Since recently purchasing a new 65nm A64 x2, I learnt that there is something about the on-die temperature sensor that makes it read incorrectly in most (all?) current software and BIOSes. Most places suggested that this is because AMD changed the way the temperature is reported, so that a simple software update could fix it. When Abit released BIOS updates for some of their motherboards (not mine yet, unfortunately) that people were reporting fixed the temperature (or at least that it reported a reasonable value, which doesn't necessarily mean the correct value), I concluded that this was indeed the reason for the wrong readings.

However, I came across a post on the CoreTemp forums that says that the sensor is actually broken. The fact that I have not found any software (CoreTemp, SpeedFan etc) that can read the temperature correctly, even if it both bypasses the BIOS and says it has been patched for Brisbane support, makes me suspicious of what's really going on.

My hope is that either it's not really broken, or if it is broken that the reported value is still a function of temperature and thus could be used to determine the actual temperature.

Does anyone here have any solid information? Has AMD said anything official? I would be a bit miffed if it turns out that I've purchased a broken CPU.

My bios reports:
CPU 65C
System 24C
PWM43C
CPU Core V 1.36V

Core Temp reports:
Core 12C complete idle, with internet surfing activity it reports 16C. voltage is 1.1

AMD CnQ Dashboard :
56C
1.1 V on CPU

Sisoft Sandra:
23C CPU
2.77 V

Speedfan:
cpu 55C
Temp 55C
VCore 1.17
Core 14C

PC Wizard 2007:
Voltage CPU 2.21v
Process Temp of 20C


__________________
Abit KN9
AMD X2 Brisbane 3600+
Scythe Ninja Rev. B, passive

I've just built up a machine, X2 3600 Brisbane, NSK2400, Biostar TA690G.

I have the the CPU undervolted to 1.1V and overclocked to CPU=240 (2.3Ghz), memory at DDR2 667 and HT at 1200. My ambient is ~20C.

BIOS reports the board at 35C and the CPU at 33C.

When I get to windows, the Biostar tools report the CPU temp at 44-48C. This does not even change when I load the CPU with SP2004 or FAH.

Everest and Speedfan both always report 40C for the CPU, it never changes, no matter what I do.

CoreTemp reports some absurdly low numbers for CPU.

HWMonitor reports the same number as the Biostar BIOS.

The heatsink is not even warm and my two case fans are running at 5V (Yate Loons) and the Heatsink fan is running at 1400RPM.

So yes, it does seem like the sensor is either not well known or just not working.

While it does seem a little dubious that a consistent temp is elusive, I cannot believe how this thing undervolts/overclocks with little to no effort and the temps are apparently in the "green" (no pun intended).

I've bought an Asus M2N-E motherboard and a 65nm Athlon 64 X2 4000+ processor in April. First I couldn't switch on the system, it's reported vga fault through beeps. After 1-2 days of fault cheking I realized that every component working separately, so I had to get an older Athlon 64 to update my boards bios to a newer version because the old one not even capable to boot with a 65nm cpu.
After this completed, I thought everything will work, but this was a mistake. I inserted the 65nm CPU and the config booted up correctly but after 2 minutes of 100% load the reported core temperature reached 120 degree C, where the motherboard shutted down the system.

The CPU cooler was just warm and it perfectly fitted to the CPU.

I sold the mobo and the cpu and now I'm using an Asrock Conroe865PE (stock P1.2 bios 7/14/2006) mobo which could boot up with a new Core 2 Duo E4400 CPU (manufactured and released in 03/2007) and also handle core temperatures correctly.

I noticed that too. The default minimum p-state is 1.1V@1GHz, but my 4000+ runs stable at a mere 0.8V at that speed, which should result in almost 50% less power used.

This is really bad news. I was thinking of upgrading some of my family’s PCs with low end 65nm X2s. I figured that with a bit of under-volting it would be possible to cool these quietly enough using the stock heatsink with the fan reduced to somewhere in the 5 to 7V range. If there’s no reliable method of reading the CPU core temps that throws this idea out of the window.
I’ve only used systems with Core (2) Duo chips for the last year and good used to how easy it is to monitor the important CPU temp (Throttling temp). I’ve forgotten what the AMD equivalent is; doesn’t it vary from chip to chip with certain revisions?

Do Speedfan and RMClock handle the reading of AMD core temps as well as they do for Core based Intel chips? There is no clock throttling for AMD as far as I remember which is another negative. Its things like these that make me prefer Intel at the platform level. I’d rather support AMD at the moment but not if their products don’t support features that are needed to help quiet computing.

This is really only an issue if you leave the PC unattended for long periods running 100% load (e.g. Folding/other DC app); if it is idle CnQ will keep temps non-fatal even without forced cooling, I would imagine.

But if the DTS is broken you have no reliable way of knowing if the CPU is close to shutting itself down completely. If it supported throttling the lack of a working DTS would be less of an issue provided that it still throttled itself.
What I’m getting at is that it’s hard to calibrate a system without accurate CPU temps. And if the CPU lacks thermal throttling the need to calibrate becomes more critical. You can always just stick a Ninja on it and not have to worry if you aren’t overclocking, but if you’re trying to squeeze every last drop of silence out of the stock heatsink/fan it becomes very difficult with no DTS or throttling.

As a comparison I played around with a Core 2 Duo at 2GHz last night with the stock Intel heatsink and it was easy to calibrate.
With the fan off it idled with the Cores at 33C. Under load with the fan at ~950 rpm the Cores hit 59C which is 26C below the throttling temp so plenty of headroom. I reduced the fan speed in Speedfan from 60% to 50% and the core temps were in the 60s and the Intel fan was very close to silent.
The Motherboard was over-volting the CPU by 0.1V so I deduce that you can cool a 2.4GHz Core 2 Duo at stock voltage very close to silently with the Intel fan at ~850 rpm. This does assume that my example is typical. This is a shocking result as far as I’m concerned and with the incompatibility situation with Scythe coolers on S775 boards seemingly chronic, the ability to use the stock Intel unit is a very big bonus in my eyes.
I’m now considering abandoning my plan to use 65nm X2s and use E4300s instead. The fact that my old S775 board with no official support for the E4400 recognized the chip and certain AM2 boards aren’t recognizing the 65nm X2s is also pushing me in this direction.

Well, a bit of news re: the DTS on my X2 Brisbane. I have been working with BIostar testing a new BIOS to fix a couple of bugs re: S3 suspend/resume, and in the latest BIOS I have, all of a sudden the CPU temps are being reported accurately via the Hardware Monitor Utility (same as the BIOS reporting).

So, it appears I now have a reliable temp reading from within XP of my CPU. Well, at least now I have three sources that all show the same reading and when volting or loading the CPU, I can see these values change predictably.

The only thing I can add is that I have all my fans at 5V (Yate-loon 120mm) with no SmartFan control and with the CPU loaded 100%, I never see anything above 44C. At idle, it's around 30C.

I can't speak to the stock retail AMD heatsink/cooler, as I purchased this Brisbane OEM, but it seems to take very little effort to keep this chip cool.

Next, I'm going to try for .8V, "FSB" 240.

I have found that Coretemeter v1.0 beta4 and Core Temp 0.95 report the very same temps. A very low 16-17 for completely idle Abit KN9 with a Brisbane X2 3600+ cooled by a Scythe Ninja B, passive. One Nesus92mm in front and Yate Loon 120mm (1500 rpm) in back. Case is a steel Ultra Wizard. Yeh, I know.
Ambient is 25.5.
CnQ reports 58. volts are 1.1
Sisoft reports 24c and 2.64 volts. The voltage is twice what the voltage is without CnQ. So I guess it is somehow reporting the voltage for both cores.
Very strange with that high of an ambient.
I would like Abit to release a new bios to straighten all this out, but I think their bios people have moved on to newer boards.

Is this typical behavior for the 65nm X2s when installed in systems with older BIOSs?

Not always,it seems:

90nm vs 65nm



so seems like it is just luck of the draw. Most recent mobos should be 65nm-ready though.

My Abit KN9/BrisbaneX2 3600 booted and ran perfectly the first time and every time since.
The only things it does not do is report the correct temperatures and recognize the floppy. I don't use RAID so I guess I don't need the floppy. Still I wish it did recognize the floppy as I might want the RAID someday. Instead I just save everything to other computers and google docs, etc.

Ok let"s find out a bit more about AMD throttling. There is certainly some sort of throttling present. ThrottleWatch has the Athlon as supported, RMClock should be more reliable.

Read it up here: "BIOS and Kernel Developer's Guide for AMD NPT Family 0Fh Processors".
It has hardware throttling and shuts itself down if that"s not enough. Nothing as efficient as Thermal Monitor 2, though.

One could still rely on the temps reported by the SuperIO via the "legacy" diode. You could calibrate it when the CPU reaches throttling.


EDIT: Interesting posts by fgw from this point on.

Thanks for that; it makes X2s seem worth a punt.

Btw, there's a fact with Intel CPUs which is often overlooked. The DTS reports how much the temperature is away from the throttling point. This point is variable for desktop CPUs. On top of it, you cannot read this value out. So DTS readings from different CPUs are useless since CoreTemp assumes 85 degree celsius most Intel desktop CPUs.
If you own a desktop Intel CPU, I would have a goal of staying at least 10 degree away from the throttling temperature. The new CoreTemp can output this delta.
See also: http://www.alcpu.com/forums/viewtopic.php?t=247

Intel mobile CPUs just report the temperature as "near" 85 or 100 degree celsius.

The actual value of the throttling point (PROCHOT_L) for AMD CPUs remains a complete mystery to me and will be part of my next experimenting

I’m aware that not all Intel CPUs share the same TJunction temperature but it’s easy to calibrate for this variation by allowing a system to reach thermal throttling and make a note of the Core temperatures being reported. I don’t think that it’s possible to read the TJunction value from the CPU which is a shame.
The next revision of the Quad cores due in the summer not only has a lower TDP but a higher TCaseMax; probably with the same TJunction though. For people running their quad core close to the throttling temperature I suppose a higher TCaseMax temp is of benefit.

I quick follow-up. The german printed magazine C't presented an application which does the same as CoreTemp. So it also assumes 85 °C for desktop CPUs. So I need more proof that this isn't 100 % right. I found it in the intel developer forums.

About AMD, the C't article refers to a revision guide which says the AMD DTS functionality is broken. It can be found here.

Thanks for that. I looked at AMD's docs but I missed that. So the on-die thermal diode still works, but have motherboard manufacturers implemented hardware to read it?

Yes, I can second that the diode is working. I have accurate CPU temps on my 3600 Brisbane using Hwmonitor (hmonitor.net), setting Diode "0". AFAIK, this is the only app that can report accurate Brisbane temps.

I can also say that the Biostar TA690G is reporting the proper CPU temp for the Brisbane in the BIOS "PC Health" settings, but none of their software for Windows is reporting correctly.

im almost sure that this is due bad mobo/bios implementation.
I have Asus M2N32-sli and a 4000+ brisbane.
Every single application that reads temp from cores (ie speedfan, coretemp, everest) are working as it should.

I just tried installing a Brisbane on an MSI K9N SLI Platinum and it gave an error message stating that it didn't support this revision of CPU and that a BIOS update might help. Of course it wouldn't go beyond that point so another CPU is needed just to flash the BIOS.
Oh well, looks like I'll have a 65 & 90 nm X2 to compare after all.

First time I've seen an AM2 socket and it's a joy to work with after LGA775; an easy install and the design requires clear area around the socket which makes plenty of room for a Scythe Ninja etc.

Abit have released a new BIOS for my motherboard to fix this, and it is indeed reporting a reasonable temperature now (32C idle, 45C load with an AC freezer 64 pro with the fan at the minimum 1000rpm). I emailed them to ask how they're measuring the temperature now and they said it was read "from the CPU internally", so that must be the diode I suppose, which means the reported temperature should be as accurate as can be reasonably expected.

Applications which read the temperature from the BIOS also report correctly now, so I can use SpeedFan to control the fan speed properly (using the "CPU" temperature, not the "core" temperature, which is from the broken DTS). Unfortunately these new 4-pin fans don't allow speed to be reduced all the way down to 0.




I gather that the DTS reports temperatures that are out by varying amounts, so maybe you just happen to have a CPU which reports a temperature in the expected range, but anything which reads the DTS should not be trusted.

maybe... but from my experience its the other way round.

bios (and SpeedFan's "CPU" sensor) reads 45idle and 70+ load which is BS imho...
but coreTemp and everest reads 26/30 idle and 42/43 load respectively for core 0/1
as i said FOR ME, DTS arent broken...

all temps are at 2.9Ghz, 1.36V and under Noctua NH-U12F