If you disable Intel TM, will your CPU melt down if cooling fails?

[Habeed]

On my ASUS sabertooth motherboard there is a BIOS option called "Intel TM" which is short for "thermal monitor". Like most overclockers, I disable it. I know this feature harms overclocks because it throttle the CPU in order to stay below TDP, or 130W used.

One thing I am curious about : with this feature off, if I were to run the cpu without a heatsink, would the CPU shut itself down (or throttle to the extreme) to prevent a meltdown, or would it melt itself into a flaming puddle like those dramatic videos of older AMD cpus a few years ago?

That is, are there other layers of thermal protection to prevent a catastrophic failure that can't be disabled by software?

I know that in the same heatsink-less test a few years back, the Intel chip did fine without failing. I believe it even booted up without a heatsink, albeit with greatly reduced performance.

 

[VirtualLarry]

It will melt down. Don't disable TM. Btw, it doesn't limit you to TDP, that's the job of the PCU, and BIOS controls how that works.

 

[taltamir]

On my ASUS sabertooth motherboard there is a BIOS option called "Intel TM" which is short for "thermal monitor". Like most overclockers, I disable it. I know this feature harms overclocks because it throttle the CPU in order to stay below TDP, or 130W used.

uh, it does not HARM overclocking... it makes overclocking SAFE... it has absolutely no negative impact on your OC and should never be disabled. if your OC actually pushes such high temps then it will damage the CPU and then your OC (and system) will fail

that being said, I am pretty sure its not called intel TM.

Oh, and I Think you heard that you are supposed to disable the intel POWER SAVING feature that throttles your CPU WHEN IDLE... not the feature that throttles it when its about to melt.

are there other layers of thermal protection to prevent a catastrophic failure that can't be disabled by software?

no, why would there need to be?

 

[PlasmaBomb]

uh, it does not HARM overclocking... it makes overclocking SAFE... it has ahttp://forums.anandtech.com/newreply.php?do=newreply&p=30709890bsolutely no negative impact on your OC and should never be disabled. if your OC actually pushes such high temps then it will damage the CPU and then your OC (and system) will fail

that being said, I am pretty sure its not called Intel TM.

It certainly was back in the P4 days....

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

Automatic thermal monitoring mechanism #1 (TM1)
Automatic thermal monitoring mechanism #2 (TM2)

Although I agree with you, turning them off is stupid... especially since control features should be more advanced now.

Oh, and I Think you heard that you are supposed to disable the intel POWER SAVING feature that throttles your CPU WHEN IDLE... not the feature that throttles it when its about to melt.

That's Intel SpeedStep, check for that in your BIOS Habeed, and play about with turning it off/on to see if it effects the overclock.

are there other layers of thermal protection to prevent a catastrophic failure that can't be disabled by software?

no, why would there need to be?

Actually there is a third system which controlls your system... Emergency overheating detector, which basically shuts the machine down if the CPU reaches ~125 °C IIRC. Huge risk of losing whatever you were working on though, and repeatedly hitting 125 °C is going to really shorten your CPU life...

PB

 

[taltamir]

It certainly was back in the P4 days....

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

Automatic thermal monitoring mechanism #1 (TM1)
Automatic thermal monitoring mechanism #2 (TM2)

Although I agree with you, turning them off is stupid... especially since control features should be more advanced now.

according to the article you linked the P4 mechanism in question COMBINES the "throttle on idle to save power" and "throttle when too hot". In modern processors the two are separate.

 

[Absolution75]

You'd likely notice a bluescreen before anything bad could happen and subsequently wouldn't be able to boot if your CPU was so hot it was burning.

 

[Ben90]

6.2.3 THERMTRIP# Signal

Regardless of whether or not Adaptive Thermal Monitor is enabled, in the event of a catastrophic cooling failure, the processor will automatically shut down when the silicon has reached an elevated temperature (refer to the THERMTRIP# definition in Table 5-1). THERMTRIP# activation is independent of processor activity. The temperature at which THERMTRIP# asserts is not user configurable and is not software visible.

Assertion of THERMTRIP# (Thermal Trip) indicated the processor junction temperature has reached a level beyond which permanent silicon damage may occur. Measurement of the temperature is accomplished through an internal thermal sensor. Upon assertion of THERMTRIP#, the processor will shut off its internal clocks (thus halting program execution) in an attempt to reduce the processor juntion temperature. To further protect the processor, its core voltage (VCC, VTTA VTTD and VDDQ must be removed following the assertion of THERMTRIP#. Once activated, THERMTRIP# remains latched until RESET# is asserted. While the assertion of the RESET# signal may de-assert THERMTRIP#, if the processor junction temperature remains at or above the trip level, THERMTRIP# will again be asserted after RESET# is de-asserted.

Your processor wont melt.

*edit*
That being said, there really isn't a reason to turn it off. PROCHOT# is the equivalent to anti-lock brakes, and THERMTRIP# would best be described as airbags. Much better to avoid the accident all together than having a safe one.

 

[IntelUser2000]

That is, are there other layers of thermal protection to prevent a catastrophic failure that can't be disabled by software?

I agree with the others in this thread. It won't melt the CPU but won't harm overclocking either.

There are two technologies to protect the CPU from overheating

TM-Thermal Monitor, kicks in when your CPU hits ~85C or so. Since its bad for your CPU running that high(not to mention your system will feel randomly slow), you shouldn't be overclocking it high enough to reach those temps, or use good enough cooling so it doesn't reach that temp.

Catastrophic failure circuitry-This is one thing you cannot disable. It kicks in when the temp increase far exceeds 85C and can burn your CPU in matter of minutes. It'll shut the chip down.

 

[Ben90]

I believe the numbers on i7s are 100*C PROCHOT# and 105*C THERMTRIP#. As far as I know Intel doesn't publish the TJMax until the processor is EOL; however, consumer testing places the TJMax around 100.

 

[Rifter]

I believe the numbers on i7s are 100*C PROCHOT# and 105*C THERMTRIP#. As far as I know Intel doesn't publish the TJMax until the processor is EOL; however, consumer testing places the TJMax around 100.

I can tell you its over 93C Thats what my i7 would hit in the summer with 30-35c ambient temps. Never failed a prime burn test even after 72hours.

To the OP do not disable it, there is no point as has been mentioned.

 

[tweakboy]

well once CPU goes to 85c it shut downs computer... hmmm

 

[Habeed]

I enabled it and tested it. Worked fine without throttling. Guess I will leave it on. 85 C is really hot and if it reaches that because there's dust on the fan or something I would rather have it clock down than damage itself.

Normally I am all for "performance at any cost" but a BSOD sure drags on the performance.

 

[Seero]

It purpose isn't to prevent melt down, but to prevent BSOD. If cooling is bad enough to cause melt down, then those won't help even if it is on. If cooling was good but are degraded due to dust or bad fan, then it will throttle the CPU to keep PC alive so you will realize the overheating problem and fix it accordingly.

 

[Ben90]

Well I snooped around on Intel's site to see if they have updated any temperature information. Unfortunately they only had the useless TCASE temperatures listed for desktop variants.

Bloomfield/Gulftown 67.9 TCASE
Lynnfield 72.7 TCASE
Clarkdale 72.6 TCASE
Mobile Nehalem Quads 100 TJUNCTION
Mobile Nehalem Dual 105 TJUNCTION
Q6600 G0 71 TCASE

I believe a Q6600 G0 has a TJUNCTION of 100*C, but for the life of me cannot find that strait from Intel. The closest I can find to an official source is from the makers of RealTemp, who apparently got that number from IDF. If current day packaging retains the same tolerances and thermal properties of the Q6600 G0 it wouldn't be too far fetched to use 95/95/100*C as the TJUNCTION of Bloomfield/Gulftown/Lynnfield respectively.
I don't believe the packaging has changed, but Intel is sure to have fiddled with the tolerances. I think its safe to say the TJUNCTION is somewhere within the 95-105*C range on todays processors.

Either way, I have been using the easy button of temperature measurements for a while now. Instead of trying to relate the Digital Temperature Sensor to an equivalent temperature on our current scales, I interpret the DTS results on its own native scale: Distance to TJMAX. No matter what the latest generations unknown TJUNCTION is set to, its much easier to know your 20 units away from it instead of all the variables in trying to convert that to *C.

There are many "expert" opinions on this forum of what a safe operating temperature is for whatever processor. By using the DTJMAX scale I take advantage of the countless man hours and expensive testing equipment that Intel uses to determine safe operating parameters without any of the paranoia or overconfidence of an "experts" opinion.

I'm not saying run your processor overvolted right up against the TJUNCTION 24/7 as you may see degraded life, but to become familiar with the DTJMAX scale as it removes a lot of the uncertainty.

 

[Dufus]

Hi Ben90, most of the newer chips have Tjmax in a register in the CPU. It's really Tj-target which AFAIK should be the minimum temperature for Tjmax. For instance with the Q6600 if Tj-target was 90°C and had a tolerance of -10°C to + 5°C then the real temperature could be anywhere between 90°C to 105°C when DTS=0 probably depending on if the manufacturing process had a good day or not. Nehalem might be plus or minus 5%, so with Tjmax (target) at 100°C that could be 100°C to 110°C, no real black and white figures though unfortunately.

I don't claim to be any sort of expert, it's just my own understanding of it. Until Intel come out and publicly publish tolerances it's remains an Intel secret and Intel seem to be very good at keeping secrets.

IIRC the Realtemp Q6600 figure is empirical.

 

[Ben90]

I could have sworn Intel published the q6600's Tjmax about a year ago. I searched and searched to no avail so I guess I must have been dreaming.

In my searching I re-read some more Intel docs and found the IA32_TEMPERATURE_TARGET register Dufus mentioned. It really didn't go into much detail and I'm confused to exactly how this works. Is this value what determines the Tjmax directly or is it just the offset of the unknown value?

 

[Dufus]

Kinda both. It's not the true temperature but meant to be taken as Tjmax IIRC. As you earlier said, it's less complicated to just use the relative measurement and not worry about absolute ones. See if this helps.

Link courtesy of user sram at hardforum.com

 

[PlasmaBomb]

I could have sworn Intel published the q6600's Tjmax about a year ago. I searched and searched to no avail so I guess I must have been dreaming.

In my searching I re-read some more Intel docs and found the IA32_TEMPERATURE_TARGET register Dufus mentioned. It really didn't go into much detail and I'm confused to exactly how this works. Is this value what determines the Tjmax directly or is it just the offset of the unknown value?

I thought that Intel published all that info after a chip EOL so it should be out there somewhere...

Kinda both. It's not the true temperature but meant to be taken as Tjmax IIRC. As you earlier said, it's less complicated to just use the relative measurement and not worry about absolute ones. See if this helps.

Link courtesy of user sram at hardforum.com

Good read, and you were indeed correct with your previous statement -

PROCHOT# trip temperature will vary from part to part

Nehalem has improved the Digital Thermal Sensor circuit
– Expanded temperature range – unlikely to ‘bottom out’
– Calibration accuracy is improved
– Slope error is reduced

Future processors may report temperatures in °C

Sounds good.

 

[Habeed]

Question : running OCCT with Intel TM on, I've noticed I can get spikes to 89 Celsius or so.

It doesn't throttle at all. Not sure what Intel TM is doing...maybe it doesn't kick in until I hit TJMax?

For those of you who make fun of the temps :

I've got pretty good cooling : this is a chip running with hyperthreading on, using a Noctua DH-14 with quiet fans, at 4.095 ghz at 1.37 volts. TBH I am happy with spikes to 89 or so...average CPU temp is 85 at most between the cores. There's also 12 gigs of ram installed, all running at 1600 MHZ cas 7. (putting more load on the IMC) Remember, this is also with OCCT, which is a load far greater than any real application. And it's survived over 40 hours of testing at these temps with 0 errors. Ambient temps in the room are often a bit high as well (as high as 85 farenheit)

It idle at 37 C, and during a game it reached 60 C at worst.

 

[Dufus]

Throttling doesn't happen via the CPU until you hit Tjmax when DTS=0, providing TM is enabled. ACPI might start throttling before that too.

There is a sticky bit (stays set until software resets it) that will get set if you hit Tjmax and that would be your best indication. IIRC Realtemp shows this as "Log" under thermal status. You might also see a "HOT" as well.

 

[Seero]

Throttling doesn't happen via the CPU until you hit Tjmax when DTS=0, providing TM is enabled. ACPI might start throttling before that too.

There is a sticky bit (stays set until software resets it) that will get set if you hit Tjmax and that would be your best indication. IIRC Realtemp shows this as "Log" under thermal status. You might also see a "HOT" as well.

Your CPU will hang as soon as you hit TjMax, throttling occurs at 5 degree below TjMax.

 

[Habeed]

Ah, 5 degrees below. That would be 95 C, which I have not managed to hit yet, even on the hottest core in a hot room with bad airflow.

Earlier posters in this thread were saying 85 C, which is why I asked.

Good to know.

 

[Seero]

Ah, 5 degrees below. That would be 95 C, which I have not managed to hit yet, even on the hottest core in a hot room with bad airflow.

Earlier posters in this thread were saying 85 C, which is why I asked.

Good to know.

I'm not an expert on chip specifics, but I believe C2D TjMax = 80, C2Q = 100. Some say coretemp has wrong TjMax because c2q TjMax is actually 80, making it sure temps higher than actual.

The sensor on die works funny. Think of it as the number when it is 0 degree and it will decrease as temp increases. As it gets to 5, throttle kicks in. When it reaches 0, all power to CPU will be cut.

 

[Dufus]

Your CPU will hang as soon as you hit TjMax, throttling occurs at 5 degree below TjMax.

If throttling occurs below Tjmax it is usually due to ACPI. If your hanging when hitting Tjmax then there is something not quite right with your system. Maybe try a BIOS update.

After a bit of digging around I found this to try and explain a little further. Pic courtesy of burebista on TPU

Tjmax is 100°C with Realtemp showing the calculated temperature and DTS value. Note how on the first core the DTS value which is 7-bit (reads 0 to 127), has wrapped around 0 to show a distance of 126 (-2) from Tjmax. The thermal status is HOT because the core temperature is currently above Tjmax. The second core shows LOG because the core was at or above Tjmax at some time between powering on the CPU and taking the screen shot. The sticky bit (LOG) has been set to record that event. If and when the first core cools down below Tjmax then HOT will change to LOG.

With todays CPUs you may find that once DTS reaches 0 it no longer wraps around. In other words it will only report a maximum temperature of Tjmax even if the temperature continues to climb higher than this.

 

[Habeed]

In any case, I have not been able to get within 5 degrees of TJMax. The hottest I have ever managed to get it was 93 degrees C on core 1, which would correlate to 7 degrees from TJMax.

And actually, these numbers are probably wrong. I've read one theory that is most likely correct : different cores in an i7 CPU have different setpoints for TJMax. This prevents them from all throttling at the same time. That would explain why my "hottest" core is always, no matter what, almost exactly 8 degrees "hotter" than my "coldest" core.

Since the temps are really distances from TJ Max, and Occam's razor suggests that the temps should be the same in each core given equal workload, this means that each core's TJMax is different.

Now that I know how Intel TM actually works, I'm going to leave it on with all future overclock builds. 95 C (or 5 from TJMax) is HOOOT. There is nothing gained by letting the CPU get hotter than that. Even the most extreme overclocker will have cooling better than that. (which means TJMax is 85 C for a C2D)