Before shutting down, thermal throttling will first slow the CPU down, and this seems to be exactly what the original poster observed. I really think this is not software but bardware related: either the temp sensor is wrong or (more likely) the "thermal interface" between heat sink and chip is not ok.
CU
HB
Agreed. Another point is that Intel's position is that the on-die sensors are not properly considered as thermometers, but are just essential components in a system to provide for correct operation, reliability, and protection. Third-party programs which display temperatures involve some degree of guesswork even as to the intended temperature reference, and aside from that ambiguity the actual sensors often display fairly substantial errors of linearity and even of on-die matching.
If you poke around in Speedfan a little, you will find there is provision to invoke a temperature offset separately for each core. This can help with the matching problem, but is not enough to deal with linearity and "missing code" type problems.
So while I'll not be too sure just how hot yours is, I agree with Bikeman which way the general evidence points.
I find the cpu temperature somewhat hard to believe - is the software reading the sensors right? I was under the impression cpu's have nowadays some kind of a safety circuit which should shut them down when they reach 90C or such. At nearly 100C it's a miracle the cpu wasn't fried up already.
The cutoff is 100C, and since the OP said that when it dropped 5C his chip went from about 35% to 100% of expected performance it;s clear that it was throttling hard when it was at 97C. The reason you want to keep your temperatures below about 70C in normal use is that at higher temperatures the chip ages faster; the throttle limits are set around the prompt kill zone (ie where you could quickly kill the chip) because in normal usage (vs what we're doing) a CPU only rarely will be at full load and hit near the max even if it's got a badly clogged heat sink, in a location with no ventilation, and is in a hot- unairconditioned room. In those circumstances actual damage is unlikely before the system is obsolete, but sudden slow downs when the CPU is needed most are readily apparent.
Yes that is what was happening with that machine Dan.
But I decided to just shut that one down and one other that is a single.
For some reason they both decided to add a HD problem to the cooling problem so I replaced them with faster machines and when winter gets here and I am inside most of the time I will see if I can get one of them to come back to life.
(I have a new SATA but they both are pre-SATA so I can't plug it in)
I also had another old timer....P4 2.5 with the same problem and it started shutting down after a couple hours and I think I will just add it to my parts stash.
I am working on making this room twice it's size so I have more room to spread things out and have a better work bench to take em apart and test things.
I replaced them with a couple triple's and a quad and they are working nicely and still have a low priced quad and a dual running that have been running for a while.
RE: Before shutting down,
)
Agreed. Another point is that Intel's position is that the on-die sensors are not properly considered as thermometers, but are just essential components in a system to provide for correct operation, reliability, and protection. Third-party programs which display temperatures involve some degree of guesswork even as to the intended temperature reference, and aside from that ambiguity the actual sensors often display fairly substantial errors of linearity and even of on-die matching.
If you poke around in Speedfan a little, you will find there is provision to invoke a temperature offset separately for each core. This can help with the matching problem, but is not enough to deal with linearity and "missing code" type problems.
So while I'll not be too sure just how hot yours is, I agree with Bikeman which way the general evidence points.
RE: I find the cpu
)
The cutoff is 100C, and since the OP said that when it dropped 5C his chip went from about 35% to 100% of expected performance it;s clear that it was throttling hard when it was at 97C. The reason you want to keep your temperatures below about 70C in normal use is that at higher temperatures the chip ages faster; the throttle limits are set around the prompt kill zone (ie where you could quickly kill the chip) because in normal usage (vs what we're doing) a CPU only rarely will be at full load and hit near the max even if it's got a badly clogged heat sink, in a location with no ventilation, and is in a hot- unairconditioned room. In those circumstances actual damage is unlikely before the system is obsolete, but sudden slow downs when the CPU is needed most are readily apparent.
Yes that is what was
)
Yes that is what was happening with that machine Dan.
But I decided to just shut that one down and one other that is a single.
For some reason they both decided to add a HD problem to the cooling problem so I replaced them with faster machines and when winter gets here and I am inside most of the time I will see if I can get one of them to come back to life.
(I have a new SATA but they both are pre-SATA so I can't plug it in)
I also had another old timer....P4 2.5 with the same problem and it started shutting down after a couple hours and I think I will just add it to my parts stash.
I am working on making this room twice it's size so I have more room to spread things out and have a better work bench to take em apart and test things.
I replaced them with a couple triple's and a quad and they are working nicely and still have a low priced quad and a dual running that have been running for a while.