you must make sure that proper airflow is
occurring around the unit or that radiation will heat up components that
are not as robust. (Drying out capacitors for example).

I have found, with all pipes active and zero airflow, the 7950GT
naturally finds 92-95c in a closed case...75c in an open case.


Thank you.
I found that the card was between 75 and 99 degrees during an hour of
gameplay.
The card is slightly OC'd but has so far only seen an increase of about
5-10 degrees compared to non OC.
The airflow of the case is sufficient but perhaps I should consider a
better cooler for the GPU.
Well, thanks again.



I neglected to mention those readings were for overclocked, passively
cooled 7950GF's. If you have an active fan already in place, and are
still reading 100c, then you need to increase circulation in the case
unless you are already getting good case temp readings. (No more than
10c over ambient). Fun! Not.

On Fri, 26 Jan 2007 14:00:27 -0500, "Mr.E Solved!"
wrote:

Hynix wrote:
Hi. Just wanted to know how high you can push a 7950GT in temperature.
The treshold is at 130 degrees celsius.
Would 100 be ok or would that shorten the life of the card
dramatically??

Thx


100c is ok, but high, for an on die temp reading during full load for a
normal lifespan. However, you must make sure that proper airflow is
occurring around the unit or that radiation will heat up components that
are not as robust. (Drying out capacitors for example).

I have found, with all pipes active and zero airflow, the 7950GT
naturally finds 92-95c in a closed case...75c in an open case.


How and under what test circumstances did you measure the on-chip
temperature? Please give complete data and be technically precise with
your answer.

thanks in advance,

John Lewis

John Lewis wrote:
On Fri, 26 Jan 2007 14:00:27 -0500, "Mr.E Solved!"
wrote:

Hynix wrote:
Hi. Just wanted to know how high you can push a 7950GT in temperature.
The treshold is at 130 degrees celsius.
Would 100 be ok or would that shorten the life of the card
dramatically??

Thx

100c is ok, but high, for an on die temp reading during full load for a
normal lifespan. However, you must make sure that proper airflow is
occurring around the unit or that radiation will heat up components that
are not as robust. (Drying out capacitors for example).

I have found, with all pipes active and zero airflow, the 7950GT
naturally finds 92-95c in a closed case...75c in an open case.


How and under what test circumstances did you measure the on-chip
temperature? Please give complete data and be technically precise with
your answer.

thanks in advance,

John Lewis

I use both Epox's Thunder Probe and USDM for managing various thermal
sensors. One type is a flat thermistor which is easily trimmed and
fitted into the most narrow of spaces. Such as alongside a GPU die. I
personally do not recommend fitting a thermistor between a heatsink and
die, but some prefer catching the tip of it inbetween for potential
gains in accuracy. Thermistors vary in accuracy and resolution, I
suspect on-die doesn't matter much where on die it is.

Once the thermistor is in place, readings are managed like any other,
and compared with nVidia's built-in GPU temperature sensor, calibrations
are easily taken at both low and high points of the temperature samples.

In fact, I have two thermistors installed, one in the method I prefer,
the other not. However, only one can be read at a time due to
motherboard IO restrictions. I am satisfied with their accuracy +/-1c
resolved to 0.1c.

Are you curious how to go about the process? Is that why you want
details? There are many enthusiast sites who will provide as much
'technical precision' as you can significantly digitize. Here's a link,
enjoy!

http://www.procooling.com/index.php?...s&disp=61&pg=1

On Tue, 30 Jan 2007 00:23:05 -0500, "Mr.E Solved!"
wrote:

John Lewis wrote:
On Fri, 26 Jan 2007 14:00:27 -0500, "Mr.E Solved!"
wrote:

Hynix wrote:
Hi. Just wanted to know how high you can push a 7950GT in temperature.
The treshold is at 130 degrees celsius.
Would 100 be ok or would that shorten the life of the card
dramatically??

Thx

100c is ok, but high, for an on die temp reading during full load for a
normal lifespan. However, you must make sure that proper airflow is
occurring around the unit or that radiation will heat up components that
are not as robust. (Drying out capacitors for example).

I have found, with all pipes active and zero airflow, the 7950GT
naturally finds 92-95c in a closed case...75c in an open case.


How and under what test circumstances did you measure the on-chip
temperature? Please give complete data and be technically precise with
your answer.

thanks in advance,

John Lewis

I use both Epox's Thunder Probe and USDM for managing various thermal
sensors. One type is a flat thermistor which is easily trimmed and
fitted into the most narrow of spaces. Such as alongside a GPU die. I
personally do not recommend fitting a thermistor between a heatsink and
die, but some prefer catching the tip of it inbetween for potential
gains in accuracy. Thermistors vary in accuracy and resolution, I
suspect on-die doesn't matter much where on die it is.

Once the thermistor is in place, readings are managed like any other,
and compared with nVidia's built-in GPU temperature sensor, calibrations
are easily taken at both low and high points of the temperature samples.

In fact, I have two thermistors installed, one in the method I prefer,
the other not. However, only one can be read at a time due to
motherboard IO restrictions. I am satisfied with their accuracy +/-1c
resolved to 0.1c.

Are you curious how to go about the process? Is that why you want
details? There are many enthusiast sites who will provide as much
'technical precision' as you can significantly digitize. Here's a link,
enjoy!

http://www.procooling.com/index.php?...s&disp=61&pg=1

Thanks for your answer. My curiosity was stimulated by some
professional experience in electronic reliability engineering and
testing.

So the GPU header temperature that you measured with the close-contact
thermistor was 95 degrees C ?

What application(s) were you running when you measured this
temperature. ?

I do not advise ever running a GPU anywhere near 100 degrees header
("case") temperature for extended periods. The statistical lifetime
profile for commercial modern-process silicon is roughly halved for
every 10 degrees C above 75 degrees C. "core" temperature. NVidia's
shutdown limits of 110-130 degrees C is there to take care of the GPU
silicon in the case of a catastrophic cooling failure (e.g: fan
seizure) and is not a recommendation ever to continuously run their
silicon at anything near those temperatures. Please also remember
that electron-propagation in silicon rapidly slows down with
increasing temperature. The timing specs for commercial silicon are
valid for up to 70 degrees C case (~ 75degrees core) temperature, and
nVidia probably uses design-models for 110 degrees device temperature
in their simulations to allow for worst-case process deviations at 75
degrees C. The variable-speed fan-cooling on the typical latest nVidia
reference boards is designed to run at a speed that keeps the GPU case
temperature around 70-75 degrees C. max -- optimum compromise
between noise and silicon-lifetime, assuming a reasonably ventilated
computer-case, of course.

Passive-cooling "solutions" that do not end up with silicon-device
temperatures on a par with their active conterparts are bad news for
long-term reliability.

John Lewis

John Lewis wrote:


So the GPU header temperature that you measured with the close-contact
thermistor was 95 degrees C ?

Yes, 95c and higher, under some conditions 100c. With zero artifacts.
Passively cooled with a sintered-powder heat pipe.



What application(s) were you running when you measured this
temperature. ?

As mentioned: Epox's USDM and ThunderProbe (Ostensibly only for Epox
motherboards)


I do not advise ever running a GPU anywhere near 100 degrees header
("case") temperature for extended periods. The statistical lifetime
profile for commercial modern-process silicon is roughly halved for
every 10 degrees C above 75 degrees C. "core" temperature. NVidia's
shutdown limits of 110-130 degrees C is there to take care of the GPU
silicon in the case of a catastrophic cooling failure (e.g: fan
seizure) and is not a recommendation ever to continuously run their
silicon at anything near those temperatures. Please also remember
that electron-propagation in silicon rapidly slows down with
increasing temperature. The timing specs for commercial silicon are
valid for up to 70 degrees C case (~ 75degrees core) temperature, and
nVidia probably uses design-models for 110 degrees device temperature
in their simulations to allow for worst-case process deviations at 75
degrees C. The variable-speed fan-cooling on the typical latest nVidia
reference boards is designed to run at a speed that keeps the GPU case
temperature around 70-75 degrees C. max -- optimum compromise
between noise and silicon-lifetime, assuming a reasonably ventilated
computer-case, of course.

Passive-cooling "solutions" that do not end up with silicon-device
temperatures on a par with their active conterparts are bad news for
long-term reliability.

John Lewis


This is all true, but in this case, the on-die temps taken in
conjunction with the GPU diode temp (under the GPU), while high are
within operational parameters and exhibit no side effects.

The sintered powder heat pipe is astonishingly effective in 'moving'
heat from one place to another, hence the success of passively cooled cards.

While excessive heat does diminish the life span of the core, the life
span of a core, thermally stressed or not, is generally longer than it's
useful life in a person's case.

I know, it's hard to accept those temps after so many years of avoiding
them...but they exist during load, and drop quickly to more traditional
levels when back to a 2d desktop.

Just to make it clearer for all: On-Die temps are not the same and are
higher than the GPU temp presented in the nvidia control panel, which is
not an 'on-die' reading, but from a diode resting under the GPU. It may
or may not physically touch it.

Idle Speculation Section: 75c @ 5years = 100c @ 3years life.

cut
Idle Speculation Section: 75c @ 5years = 100c @ 3years life.

Not so idle. Who'd want a card 3 years old anyways?