Acceptable CPU temps?

[Doc]

Just got an E8400 3gig Core2 Duo CPU to run on a Gigabyte GA-P35-DS3L Rev 2.0 board.

I've been fooling with overclocking, currently have it running at 4.0 gig. Using an OCZ Vendetta 2 cpu cooler.

I've seen it go as high as 68deg during gameplay. Is this high or what you'd expect?

Per Realtemp the lowest it will register is 38 deg at idle. This makes me wonder if the reading is a bit high, I see others report lower idle temps - like 30 deg. As of yet I haven't calibrated it, need to get a thermometer to measure the case temp.

Thanks for all input.

 

[Paul]

Just got an E8400 3gig Core2 Duo CPU to run on a Gigabyte GA-P35-DS3L Rev 2.0 board.

I've been fooling with overclocking, currently have it running at 4.0 gig. Using an OCZ Vendetta 2 cpu cooler.

I've seen it go as high as 68deg during gameplay. Is this high or what you'd expect?

Per Realtemp the lowest it will register is 38 deg at idle. This makes me wonder if the reading is a bit high, I see others report lower idle temps - like 30 deg. As of yet I haven't calibrated it, need to get a thermometer to measure the case temp.

Thanks for all input.

At the high temperature end, what's important is whether it
is throttling or not. You lose performance if the processor
is thermally limiting itself.

http://ixbtlabs.com/articles2/cpu/intel-thermal-features-core2.html

The digital temperature readout, is closest at the high temp end.
If it reads 68C, it could be 67C or 69C.
If it reads 40C, it could be 30C or 50C.
So the readings are closer on the high end (as presumably the
trip point is calibrated at the factory). Someone did the
experiment, of measuring the CPU carefully, and that's what
they found as a general principle. It's "looser on the low end",
where a precise measurement isn't important.

And an indirect proof of that (seen with other thermal measurement
solutions in the past, on older processors), is when your temp
readout is "sub-ambient" on an air cooled system. Which tells
you the thing is way off. 38C is like, human body temperature,
so not exactly boiling hot. Whereas 68C would be uncomfortable
to the touch (you'd pull your finger away within two seconds).
The air in the computer case itself could be around 35C, depending
on the level of air movement through the case, so at the low end
it really can't get much lower with respect to the computer case
air temperature. If you were reading 30C on the processor, for that
to be possible, the air temperature inside the case would have to be
lower than that. Which you can do, if the room is uncomfortably cold.

Paul

 

[Doc]

So the readings are closer on the high end (as presumably the

trip point is calibrated at the factory). Someone did the

experiment, of measuring the CPU carefully, and that's what

they found as a general principle. It's "looser on the low end",

where a precise measurement isn't important.

Okay, I appreciate the info. How high a temp do you think is acceptable?

 

[Paul]

Okay, I appreciate the info. How high a temp do you think is acceptable?

The acceptable temperature is the throttle point.

In the past, a Tcase of 65C was considered to be a good place to be.
If we add 25C for the case to junction drop, 65+25=90C Tj or junction
temperature. And the E8400 probably has a Tjmax of 100C. It should start
to throttle when it gets to that point. So perhaps 10C below throttle
point, might be a good metric to use.

On my stock E8400, running Prime95 (small FFT) at 3GHz

Tj_Core0 Tj_Core1
Idle temps 42C 35C
Prime95 temps 55C 52C

There's still a fair distance before I get to 90C for Tj.

My heatsink is bolted to the motherboard, so there's no question
how much pressure the heatsink is getting in terms of normal force.
I used the stock cooler on my other PC (that cooler with the push
pins). And one pin is always a bit flaky on that thing. But the cooler
on this PC, bolts right through the motherboard, so it's a bit more
difficult for it to break loose.

If your CPU is really too hot, the overheating is controlled by the
CPU reducing the effective clock rate. And the throttle bit gets
set. And the throttle bit is set, when you reach Tjmax (whatever
that temperature happens to be, for each family of processor).

In terms of damage to silicon, when I asked the question of our
cell library guy at our fab, our CMOS starts to "cook" at 130C.
Maintaining that high a junction, for a prolonged period of time,
can cause a parametric shift. And with a high overclock, it also
makes electromigration worse. (My fab days were many years ago now,
and much has changed since then. And that silicon fab is even gone now,
as are all the staff.)

The other limiting factor, is the IC package used for the processor.
At one time, ceramic packages were quite popular. My first project,
a majority of the chips were in ceramic, and the ceramic could take
as much heat as the silicon could. Now that cheesy "organic" packages
are used for ICs (no longer a hermetic glass frit seal), the package
is the thing that gets damaged first. And that's where the 100C limit
comes from. The package would start to degrade, if the temperature
gets too high. And that could happen, before the silicon is damaged.
So 90C is a decent distance from 130C at least. And 10C before
the package is getting close to its suggested long term 100C Tj limit.

Paul

 

[Doc]

My heatsink is bolted to the motherboard, so there's no question

how much pressure the heatsink is getting in terms of normal force.

I used the stock cooler on my other PC (that cooler with the push

pins). And one pin is always a bit flaky on that thing. But the cooler

on this PC, bolts right through the motherboard, so it's a bit more

difficult for it to break loose.

Same here. I got an OCZ Vendetta2 with a backplate. Big monstrosity with aluminum fins with copper pipes running through them which seems to be the basic configuration of "better" air coolers.

So from what you're saying, assuming Realtemp is in the ballpark at the high end, 68 deg shouldn't particularly stress the chip?

Still, 68 is higher than what I've seen others report with an E8400, even cranked to 4 gigs. I might try reapplying the thermal paste. I've seen mixedreports on the usefulness of polishing the surfaces of the cooler and CPU backplate, the theory being if the surfaces are closer to perfectly flat you get more/closer contact and better heat transfer.

 

[Darklight]

Just got an E8400 3gig Core2 Duo CPU to run on a Gigabyte GA-P35-DS3L Rev
2.0 board.

I've been fooling with overclocking, currently have it running at 4.0 gig.
Using an OCZ Vendetta 2 cpu cooler.

I've seen it go as high as 68deg during gameplay. Is this high or what
you'd expect?

Per Realtemp the lowest it will register is 38 deg at idle. This makes me
wonder if the reading is a bit high, I see others report lower idle temps
- like 30 deg. As of yet I haven't calibrated it, need to get a
thermometer to measure the case temp.

Thanks for all input.

go here

http://www.cpu-world.com/CPUs/Core_2/Intel-
Core%202%20Duo%20E8400%20EU80570PJ0806M%20-%20AT80570PJ0806M%20%28BX80570E8400%29.html

 

[Doc]

I think I've been running the cpu voltage higher than it needs to be. It's my first attempt at OC'ing and I went by what one particular site was posting - 1.4v for stable running at 4.0 gi but I think I can back off from that.. Seems okay so far at 1.3v and the temps are lower. I'll try even lower and run a Prime95 test.

 

[Paul]

I think I've been running the cpu voltage higher than it needs to be. It's my first attempt at OC'ing and I went by what one particular site was posting - 1.4v for stable running at 4.0 gi but I think I can back off from that. Seems okay so far at 1.3v and the temps are lower. I'll try even lower and run a Prime95 test.

Yes, dropping the voltage and checking with Prime95, are
good things to try. For my E8400, I think the recommendation
was to not use more than 1.4V long term. So cranking Vcore
down a bit and doing stability testing, is a good idea.

The "normal" overclock sequence, is to work in a stairstep
way with voltage and clock speed. Start with stock voltage,
disable Intel SpeedStep (on my Asus motherboard, also disable
C1E). Check that the clock stays at 3GHz, even when the system
is idle. You want the thing to run at a constant condition,
while you're doing experimental runs.

Now, increase the CPU clock. Use small increases in BCLK.
At some point, it'll crash after ten minutes or so. You
can use a Linux LiveCD, rather than your Windows hard drive,
to test that. If a Linux LiveCD crashes, you're less likely
to corrupt a Windows hard drive. You can run a Linux LiveCD
with no hard drive connected, if you want.

Then, increase the voltage a notch, and retest at the
same clock speed. Now, you'll find you can increase the
clock 5, 10, 15MHz more on BCLK, before it becomes unstable
again.

You end up with a stairstep graph, of voltage and frequency.
The slope of the line, allows you to roughly predict how
much voltage, to get to a particular speed. And that
allows you to select a (clock, voltage) operating point,
which uses minimal voltage for stability.

This example is *not* a CPU chart, but it's meant to show someone
who is preparing their stairstep graph, to help predict
a good voltage/speed combo. You can see a "trend line"
here, and the trend helps you predict whether you're
going to be stable or not. You want to select final
conditions, which are a "dot or two" below the line.
And then, Prime95 should run for hours, with no errors showing.
If you leave the computer at the "ten minute stable" set of
conditions, that's not really useful for normal everyday
usage (crashing your Windows every ten minutes, you
couldn't get much work done).

http://tpucdn.com/reviews/NVIDIA/GeForce_GTX_680/images/clock_vs_voltage.gif

The trend line on a CPU, doesn't have to be linear like that.
Each CPU is different. Some older CPUs would exhibit a
"wall" behavior, where at a certain clock, you could
add more and more VCore, and no additional clock
headroom resulted. Not every CPU you buy, is a "winner".
If your graph shows a "wall" forming, well, you would
be advised to avoid the wall portion of the graph for
long term usage. As it's not going to be very stable
there. (If the CPU ages a bit, you'd need to re-adjust
things.)

Paul