SPCR

http://www.techpowerup.com/?51419

Don't have one myself so I can't judge the legitimacy of the claim, but apparently the E8400 erroneously sends cooler temp data then it actually is running at. This could explain how these processors seem to run so cool under load, but I have no idea. Can anybody confirm?

ocforums, abit forums, xtremesystems forums all have extensive info on this.

some people aren't having problems, but it seems that the first batch of e8400s may have flawed temp sensors. somebody traded in for a newer cpu batch and it fixed the problem but not totally. there also may be an issue with these cpus requiring single rail PSUs, even though they're supposed to run cooler. It may be a problem with the mobos though, bc people with ga-p35-ds3ls aren't having problems that i've seen yet. I got the abit ip35 pro ready to be hooked up with my e8400 once I know that the temps will be ok so I don't damage my system. Some people are reporting that a flash of the BIOS fixes their problems with that board, and some aren't. It's so confusing and a lot more testing needs to be done. I'm glued to this topic bc i want my system up and running. I think I'll just try the new BIOS, see if it works and rma the cpu if it doesn't. That seems to be the best route so far.

edit: everest seems to be giving the accurate temps much more than other programs.

There is no flawed sensor. Some motherboards just don't handle the sensor correctly.

An E8400 used in different motherboards from different manufacturers will report wildly different temperatures.

Abit is one manufacturer that is known to handle the E8400 sensor incorrectly. They are working to fix the problem, but an update is not expected until after the Chinese new year.

KenAF wrote:
Abit is one manufacturer that is known to handle the E8400 sensor incorrectly. They are working to fix the problem, but an update is not expected until after the Chinese new year.

They've released a couple BIOS updates already, but only some people are reporting success.

I don't think it's the motherboards.

One guy went through three E8400 CPUs, and they all reported different temperatures and were "locked" on idle.

Isn't this sensor on-die and read via a CPU register? That should keep the motherboard out of the equation.

Matija wrote:I don't think it's the motherboards.

One guy went through three E8400 CPUs, and they all reported different temperatures and were "locked" on idle.

The "locked" temperatures is what some motherboard manufacturers -- such as Abit -- are doing as a temporary solution until they implement proper support.

Every E8400 CPU will have a slightly different stock operating voltage and temperature. The only shared attribute of all CPUs is that they operate below a certain temperature and below a certain max voltage.

cpu's have on chip diode junctions used for temp sensing. These connections are pinned out and are read by the mobo's temp management chip.

So,

- the diode junction could be poorly designed
- the mobo temp management chip could have a bad curve for the diode over temp.
- could be a combination of the two.

Don't know the details of the 8400, but, this is how it works in general.

Does anyone know if this affects all Wolfdales, or is it just the E8400 (3GHz) speed grade?
Also, what exactly is the current experience on motherboards? Are there specific vendors that people have not currently run into these problems?

It looks like it affects all the new Intel E8x00 series processors at the moment.

It's not really that they're actually overheating though. It's just that the temperatures don't display correctly in monitoring software such as Coretemp 0.96.1.

Both my E8500 CPUs (used with Gigabye GA-P35-DS4 and GA-X38-DS4 motherboards) show unreliable temperatures. They don't appear to have a heat problem: The CPU heatsinks are cool to the touch, and the air coming out of the case exhaust fan is barely warm.

There have been a few mentions of some QX9650 processors having the same problem too:

http://www.xtremesystems.org/forums/sho ... p?t=171381

All Core2 CPUs have two internal temperature sensors: a thermal diode that is pinned out to the motherboard and measured by an external chip, and a DTS (digital thermal sensor) that is calculated inside the chip and read by software from a CPU register.

The diode is provided for backward compatibility with P4 CPUs, and has inherent reliability issues, not least of which is noise susceptability which requires very tight layout rules.

DTS was designed to move the burden of accurate sensing from the motherboard/system vendor to Intel. It worked very well in Conroe and Allendale, once people understood that it reported the headroom below throttling as opposed to the absolute temperature.

Many motherboards provide a third way to measure CPU temperature: an analog sensor mounted on the board near the CPU socket. This is very unreliable though, as the chip runs much hotter than the board it's mounted on. In SpeedFan, this is usually called something like "CPU", whereas the other sensors are called something like "Core".

cmthomson wrote:All Core2 CPUs have two internal temperature sensors: a thermal diode that is pinned out to the motherboard and measured by an external chip, and a DTS (digital thermal sensor) that is calculated inside the chip and read by software from a CPU register.

The diode is provided for backward compatibility with P4 CPUs, and has inherent reliability issues, not least of which is noise susceptability which requires very tight layout rules.

DTS was designed to move the burden of accurate sensing from the motherboard/system vendor to Intel. It worked very well in Conroe and Allendale, once people understood that it reported the headroom below throttling as opposed to the absolute temperature.

Many motherboards provide a third way to measure CPU temperature: an analog sensor mounted on the board near the CPU socket. This is very unreliable though, as the chip runs much hotter than the board it's mounted on. In SpeedFan, this is usually called something like "CPU", whereas the other sensors are called something like "Core".

So what are the programs which read DTS?

-Robert

valnar wrote:So what are the programs which read DTS?

Lots of them. CoreTemp, RMClock, Everest Ultimate, SpeedFan (latest version), TAT, etc.

cmthomson wrote:

valnar wrote:So what are the programs which read DTS?

Lots of them. CoreTemp, RMClock, Everest Ultimate, SpeedFan (latest version), TAT, etc.

Most of those programs currently don't work with the E8x00 processors.

I had a quick try with the ones you listed to see what happens:

Coretemp 0.96.1 displays CPU core temperatures, but those readings are the ones most people are reporting as inaccurate.

RMClock 2.30.1 can't identify the CPU and doesn't display any voltages or core temperatures.

Everest Ultimate (build 1257 beta trial version) can display CPU core temperatures but they're the same as the ones from Coretemp 0.96.1.

Speedfan 4.33 doesn't display the CPU core temperatures at all.

TAT tool is unable to even load as it doesn't recognise the CPU.

The Core Temp thread on XS:
http://www.xtremesystems.org/forums/sho ... 38&page=13

Somebody had a 45nm quad with stuck temperature readings. He switched it out and the new one seems to work fine.

Since thermal throttling is triggered by the DTS readings this could potentially turn out to be a major problem. I suspect that it isn’t and that thermal throttling works and Intel just didn’t bother fully testing the DTS at lower temperature ranges which are irrelevant to thermal throttling.
Has anyone seen any evidence to suggest that thermal throttling is broken on these CPUs? Faulty temperature readings is a pain but nothing compared to broken throttling.

I currently use CoreTemp and RMClock to calibrate my DTS temps and thermal throttling point and then use Speedfan to control fan speeds in relationship to its DTS temps which often need to be given an offset. Without the ability to do this I will pass on upgrading to Wolfdale.
It seems strange that Intel chose to change the way that DTS readings are read to the point that nothing can read them!

On the other hand, I've read in a forum thread somewhere (HardForum?) that someone had a normal E8400 (meaning, normal temp readings), and after about a week they got stuck at 15C above what they used to be.

Something's wrong with the CPUs, which is a damn shame. That said, it's not a big deal unless you are obsessed with temp readings - the CPUs are perfectly safe, even when overclocked.

I have an E8400 in an Abit IP-35 Pro, and I just installed a Scythe Ninja with the Thermalright Bolt-Thru. I don't have Windows installed on the PC yet (was trying out Ubuntu), but I am concerned about this issue.

How am I supposed to know if I installed the heatsink correctly if the temp reported is not correct? I followed the video here, and I made sure that I did not tighten it too much, i.e. I did not tighten the screws all the way down. Just enough so that the heatsink seemed not to move much. Also the board is vertically mounted.

Should the bios report the correct temp. Right now it is reporting only 9C, is this correct?

I will be installing XP on it soon, and I only plan on overclocking to 3.6GHz. Once I have XP installed I will report back.

sdheda wrote:How am I supposed to know if I installed the heatsink correctly if the temp reported is not correct? I followed the video here, and I made sure that I did not tighten it too much, i.e. I did not tighten the screws all the way down. Just enough so that the heatsink seemed not to move much. Also the board is vertically mounted.

With older CPUs too much pressure could potentially crack the fragile CPU core. The newer processors, with a built in heatspreader, can take a more secure fitting.

The whole point of using the bolt through kit and backplate is to get more pressure between the CPU and CPU cooler. If you can move the heatsink easily it's not really tight enough. Tightening the spring loaded screws right down would be better.

If anything you could do with more spring tension than that: with a Thermalright Ultra 120 Extreme (which uses the same bolt through kit) you often see people using a shim to get even more pressure.

http://www.xtremesystems.org/forums/sho ... 499&page=2

my 9450 quad..indeed houston..there's a problem. this is nothing though..when i first installed it saturday it was reading 212-214C all day..this is nothing:)

[/img]

Just an update.

I installed XP and indeed my temp reading is also stuck on 44C. Bios, however, is reporting only reporting 9C.

Also what is a safe CPU Vcore for an E8400. I have is at 1.33 and still cant get it stable. Of course, CPU-Z is reporting a lower value.

1.5 is what i've seen as outside, but i wouldn't go over 1.4 max, unless you're going for a suicide run:) i'd try upping other voltages that your mobo supports rather than going over 1.35-1.4

TENAX wrote:1.5 is what i've seen as outside, but i wouldn't go over 1.4 max, unless you're going for a suicide run:) i'd try upping other voltages that your mobo supports rather than going over 1.35-1.4

There are four other voltages, which one should I try?

CPU VTT
MCH
ICH
ICHIO

have you gone to the universal abit forums? lots of good info on there including a thread with recommended voltages.. i'm at work..don't have a link handy but there's good stuff there for this board:)

It could easily be your memory timings causing instability. If the memory is having problems with its timings that will cause it to fail any stress tests, even if you increase the CPU vcore to very high levels.

You may need to increase the memory voltage and/or lower the memory speeds whilst relaxing the memory timings (a higher number means the memory is running slower).

Have a look at this informative Abit IP-35 Pro thread too:

http://www.xtremesystems.org/forums/sho ... p?t=147163

The last few pages of the thread have quite a bit of discussion of the best voltage settings to use for overclocking "Wolfdale" processors.

Where I posted that list of programs above that don't work you need to keep an eye out for newer versions of the programs. RMClock and Speedfan both have newer beta versions that appear to support "Wolfdale" processors now.