Brisbane is broken (possibly)....

C

ClariorHincHonos

[H]ard|Gawd
Joined
Jun 25, 2007
Messages
1,972
After experiencing some really odd temps with my two 3600 Brisbane procs I went searching for some answers. While it's not confirmed I'd say the conclusion this guy has come to is pretty damn accurate.

http://forums.amd.com/forum/messageview.cfm?catid=22&threadid=85837&enterthread=y

Thank you to all who have made suggestions. All the the programs I have tried can be lumped into two categories:

1) Those that read the voltage drop across the internal silicon diode apply equations to that voltage that do not work for the Brisbane socket AM2 CPU. All of them show temperatures below the ambient temperature inside the case. Right now, for example, ASUS PC Probe II shows 22C.

2) Those that read the internal digital register (Digital Temperature Sensor) read 3C to 6C, obviously not correct with a mother board temperature of about 37C.

The bottom line is that temperature sensing in the AMD Athlon X2 Brisbane (socket AM2) series of chips is broken. AMD's own technical staff are not permitted to confirm that, but they also won't deny it. They have directed me to the same programs suggested here, none of which display reasonable answers.

I'm still waiting for news from AMD, or from anyone here who has a solution for that specific CPU family.
Click to expand...

Now my temps aren't off that much anymore after Abit released an updated BIOS-but another user pointed out to me the other day that 53*C(load) on a 65nm 3600 OC'd to 2.47 is a little high, especially when using a Freezer 64 Pro. I starting thinking about that and checked the mating and even went as far as to clean and reapply my AS5. Still high. I'm thinking there's been a "rough" calculation developed for the temperature readings, and that's what is being used. Or maybe I'm crazy? Opinions?
 
Get an external calibrated temp sensor and place it as close to the CPU as possible.
You could still be out a fair bit due to lack of direct contact with the die, but at least you will know the ballpark your temp readings should be.
Test for a range of temperatures to see the differences and you will get an idea of how the reported temps compare to real life.

To calibrate the results better....
To get a better idea of the difference between the die temps and the reported temp from an external sensor, try the external sensor on a PC that is known to report approximately the correct temps (preferably with the same heatsink and paste).
You should be able to see how much temp drop there is from the die to where the external temp sensor is placed. You can then use this as an approximate temp to add to your final results on your PC that is playing up.
BTW, the temp difference is likely to increase with higher temps and decrease with lower temps so you might want to get some baseline data for high and low temps.
 
Thanks-I may look into doing that. The only problem I see is that I don't have a known good AM2 proc, but I guess the baseline work could be done with any processor.
 
If they are so unreliable, then why the widespread use? I'm sure it's not 100% accurate, but I would think they are somewhat close (unless we are talking about Brisbane of course).
 
They are good for getting your own idea of whether your system is running hotter or colder than previous.
The accuracy can vary quite a bit between CPU's so its hard to put strong faith in the real world results from them.

On die sensors are good enough for general CPU tweaking as long as you know what is too hot.
More advanced tweakers might need to have better temp figures, depending what they are trying to achieve.
 
ClariorHincHonos said:
If they are so unreliable, then why the widespread use? I'm sure it's not 100% accurate, but I would think they are somewhat close (unless we are talking about Brisbane of course).
Click to expand...

they're fine for relative measurements. if it's normally 40c and then one day it's 55, you know something is up. comparing between cpus is rather pointless though :(
 
So I had a quick simple idea that while not scientific and 100% accurate gave me some idea of how off the temps are. I was looking at the Brisbane temp on my rig again and it dawned on me that I might be able to get a general idea of temp by using a regular thermometer. So I grabbed the one I have and started with the heat pipe extending out of the top of the heatsink-this area should be fairly cooled off, but with the core temp at 53-55*C (131*F)according to Everest, should still be a bit warm. It wasn't, temp was around 93.6*F. Ok, moving down the chain, I tried the heatsink base-96.4*F. Next, I tried the heat spreader side, which I could get to. This measured at 99.9*F.

So, while this is by no means scientific, I would think that if the core temp was 131*F the thermometer would have read a higher temp at all three points, especially the head spreader side. Opinions?
 
I will concur you are on the right track.
Its hard to know just how big a difference you are getting though.
With everything being uncalibrated and the thermal connection quality unknown, I couldnt wager a guess as to the real temps.

Your own judgement is your best guide. It does look like there is a big discrepancy between the reported and actual temps from what you have done but I couldnt possibly say that you have obtained any form of near accurate results. Maybe you have, but I dont know.

If what you have found is correct, its not a huge surprise.
I had an Asus motherboard for my old Athlon XP that reported anything from 6C to 20C out !!
The reported temp hardly changed at all despite massive changes in real temp.

Good luck getting a better reading :)
 
I called AMD myself about this today. They blew me off and blamed my motherboard. My CPU is, according to Everest, sitting at 16C (60F) in a 74F ambient room on air. If I put it under load, it shoots to 62C.
 
Caffeinated - I was actually the who talked to you and it sparked my interest and so I went to and started poking around with one of our FAEs and my L2 guy. Both pretty much told me the same thing. The processor's do not have anything that tells the computer "HAY I AM RUNNING AT 50 C LOL"; it all depends on the voltages and other electrical parameters from the processors which the motherboard then uses to convert to a temperature readout. When the 90 nm processors were released to replace the 130 nm processors, motherboards had the same problems as the BIOS's weren't capable of translating temperatures properly using the old equations due to the change of the die size. Same problem now with the 65 nm swap - the motherboard manufacturers needs to release a proper BIOS that would provide you with the proper temperature readout which basically means the BIOS needs to have proper conversion equations in it.

Our labs M2N32-SLI has the latest possible BIOS and it reports temperatures around the average norm even on brisbane cores.

ALSO - be aware that the temperature you see in the BIOS can be off up to 20% compared to the actual temperature. The best way to get a proper temperature would be with an infrared thermometer pointed at the thermal lid from the side of the motherboard while the computer is on.

And another interesting fact I found out - processor have 3 (i believe) pins that are dedicated to temperature conversion for the motherboard. Often times, motherboard manufacturers use only 2 or even 1 of the pin available and thus sometimes the readouts may be incorrect.



Cliffs for those who are lazy:

1. The procs are fine.
2. Motherboards are not calibrated to read 65 nm temps properly.
3. ...
4. No profit.




That said :eek:
 
Then why are there people on every forum on the web griping about this? Are all motherboards broken? Because if they are...then it isn't the motherboards (or the processors don't fully conform to the spec). Look on the AMD website for crying out loud, or use google.
 
Caffeinated said:
Then why are there people on every forum on the web griping about this? Are all motherboards broken? Because if they are...then it isn't the motherboards (or the processors don't fully conform to the spec). Look on the AMD website for crying out loud, or use google.
Click to expand...

First of all - not all motherboards have that problem - as I mentioned my ASUS M2N32-SLI DELUXE and the ECS board I have in the lab run perfectly fine and report proper temperature readouts with the latest BIOS. There isn't anything in the processor, as I mentioned, that says "HAY - THIS IS THE TEMP OF THE PROC ROFL" in a direct number to a probe in the BIOS; it all has to do with proper calculations and equations. It is POSSIBLE AMD provided the wrong conversion figures or something along those lines - I cannot speculate nor comment nor even guess on that.

I am going to email my Abit contact.
 
Look, I don't want to get into an argument with you, but I reported the issue because bitching about it on a forum apparently doesn't help (because apparently this is the first you've heard of it). If it is truly an issue with Abit, then ok, but it just seems odd that this has been going on for months with Abit, Asus, Biostar and others.

If I sound annoyed, it is because I am annoyed at the situation, not with you personally. It just seems like it is yet another situation where everyone is pointing fingers at everyone else. Abit blamed you guys, you guys are blaming the motherboard makers.

I guess it doesn't really matter, since my warranty is voided anyway because god forbid I used an after market HSF in an attempt to fix a problem that wasn't even really there, despite what my mb, CoreTemp and Everest reported.
 
I didn't say the warranty is voided because you used a third party HSF; in fact I didnt even note that in the ticket nor did I note that you overclocked. I am just stating FYI information which you should sorta catch on that if you REALLY have a problem with the proc (no boot, blow up, yada yada), you don't want to mentione either one of the above things.


ALSO - I went on and badgered our FAE a bit more (and I got things chucked at me) - the kernel development guide for our processors provides all the updated information for the motherboard manufacturers on how to design their BIOS; it is updated with information for all the BIOSes so it is up to them to look at it and interpret the information properly. Yes, there is a chance that the information is incorrect but the chances of that are EXTREMELY SUPER DUPER low because everything gets triple checked before being posted as all of our partners read that information for their kernel development.

Now when it comes to overclockig - you will most DEFINITELY get incorrect temperatures since you are changing the operating voltage of the processor. Remember - the motherboard relies on processor parameters along with predefined information to give you the proper readouts ON TOP of the fact that you should expect up to a 20% variance on your temperature readout.



I emailed my Abit guy to see what he has to say about it but I expect he'd probably say "COME AGAIN?".
 
VanFanel89 said:
I didn't say the warranty is voided because you used a third party HSF; in fact I didnt even note that in the ticket nor did I note that you overclocked. I am just stating FYI information which you should sorta catch on that if you REALLY have a problem with the proc (no boot, blow up, yada yada), you don't want to mentione either one of the above things.


ALSO - I went on and badgered our FAE a bit more (and I got things chucked at me) - the kernel development guide for our processors provides all the updated information for the motherboard manufacturers on how to design their BIOS; it is updated with information for all the BIOSes so it is up to them to look at it and interpret the information properly. Yes, there is a chance that the information is incorrect but the chances of that are EXTREMELY SUPER DUPER low because everything gets triple checked before being posted as all of our partners read that information for their kernel development.

Now when it comes to overclockig - you will most DEFINITELY get incorrect temperatures since you are changing the operating voltage of the processor. Remember - the motherboard relies on processor parameters along with predefined information to give you the proper readouts ON TOP of the fact that you should expect up to a 20% variance on your temperature readout.



I emailed my Abit guy to see what he has to say about it but I expect he'd probably say "COME AGAIN?".
Click to expand...

I have been around long enough to know that overclocking voids the warranty. At stock settings, though, the temperature readings are flaky. Take a figure (and this is an accurate stock reading) of 57 F (14C) in a 70 F room. You know that it can't possibly be less than ambient, therefore it is likely, best case scenario at least 5 or 6 degrees more (and likely at least 10). That would be 75. 57 to 75 is more than 20%. Something isn't right.

Do a search in this particular subforum for "Brisbane" and "temp". See how many threads you come up with. Go to Anandtech and do the same, and then just do a google.
 
Caffeinated said:
I have been around long enough to know that overclocking voids the warranty. At stock settings, though, the temperature readings are flaky. Take a figure (and this is an accurate stock reading) of 57 F (14C) in a 70 F room. You know that it can't possibly be less than ambient, therefore it is likely, best case scenario at least 5 or 6 degrees more (and likely at least 10). That would be 75. 57 to 75 is more than 20%. Something isn't right.

Do a search in this particular subforum for "Brisbane" and "temp". See how many threads you come up with. Go to Anandtech and do the same, and then just do a google.
Click to expand...

Like I said - I am not doubting that there is a problem; but I am confident that the nature of the problem is not rooted at the processor.
 
What would be nice is if AMD (and others) would publicly stand behind these product "bugs", maybe it would cut though some of the finger pointing in the industry. If it was dropped in the lap of the companies responsible some of these problems would never see the light of day. Never happen though.
 
VanFanel89 said:
Like I said - I am not doubting that there is a problem; but I am confident that the nature of the problem is not rooted at the processor.
Click to expand...
While I don't doubt or question the integrity or intelligence of you or your coworkers, I doubt we'd ever hear AMD confirm it was an issue-it'd be a PR nightmare, especially right now.

Back to the issue at hand-Abit released an updated BIOS for the NF-M2 nView (I even hounded them via email about it)-which is still wrong by what I've found above. My Biostar Tforce 7025-which is a new chipset which should have a fairly new BIOS-doesn't report a "CPU" temp at all. I get an "AUX" temp and my HDD temp. I've overclocked both-but not touched the voltage (which is the great thing about these chips). Both reported abnormal temps before overclocking. Now I can't talk much about Abit because I've only owned one board-but as far as Biostar goes the Tforce motherboards have always been great (this is my 4th) and I've never had an issue with temps. It's then surprising to me that they would go from excellent temp reporting to non-existent. Makes me wonder if there was a reason why.

Either way I don't think we'll see a good fix. I do wonder if it's been fixed in the BE chips, but I would assume not. With AMD moving to Barcelona soon I doubt it's a priority (if there is an issue).
 
AMD can't make motherboard OEMs write new BIOS to properly read temps. (Although it is in their best interest to do so if they want to sell more boards) I don't think there is anything wrong with the procs themselves, but rather the motherboards don't know how to read the smaller nm process CPUs without a BIOS upgrade. The BIOS revisions on these boards are trying to read the procs as if they are .90 nm procs. A lot of boards *are* reading Brisbane temps properly, including several Asus and Gigabyte boards we have used at work recently.

Abit boards in particular have been problematic. I have one of the Abit NF-M2 Nviews and it took months before Asus released a BIOS revision that properly displayed Brisbane temps. And that BIOS revision can still only be found on the Taiwanese website ... you won't get it by using AbitFlash. If these motherboards do read temps properly after a BIOS update, then isn't it logical to assume the problem is with BIOS and not the proc itself?
 
ClariorHincHonos said:
While I don't doubt or question the integrity or intelligence of you or your coworkers, I doubt we'd ever hear AMD confirm it was an issue-it'd be a PR nightmare, especially right now.
Click to expand...


Thing is - there's nothing really to cover here by AMD and it's not really AMDs fault... the only reason, in my opinion, for AMD not to publicly point it out is to help the motherboard manufacturers to save face and not really put them in the spotlight. But it would not really be something that we'd say "LOL ON DONT KNOW WTF IS GOING ON ROOFLES"...
 
VanFanel89 said:
Thing is - there's nothing really to cover here by AMD and it's not really AMDs fault... the only reason, in my opinion, for AMD not to publicly point it out is to help the motherboard manufacturers to save face and not really put them in the spotlight. But it would not really be something that we'd say "LOL ON DONT KNOW WTF IS GOING ON ROOFLES"...
Click to expand...
You never did tell us what your Abit guy said. I'd be interested in hearing that.
 
hunter33 said:
Abit can't make motherboard OEMs write new BIOS to properly read temps. (Although it is in their best interest to do so if they want to sell more boards) I don't think there is anything wrong with the procs themselves, but rather the motherboards don't know how to read the smaller nm process CPUs without a BIOS upgrade. The BIOS revisions on these boards are trying to read the procs as if they are .90 nm procs. A lot of boards *are* reading Brisbane temps properly, including several Asus and Gigabyte boards we have used at work recently.

Abit boards in particular have been problematic. I have one of the Abit NF-M2 Nviews and it took months before Asus released a BIOS revision that properly displayed Brisbane temps. And that BIOS revision can still only be found on the Taiwanese website ... you won't get it by using AbitFlash. If these motherboards do read temps properly after a BIOS update, then isn't it logical to assume the problem is with BIOS and not the proc itself?
Click to expand...
Could you clarify your post a bit? Your first paragraph talks as if Abit isn't responsible for making the BIOS themselves. Also, Abit released a BIOS (as I've mentioned before) for the nView-not Asus. Unless you used an Asus BIOS on a Abit motherboard. Basically could you explain this a bit more? In it's current form it's not making much sense to me.
 
ClariorHincHonos said:
You never did tell us what your Abit guy said. I'd be interested in hearing that.
Click to expand...

I actually completely blanked out and confused my Abit guy with a Gigabyte guy :eek: :eek: :eek:

But the gigabyte guy said he forwarded my email to the R&D team in Taipei and hopefully I'll hear from them by tomorrow, otherwise - I'll come back in 3 months and let you guys know :p
 
ClariorHincHonos said:
Could you clarify your post a bit? Your first paragraph talks as if Abit isn't responsible for making the BIOS themselves. Also, Abit released a BIOS (as I've mentioned before) for the nView-not Asus. Unless you used an Asus BIOS on a Abit motherboard. Basically could you explain this a bit more? In it's current form it's not making much sense to me.
Click to expand...

It was a typo, I meant to say AMD instead of Abit.
 
hunter33 said:
It was a typo, I meant to say AMD instead of Abit.
Click to expand...
Gotcha-that's what I thought might have happened but was not sure. Thanks for clarifying.

We're not expecting AMD to write the BIOS-I am fully aware that the motherboard manufacturers should do that. However, it seems that a good majority are having issues writing BIOS's that detect the temperature correctly. I understand the initial issues-the old stuff wasn't calibrated. But the new Abit BIOS is still a bit off, and as I mentioned the 7025 motherboard I've had for a few weeks doesn't report a "CPU" temp at all. It seems to me something is making it awfully difficult for them to get the correct temp. And if what VanFanel89 said is correct about the proper information being available to OEMs....then what's so difficult? I'm not saying programming is easy (it's damn hard), but when programmers are given the specifics from the processor manufacturer and have had experience with the architecture.......I'm a little puzzled why we still have inaccurate temps.
 
http://www.alcpu.com/CoreTemp/howitworks.html according to its own explanation, core temp does not need the mobo to interpret the temperature. i am certainly no cpu expert so i can't make any concrete conclusions, but this bug seems to be the processor and not the bios, which reads a totally different sensor.
 
You must log in or register to reply here.
Back
Top