Intel to disclose the maximum Tjunction values for 45nm Core 2s

Oh, and here's the presentation. http://intel.wingateweb.com/US08/pub...TS001_100r.pdf

Originally Posted by msabo

Here's the summary. Only 45nm processors not 65nm.

Desktop TJ For Dual and Quad Core CPUs
45nm Desktop Dual-Core Processors TJ
• Intel® Core™2 Duo processor E8000 and E7000 series 100°C

45 nm Desktop Quad-Core Processors
• Intel® Core™2 Quad processor Q9000 and Q8000 series 100°C
• Intel® Core™2 Extreme processor QX9650 95°C
• Intel® Core™2 Extreme processor QX9770 85°C

Doesn't Realtemp use 90C? And Coretemp 100?

So Coretemp was right all along?

Or is it 95 and 105 and they split the difference - I'm not at home to check right now.

Originally Posted by Forceman

Doesn't Realtemp use 90C? And Coretemp 100?

So Coretemp was right all along?

Or is it 95 and 105 and they split the difference - I'm not at home to check right now.

Real Temp was 95 for my Q9450, I guess I'll be nudging it up to 100.

I had reltemp set to 105C for my E8500! I also had to use -5 C offset with Speedfan 4.35 final. The E8500 is even cooler chip than I thought.

so the question is now with the 7000 and 8000's is

this is just the fan throttle up point right?
so do that mean 80 to 90 C is totally fine to be running at?

Originally Posted by Elios

so the question is now with the 7000 and 8000's is

this is just the fan throttle up point right?
so do that mean 80 to 90 C is totally fine to be running at?

Actually, the official TJmax value will enable calculation of core temp correctly.

Actually, the official TJmax value will enable calculation of core temp correctly.

Actually it does not.

Carefully read slide 7 ( Sensor Calibration) and think about what it is saying in regards to picking a Tj for your monitoring software. Hint, its the bit about "Each device is individually calibrated" so if you want you can grab the numbers from the slide 13 and run away all happy and everything will be fine, but you still dont understand what is really going on. Delta to Tjuncion is still the only really meaningfull number until Nehalem. Why do I say that ? Read on.

This is interesting, Nehalem will have a software register were we can read the factory calibrated temp where PROCHOT is asserted (typically called Tj Max). From slide 14:

Software visible register contains the target TJ
–A new feature in the Intel®Core™i7 processor is a software readable field in the IA32_TEMPERATURE_TARGET register that contains the minimum temperature at which PROCHOT# will be asserted. The PROCHOT# activation temperature is calibrated on a part-by-part basis and normal factory variation may result in the actual activation temperature being higher than the value listedin the register. PROCHOT# activation temperatures may change based on processor stepping, frequency or manufacturing efficiencies.


"Note again that this value (like slide 7) is adjusted on a part by part basis for each individual chip and thus any one generic Tjmax value, while being close enough, is still not the actual value to use. But on Nehalem at least we can see what they put in at the factory. (Keeping in mind that all Tjmax means it that the CPU thinks its hot and signals the fans to speed up 100%.

It would have been very interesting if they had given values on the true Tj catastropic trip point at which temp the cpu will shutdown.

Franly little new here on existing CPUs other than to throw out some target numbers for the 45nm chips that by their own data/slides are not absolute values anyway due to factory calibration. eh. its something and will make a lot people happy, I am not impressed of a rehash of info found in old Intel tech journals and datasheets. The Nehalem tidbit is the real nugget and UncleWebb gets to rewite his excellent software yet again once they tell us the register address.

Actually slide 8 is of far bigger concern to me. It says, “Slope error overshadows calibration error at lower temperatures”. This basically means that in the normal operating range, the DTS is simply not accurate!!! So what it’s saying loud and clear is, Don’t use DTS data to calculate temperature! DTS should be used for fan speed control & thermal solution failure detection only.

Just look at the “slope induced error” in your normal operating range on the graph. The slope induced error issue dominates this problem space. Screw it, I’m sticking with Tcase.

there is still the question of what is a "safe" value to run at under full load
if the Tj is just when the CPU sends the PROCHOT#
then is any value under it ok to run at?

and 100c seems really high to me to just be were your throttling up the fan imo
i really would never like to see a CPU get over 70c and 80c as top end limit with 100c maybe being the CPU throttle back temp

Originally Posted by Elios

there is still the question of what is a "safe" value to run at under full load
if the Tj is just when the CPU sends the PROCHOT#
then is any value under it ok to run at?

and 100c seems really high to me to just be were your throttling up the fan imo
i really would never like to see a CPU get over 70c and 80c as top end limit with 100c maybe being the CPU throttle back temp

I suggest you use the thermal spec for your processor if you are looking at "safe" operating temperatures. My Q9450 has a thermal spec of 71.4°C, as measured by Tcase. So I keep it under 65C, to allow for Tcase inaccuracies. And to BillParrish point, you might also keep an eye on Delta to Tjuncion if you are doing some serious stress testing.

The 71.4°C refers to the Tcase temperature at maximum TDP. Delta to tjunction max gives you the amount of headroom you have till the onboard safety features determine the processor to be "too warm" and starts measures to reign it back in before shutting off.

Originally Posted by msabo

I suggest you use the thermal spec for your processor if you are looking at "safe" operating temperatures. My Q9450 has a thermal spec of 71.4°C, as measured by Tcase. So I keep it under 65C, to allow for Tcase inaccuracies. And to BillParrish point, you might also keep an eye on Delta to Tjuncion if you are doing some serious stress testing.

so lets say that my CPU is the same 71.5C
how do you read the Tcase if the only way to get a temp out of the CPU is from the Tjmax delta?

Originally Posted by Elios

so lets say that my CPU is the same 71.5C
how do you read the Tcase if the only way to get a temp out of the CPU is from the Tjmax delta?

Tcase is a seperate sensor.

then how do i read that one?

Originally Posted by msabo

Tcase is a seperate sensor.

No, it's not. It is the temperature measured at the top center of the heatspreader. There is no way to measure it directly without modding your chip, which is why everyone uses Tj.

Originally Posted by msabo

Actually slide 8 is of far bigger concern to me. It says, “Slope error overshadows calibration error at lower temperatures”. This basically means that in the normal operating range, the DTS is simply not accurate!!! So what it’s saying loud and clear is, Don’t use DTS data to calculate temperature! DTS should be used for fan speed control & thermal solution failure detection only.

Just look at the “slope induced error” in your normal operating range on the graph. The slope induced error issue dominates this problem space. Screw it, I’m sticking with Tcase.

I agree. I have a slightly different view of what the impact is but do not disagree with you at all. To me, notice how the error gets smaller as you approach Tj (god I wish they would start calling that Tprochot). The major thing I get out of it is that using the DTS to determine/say/brag/whatever your idle temps are is pure shite. How much shite Intel does not bother to tell us. Probably at least a couple of degs of shite.

Still, dont get me wrong, DTS regarless of its shortcomings is a hella lot better than using artic adhesive and glueing a thermistor to the side of the IHS. I am just not happy unless I got something to bitch about

Originally Posted by Forceman

No, it's not. It is the temperature measured at the top center of the heatspreader. There is no way to measure it directly without modding your chip, which is why everyone uses Tj.

Where did you get that idea? All Core 2 Quad and Core 2 Duo processors have two different types of temperature sensors. One is a CPU case thermal diode which is located within the CPU die between the cores. The case thermal diode is used to measure the temperature of the CPU case (Tcase). Tcase is what utilities, such as Everest, report as the CPU temperature. You don't need to mod your chip to get Tcase.

The second temperature sensors are the Digital Thermal Sensors (DTS) located within each Core. The DTS measures distance to PROCHOT activation. Utilities such as Real Temp and Core Temp attempt to use DTS to derive core temperatures.

Originally Posted by msabo

Where did you get that idea? All Core 2 Quad and Core 2 Duo processors have two different types of temperature sensors. One is a CPU case thermal diode which is located within the CPU die between the cores. The case thermal diode is used to measure the temperature of the CPU case (Tcase). Tcase is what utilities, such as Everest, report as the CPU temperature. You don't need to mod your chip to get Tcase.

The second temperature sensors are the Digital Thermal Sensors (DTS) located within each Core. The DTS measures distance to PROCHOT activation. Utilities such as Real Temp and Core Temp attempt to use DTS to derive core temperatures.

That is completely wrong. There is only one sensor on a processor, and that is the DTS located inside the core itself. The temperature Intel refers to as Tcase is measured external to the processor on the top center of the integrated heat spreader. I'd find the Intel reference document that specifies this, but that would require searching, when I'm quite sure that BillParrish can quote that baby out by heart, given all the times he's done it on these forums. The temp reported as CPU temp by Everest, et al, is probably a sensor located somewhere on the motherboard. But there is unquestionably only one sensor on the CPU chip itself.

Okay, I postponed lunch for a minute to locate this - it is specifically about the E7500 but the section that matters is in all of them.

http://www.intel.com/Assets/PDF/desi...501-298647.pdf

Read Section 5 of this document, it is pretty clear about how to measure Tcase. I'll quote the relevant parts:

"To ensure functionality and reliability, the Tcase of the MCH must be maintained at or below the maximum temperature specifications as noted in Table 1 and Table 2. The surface temperature at the geometric center of the die corresponds to Tcase. Measuring Tcase requires special care to ensure an accurate temperature measurement."

It then lays out procedures for placing the thermocouple to measure this, which involves, according to their procedures, milling a hole in the heatsink.

http://www.intel.com/design/processo...hts/315592.htm
Section 5?
http://www.intel.com/design/processo...nex/315594.htm
Appendix D?

Unless I'm misreading those.

Would have been nice to get the 65nm values too