opt33

LOL, I think grunion hit the nail on the head with having to use common sense. I would expect numerous errors in both coretemp and realtemp, except for those exact cpu batches that have been measured at IHS at undervolted, underclocked state with no heatsink in realtemps case or on those cpus where tjmax mobile is near throttling temp point of desktops in coretemps case. And that assumes intel is not changing tjmax within same batch.

But you guys with 15C difference, tjmax 100 versus 85, should be easy to find out.

Grunion suggested one method.

Another, less risky, slightly less accurate..but good enough to find 15C difference. Of course you know it may be neither 100 or 85, and actually in between. Calibrate your cpu temp using speedfan, with cpu in undervolted, underclocked state against internal ambients. Then use load like blend, while on large ffts (not large in place as that will be high load)...then compare cpu to core temps, see what makes sense.

You can compare temps at ambients, but on 45nms, die sensors do not scale linearly, not accurate until reaching load temps, dont know how much better 65nm is

GigaByte

I did that, it seems like that would work but it don't. My E4500 throttled at 97c and shutdown at 101c, as Tj max is SHUTDOWN temp not throttle. How do we know if the temp the software is showing is correct? If its using Tj max 85c it would show temps at 85c give or take 1 degree, same time if it was using 95c or 100c Tj max. If you use one of those lazer thermometers it also woulden't show what temp as its only touching the IHS (assuming you take the heatsink off) or the die if you took the IHS off. It won't be able to read the temp of the actual cores where the DTS is.

__________________

Refresh

The IHS doesn't get hot at all. Mine is currently at 6C, while the cores are at 24C

grunion

Yeah, the more I thought about it the more it made less sense

No matter the tj, 85 or 100 the delta to tj remains the same.

__________________

Megahalems > TR Ultra Extreme
How do I uninstall Video Card drivers?
How do I see my in game temps?

Retrospekt

Haha, yeah it seemed like a great idea at first, but the experiment used what we were testing in it. If that makes sense... lol

__________________

opt33

Whether you calibrate core temps at ambient (difficult from nonlinear/inaccurate readings of many DTS sensors at low temps) or at load, or at DTS=0, you obviously have to measure/approximate absolute temps.

Either use cpu temp calibrated to ambient or take off heatsink and measure temp of IHS. When cpu is idle IHS/cpu temp should approximate core temp within 1C, I have linked to several articles showing this, but apparently no one has read them...so here is the highlights. (Grunion is correct in using undervolted, underclocked state with low/no cooling, to allow idle speed (<1C gradient) to take you to temp range where DTS sensors are most accurate)

First pic...see graph below from intel. Each data point represents a "popular program/benchmark" intel tested at full load, steady state. At idle, the gradient is near zero. ~15% of those programs/benchmarks at load, have a near zero, less than 0.1C gradient. 1/3 at load have gradient less than 1C. Highest gradient is 5.2C at TAT load.
http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf

Second pic...this is intel testing temps at idle loads, and note gradient tj4 versus tcase...0.1C gradient from tj across the molded cap IHS.
http://www.flomerics.com/flotherm/te...apers/t324.pdf

Independent verification from someone who drilled a hole to the die, placed a calibrated thermocouple on die, then checked it against another calibrated diode on casing...0.4C difference, same as what intel is telling you.
http://www.overclockers.com/tips443/

Only at high loads, especially TAT, encoding, etc., does a significant gradient occur. This 5C or 10C or 15C gradient theory that people have been brainwashed into believing must always exist, is because people have believed that tjmax was accurate, and then worked backwards with circular reasoning to prove gradient using that tjmax, when tjmax was in fact inaccurate. Very high gradients only occur during increasing load, for example first couple seconds after pushing start button on orthos, not at steady state, and certainly not at idle loads.

Attached Thumbnails

   

TheSubtleKnife

I ran orthos small FFT's with my fan unplugged, and this is what i got:

Looks like the TJmax for M0's are around 95, because thats when i got the prochot notice, which is tripped when the temp hits the TJmax, from what I understand...

__________________

Retrospekt

So RealTemp was actually right with your tjmax? I'll have to try this when I get home...

__________________

alwaysAMD

You are very brave, TheSubtleKnife.

__________________

opt33

You forgot one important thing, but first...

You are correct in your assumption that at DTS=0, TCC (throttling) occurs, as is illustrated on orthos mhz. Shutdown occurs at 20-25C beyond that point, near 120C, see quotes below and source.

"Readings from the DTS are relative to the activation of the
TCC. The DTS value where TCC activation occurs is 0 (zero)."

"In the event of a catastrophic cooling failure, the processor will automatically shut down when the silicon temperature has exceeded the TCC activation temperature by approximately 20 to 25 °C."
http://download.intel.com/design/pro...x/31873401.pdf

But what you forgot was a method to measure absolute temp while you tested. Had you been using coretemp, it would have displayed temp at 105C when throttling occurred, and you would have concluded that coretemp was correct. When you mess around with tjmax, setting it higher or lower, it does not alter the throttling point, it only alters the temp that is displayed when the throttling occurs. You could run coretemp and speedfan and realtemp at same time, do it again, and you would see at throttling, coretemp reads 105C, speedfan reads 100C, and realtemp reads 95C, and all 3 will accurately read DTS of 0 at that point.

Instead, if you do not have temp probe, thermocouple, or IR gun, then you need to calibrate your cpu temp on speedfan so it equals internal case ambients (ambients +1-2C depending on internal case cooling) when your cpu is undervolted/underclocked.

1)set your cpu at 1.15 or 1.2 volts (less for 45nm), 6x200 mhz, and boot windows, and keep at idle.

2) check cpu temp, and if reads 20 and ambients are 25, internal ambients 26, then you need to always add 6C to cpu, speedfan can do this, or just remember the difference.

3)Then you really dont need to run temp all the way to DTS=0, just run up to DTS=30 or so (change coretemp settings to read distance from tjmax, ie DTS). This needs to be done at idle to prevent any significant gradient, with heatsink off or at least with fan off (otherwise temps wont reach as high as necessary at idle). Dont use orthos, as at load you may create a gradient from core to cpu. If cpu temp is 65, when DTS=30, tjmax is 95. But basicically just have to add DTS displayed (accurate regardless of what program you use) to cpu (calibrated temp) at idle. This will not work at low temps, as DTS sensors can be inaccurate at low temps, so need to get DTS up to around 35 to 25.