Max Tjunction secret revealed soon

[bryanW1995]

http://www.anandtech.com/weblog/showpost.aspx?i=478

for those of you who don't know what this is about, all current cpu core temperature readings for core2-based processors are a complete guess. intel has never publicly disclosed tjmax for some stupid reason, vexing many of us for a very long time now. thank god, no more setting real temp to display distance to tjunction!!!

edit: thanks jaredspace, the blog has been updated with the 45nm tjmax specifications:

Update (8.21.2008):

45nm Desktop Dual-Core Processors
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


edit #2: unclewebb has some very insightful comments over at XS on this issue. The way I read this now, even at 1c below tjmax there is error being introduced into the system, and every sensor is off to a different degree. I hate doing it, but I might actually go through the hassle of calibrating my cpu now...:| Anyway, here is the link:

http://www.xtremesystems.org/f...d.php?t=179044&page=84

 

[myocardia]

So, Lord Poopy Pants has finally decided to give the whole truth, huh? As far as I'm concerned, he can go crawl back under his rock. Interesting? Yeah, but anytime something arrives 18 months too late, it might as well never have arrived. At least that's the way I see it.

 

[Idontcare]

It's nice of them to do this finally but it really is worth scrutinizing the background of the situation regarding why this info has been withheld to-date.

 

[Lonyo]

Originally posted by: Idontcare
It's nice of them to do this finally but it really is worth scrutinizing the background of the situation regarding why this info has been withheld to-date.

Yes. To try and work out whether this sort of thing will happen in the future as well, and whether it might happen with relation to other things.
If something has been done which arguably causes issue (even if only for a minority of users), it's worth trying to work out why to try and work out how it may impact things in the future.

 

[QuixoticOne]

It is nice that they're finally doing this, but I give them a raspberry for doing it WAY too late.

Also I find it fairly pathetic that a $5 timex can tell time better than my $1000 PC (in the hardware RTC),
and that a $2 thermometer can tell temperature better as well.

Would it have killed them to but an accurate (correlated to reality) clock and temperature sensors in these $200 chips?
They should also have standardized on adding SMBUS / I2C headers on the motherboard so one can install more thermal monitors where desired.

 

[Idontcare]

Originally posted by: QuixoticOne
Would it have killed them to but an accurate (correlated to reality) clock and temperature sensors in these $200 chips?
They should also have standardized on adding SMBUS / I2C headers on the motherboard so one can install more thermal monitors where desired.

Not disagreeing with the validity of your requests, I wish the very same as well, but the reasoning why it hasn't been done is probably straightforward economics.

Even if it only increased the production cost of a $200 CPU by $2 to add in a satisfactory temperature sensor that action must be justified by a market price reaction (i.e. will customers pay an extra $2-$4 for that temp sensor?) otherwise the net result is Intel management is being wasteful of their shareholders equity (increasing production costs without increasing revenue or profits).

It's easy to say they make sooo much money already so why can't they just throw us a bone and incorporate a better thermal sensor, but someone somewhere in Intel with that decision making authority is ultimately going to have to justify that cost increase activity to the shareholders (or the Intel executives who must make such justifications, etc etc) and who the heck wants to put that line item on their resume and lose their chances at an annual bonus?

Intel Boss: So Ted, what's your #1 revenue impact project this past year?

Ted: I executed a project that led to the inclusion of a more precise and accurate temperature sensor, raising our production cost by a mere $2/CPU and delighting thousands of enthusiasts worldwide!

Intel Boss: And did this cost increase activity result in a commensurate or better increase in ASP's?

Ted: Uh, well that wasn't exactly what the delighted enthusiasts were looking for, they wanted it for free in addition to the planned price-cuts we discussed last year...

Intel Boss: Ted, your fired, but we'll give you a $5k signing bonus if you convince AMD to hire you and put your brilliance to work competing against us :thumbsup:

 

[Tempered81]

Wonder what the actual TJMax is going to be for E8400?

95 or 105?

I'll bet it's much more complicated than this and DTS reading of Tcase & distance to TJmax has more to do with OC & temp readings.

 

[geokilla]

BETTER LATE THAN NEVER!

I'd be surprised if Intel releases something like the TJMax can be 140C. We'll probably see half the people on the forums here raising their vcore a notch or two just to get a bit more performance =P.

Question. What's the difference between the DTS on 45nm and 65nm processors? Why won't they release the TJMax for 65nm processors as well?

 

[Tempered81]

because it's generally accepted that it's 105C on 65nm.

 

[Idontcare]

Originally posted by: geokilla
Question. What's the difference between the DTS on 45nm and 65nm processors? Why won't they release the TJMax for 65nm processors as well?

http://www.patentstorm.us/patents/7228508/fulltext.html

http://www.patentgenius.com/patent/7228508.html

Well it appears the IC hardware involved in creating the DTS is a bipolar transistor and some circuitry designed to measure voltage/current output from the bipolar xtor which is computed to yield a band-gap result which has already been correlated to temperature.

So given that the xtor materials of construction are vastly different between 65nm (doped poly-si for gate, nitrogen doped silicon dioxide for gate oxide) and 45nm (bi-layer metal gate stacks, bi-layer hi-k for gate oxide) it would not be too surprising if it turns out to have just taken Intel a bit longer to figure out a robust bipolar xtor based thermal sensor characteristics.

I have no idea if the material changes threw a major monkey-wrench into their existing DTS working knowledge, but at the same time I can't imagine a bigger change than what 45nm brought to the table to cause a delay in them mastering their own DTS hardware. Would be nice if an Intel guy who works on DTS could comment, but then again that's how you lose your job so I doubt we'll know the story for a few more years yet.

 

[Tempered81]

Idontcare, maybe the new materials could be the reason some sensors were borked upon 45nm release. They never experienced stuck/broken sensors with 65nm generation, correct?

 

[VirtualLarry]

I never heard anything here about stuck thermal sensors on 65nm.

 

[aigomorla]

Originally posted by: Idontcare
Intel Boss: So Ted, what's your #1 revenue impact project this past year?

Ted: I executed a project that led to the inclusion of a more precise and accurate temperature sensor, raising our production cost by a mere $2/CPU and delighting thousands of enthusiasts worldwide!

Intel Boss: And did this cost increase activity result in a commensurate or better increase in ASP's?

Ted: Uh, well that wasn't exactly what the delighted enthusiasts were looking for, they wanted it for free in addition to the planned price-cuts we discussed last year...

Intel Boss: Ted, your fired, but we'll give you a $5k signing bonus if you convince AMD to hire you and put your brilliance to work competing against us :thumbsup:

IDC you dont know how much i was cracking up on this one.

incase you guys are wondering IDC did work for a chip maker.

so his statement has some truths. rofl!

 

[Idontcare]

Originally posted by: jaredpace
Idontcare, maybe the new materials could be the reason some sensors were borked upon 45nm release. They never experienced stuck/broken sensors with 65nm generation, correct?

Yep, it would be my number one suspect.

Kinda like how all those early Pentium4's had hyperthreading in them but they were just ever so slightly borked so they kept it disabled until a later mask respin in which the logic was finally fixed. Was just low enough priority that it didn't hold up the initial chips.

So too with DTS on 45nm I suppose as they could sell desktop chips with borked DTS and not have it be a problem until they wanted to sell mobile parts that need a robust DTS implementation. (just my speculation, I have no info to know for certain)

 

[KingstonU]

Originally posted by: Idontcare

Intel Boss: So Ted, what's your #1 revenue impact project this past year?

Ted: I executed a project that led to the inclusion of a more precise and accurate temperature sensor, raising our production cost by a mere $2/CPU and delighting thousands of enthusiasts worldwide!

Intel Boss: And did this cost increase activity result in a commensurate or better increase in ASP's?

Ted: Uh, well that wasn't exactly what the delighted enthusiasts were looking for, they wanted it for free in addition to the planned price-cuts we discussed last year...

Intel Boss: Ted, your fired, but we'll give you a $5k signing bonus if you convince AMD to hire you and put your brilliance to work competing against us :thumbsup:

This post made my day, merci buckets

 

[Tempered81]

45nm Desktop Dual-Core Processors
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

So realtemp & Coretemp are wrong.

If realtemp says 50C and coretemp says 60C your actual CPU temp (e8400's at least) was 55C.

haha

 

[VirtualLarry]

So what is tjmax for 65nm? I think 100C is correct too, but i'd like to hear it officially.

 

[VirtualLarry]

So what is tjmax for 65nm? I think 100C is correct too, but i'd like to hear it officially.

 

[geokilla]

Aw man, it's not 140C? That sucks =(

Where'd you get the info? I can't find it on the AnandTech page.

 

[kpo6969]

Originally posted by: VirtualLarry
So what is tjmax for 65nm? I think 100C is correct too, but i'd like to hear it officially.

http://www.xtremesystems.org/f...3236208&postcount=2078

http://www.xtremesystems.org/f...3236637&postcount=2089

 

[Idontcare]

Originally posted by: geokilla
Where'd you get the info? I can't find it on the AnandTech page.

See page 13.

http://intel.wingateweb.com/US.../SF08_TMTS001_100r.pdf

 

[bryanW1995]

for anybody interested in this topic, unclewebb is going to town right now at xs:


http://www.xtremesystems.org/f...?p=3236208#post3236208


****ing intel!!!

 

[nevbie]

Haha, the Extreme processors are made to look cooler than the standard variants. Probably makes buyers of the Extremes happy when they see lower temperatures in monitoring programs than people with non-Extremes.

..actually, if someone has both an Extreme and a locked version of the same physical processor type, he/she could use the exact same conditions.. same other parts, same cooler, as similar cooler install as possible and same multi&FSB&vcore&other settings and check if the distance to maximum temp has the difference it should, between the processor models. Even the twistedness of the temperature scale could be measured, but that would need more samples and too much work.

 

[error8]

So my E7200 hits 70 C instead of 65 C that RealTemp used to show me. Well that is scarry. Oh well, let's hope it's not going to die on me.

 

[Idontcare]

Originally posted by: nevbie
Haha, the Extreme processors are made to look cooler than the standard variants. Probably makes buyers of the Extremes happy when they see lower temperatures in monitoring programs than people with non-Extremes.

..actually, if someone has both an Extreme and a locked version of the same physical processor type, he/she could use the exact same conditions.. same other parts, same cooler, as similar cooler install as possible and same multi&FSB&vcore&other settings and check if the distance to maximum temp has the difference it should, between the processor models. Even the twistedness of the temperature scale could be measured, but that would need more samples and too much work.

That's an unfortunate viewpoint that can be taken here, but that is a "tail wagging the dog" viewpoint.

Yes an unnaccounted for low Tj does make the QX's appear to be running cooler than they probably are with the currently incorrectly calibrated programs, but that was hardly Intel's intentions when they elected to make the Tj-max for QX's so much lower than the non-QX chips.

The story here is that the these QX's shutdown (thermal throttle) at much lower temperatures than the non-extreme quads.

Meaning you really must be on top of your cooling system in order to avoid thermal throttling with these QX's. In theory a Q9650 will continue to operate at 96C while a QX9650 will thermal throttle at 96C.

Not that anyone runs their CPU's at 96C but it begs the question why did Intel lower the Tj-max on these extreme quads? Are they unstable at these higher temps whereas the non-QX chips are stable? (for their speedbin)

And what do we call a CPU that is stable at X.X GHz but only if the temps remain under a threshold value that is lower than other chips maximum operating temp? I usually recognize those chips as being "overclocked"...sooooo oh my, I guess Intel's QX chips are "factory overclocked"...stable at the speedbin only if they thermal throttle before hitting the higher operating temps that the non-QX chips can stably operate at.

edit: forgot to add my conclusion - so in conclusion I believe the reason it took Intel this long to publish what they knew >1yr ago is that this data betrays the fact that Intel's QX's are factory overclocked by way of manipulating the Tj-max value.

I suppose they could release a QX9950 at 4 GHz with a Tj-max of 60C, but that wouldn't exactly encourage an enthusiast to spend $1k on it knowing the TJmax was so low and thermal throttling was always right around the corner. Definitely explains why everyone had considerable issue with the QX9770 (Tj-max is another 10C lower) when the reviews first came out, now I understand why Intel was so silent on the topic.