>I suspect he has Magny Cours processors. They are multi-chip modules with two CPU chips per >package. /proc/cpuinfo will >tell you which cores are in which sockets
Yes. I think "physical id" shows the socket number
>$ numactl --cpunodebind=0 mprime -b4 -t
>You should see the temperatures diverge for the sensors in the same socket.
>You should see the temperatures diverge for the sensors in the same socket.
Yes I saw that the first two temperature are the same but different from other two
// Naderan *Mahmood;
From: Philip Pokorny <ppokorny@xxxxxxxxxxxxxxxxxxxx>
To: Jean Delvare <khali@xxxxxxxxxxxx>
Cc: Clemens Ladisch <clemens@xxxxxxxxxx>; lm-sensors <lm-sensors@xxxxxxxxxxxxxx>
Sent: Tue, October 26, 2010 6:34:22 PM
Subject: Re: k10temp is detected but not shown
On Oct 26, 2010, at 7:15 AM, Jean Delvare <khali@xxxxxxxxxxxx> wrote:
> Hi Clemens,
>
> On Tue, 26 Oct 2010 15:53:29 +0200, Clemens Ladisch wrote:
>> Jean Delvare wrote:
>>>>> Doh. You are aware that Core i7 CPUs are made by Intel and Family 10h
>>>>> (aka K10) CPUs are made by AMD, aren't you? They are different CPU
>>>>> models, they don't have to be designed the same way.
>>>>
>>>> I know that :)
>>>> I thought maybe that is a standard and each cpu temperature must be show
>>>> separetly.
>>>
>>> There is no such standard.
>>
>> And on the newest CPUs, Intel has switched to one temperature sensor per
>> package.
>
> Really? Starting with which CPU model?
>
>>> On Tue, 26 Oct 2010 06:06:53 -0700 (PDT), Mahmood Naderan wrote:
>>>> One more question, Is it possible to find out which two modules are tied to one
>>>> processor?
>>>
>>> Modules?
I suspect he has Magny Cours processors. They are multi-chip modules with two CPU chips per package. /proc/cpuinfo will tell you which cores are in which sockets.
>>>> Is it true that [00c3 and 00cb] are for one physical cpu and [00d3 and 00db] are
>>>> for the other cpu?
>>
>> I conclude from this question that you have two CPUs, and that each CPU
>> has two temperature sensors.
>>
>> The temperature sensor is part of the PCI device that associated with
>> the CPU's internal northbridge. IIRC the six-core CPUs have two
>> northbridges, for whatever technical reason, so they present two
>> temparature sensors to the OS.
>>
>> It looks as if the two sensors in each CPU have the same value. Is this
>> always the case?
>>
>> (I don't know where the c3/cb/d3/db names come from.)
>
> This is the PCI bus number in compact hexadecimal form:
>
> 00:18.3 -> 00c3
> 00:19.3 -> 00cb
> 00:1a.3 -> 00d3
> 00:1b.3 -> 00db
>
> We had to find a unique ID to differentiate between them, and that ID
> had to fit in an int to make libsensors happy.
>
>>> All entries in cpuinfo look the same, so I can't explain why some have
>>> the critical limit and some don't.
>>
>> The limit is set by the BIOS. I'd guess the BIOS knows that two sensors
>> are actually the same, an so doesn't bother to set the limit on the
>> superfluous ones.
>>
>> I'll look into the datasheet to see if this situation (two northbridges
>> per CPU) is documented and can be detected so that the driver can ignore
>> these superfluous sensors.
>
> If these sensors exist, I see no valid reason to ignore them. Two
> sensors can be useful, to slightly increase the resolution by combining
> them, or simply in case one of them get unreliable for whatever reason.
>
If these are Magny Cours, then they are definitely different sensors and they come from the different CPU chips in the package. If you ran a load on only one CPU chip:
$ numactl --cpunodebind=0 mprime -b4 -t
You should see the temperatures diverge for the sensors in the same socket.
Phil P.
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
_______________________________________________ lm-sensors mailing list lm-sensors@xxxxxxxxxxxxxx http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
