Safe socket temperatures

I'm overclocking on an ASRock 970M Pro3, but I think the the following question concerns all motherboards: What is the maximum safe temperature for the socket?

The general truth concerning AMDs sockets seems to be that you should stay below 70 C. However, there doesn't seem to be much vendor provided fact on the matter. Does it depend on the motherboard model, is anything that's still stable ok, can the motherboard be damaged if the temperatures goes beyond a certain limit?

The problem from a noob's perspective is that it's easy to miss the socket temperature altogether. Speccy doesn't show it, AMD Overdrive doesn't show the thermal margin for the socket, HWMonitor doesn't label it (at least not on my motherboard) as "socket" but as "CPUTIN"... So, if you don't do your research and only trust, say, AMD Overdrive, there's no telling what temps the socket could reach. I've even seen a case where a user didn't understand what socket temperature of 79 C on his ASRock 970M Pro3 could mean and refused to accept there were any issues with temperatures.

The reason I'm looking for possible official figures is that I happened to read somewhere that socket temperature readings were pretty reliable when the CPU was idle, but under load they became less accurate. The 70 C I see quoted (and have quoted myself) may be nothing more than a result of confirmation bias combined with unrealiable sensor readings. Maybe 75 C is in fact still safe, or then again maybe 65 C is too much for my motherboard. So, does anybody know what the real socket temperature limits are and how much do they depend on the motherboard you are using?

Thanks for the links, but I don't think they provide a conclusive answer and some of the comments in those threads can be even misleading. Let's take the answer from AMD, as that's supposedly the best source of information:

"Concerning your question regarding the temperatures with your processor. (1090) the maximum temperature threshold is 62 Celsius which set for the internal die (core) temperature of the chip. The core temperatures have an equational offset to determine temperature which equalizes at about 45 Celsius thus giving you more accurate readings at peak temperatures. The hindrance in this is the sub ambient idle temperature readings you speak of.

The silicon and adhesives used in manufacturing these processors has a peak temperature rating of 97+ Celsius before any form of degradation will take place. The processor also has a thermal shut off safe guard in place that shuts the processor down at 90 Celsius.

The Cpu temperature is read form a sensor embedded within the socket of your motherboard causing about a 7-10 Celsius variance form the actual Cpu temperature, which may be what you are reading about on the net.

You can use an application called AMD overdrive, that will allow you to monitor your temperatures accurately.

As long as your core temperature has not exceeded the high side of the 60 degree mark for extended periods of time you should be ok. 62 degrees holds a generous safety net to begin with."

I have the exact same CPU and according to the thermal margin given by AMD Overdrive the Tmax is not 62 C but roughly 70 C. So, AMD Overdrive says that the core/package temperature can safely be ~70 C, but the guy from AMD claims this is too high for extended periods. Which one can I trust? After all, he also states that no form of degradation will take place (due to heat) at those temperatures. Why shouldn't 70 C be okay (if it's stable)?

Regarding the socket temperature, I can understand why it might differ somewhat from the computational value that is the CPU temperature. However, on my system the difference is currently up to 18 C when running P95 (CPU maxes out at 45 C, socket at 63 C). This makes you wonder which figure actually is more descriptive. After all, I'm just reaching the CPU temperatures after which the computational temperature should be meaningful. Then again, the socket temperature sensor may just be of poor quality.

In any case, this much we know:

1. The CPU temperature is not the same thing as the socket temperature.
2. The socket temperature is measured by an actual sensor inside the socket.
3. The CPU temperature on (modern?) AMD CPUs is not measured, but calculated by an algorithm that is reliable only above ~45 C.

The question is, does the socket temperature sensor provide any meaningful information - under load or otherwise? I've seen experienced overclockers telling how they could push their system further by putting an extra fan on the backside of their motherboard to cool the socket. Did it also lower the CPU temperature, or does the temperature of the socket directly affect how, for example, the surface components on the backside of the motherboard work?

PetrolHead wrote: Thanks for the links, but I don't think they provide a conclusive answer and some of the comments in those threads can be even misleading. Let's take the answer from AMD, as that's supposedly the best source of information: "Concerning your question regarding the temperatures with your processor. (1090) the maximum temperature threshold is 62 Celsius which set for the internal die (core) temperature of the chip. The core temperatures have an equational offset to determine temperature which equalizes at about 45 Celsius thus giving you more accurate readings at peak temperatures. The hindrance in this is the sub ambient idle temperature readings you speak of. The silicon and adhesives used in manufacturing these processors has a peak temperature rating of 97+ Celsius before any form of degradation will take place. The processor also has a thermal shut off safe guard in place that shuts the processor down at 90 Celsius. The Cpu temperature is read form a sensor embedded within the socket of your motherboard causing about a 7-10 Celsius variance form the actual Cpu temperature, which may be what you are reading about on the net. You can use an application called AMD overdrive, that will allow you to monitor your temperatures accurately. As long as your core temperature has not exceeded the high side of the 60 degree mark for extended periods of time you should be ok. 62 degrees holds a generous safety net to begin with." I have the exact same CPU and according to the thermal margin given by AMD Overdrive the Tmax is not 62 C but roughly 70 C. So, AMD Overdrive says that the core/package temperature can safely be ~70 C, but the guy from AMD claims this is too high for extended periods. Which one can I trust? After all, he also states that no form of degradation will take place (due to heat) at those temperatures. Why shouldn't 70 C be okay (if it's stable)? Regarding the socket temperature, I can understand why it might differ somewhat from the computational value that is the CPU temperature. However, on my system the difference is currently up to 18 C when running P95 (CPU maxes out at 45 C, socket at 63 C). This makes you wonder which figure actually is more descriptive. After all, I'm just reaching the CPU temperatures after which the computational temperature should be meaningful. Then again, the socket temperature sensor may just be of poor quality. In any case, this much we know: 1. The CPU temperature is not the same thing as the socket temperature. 2. The socket temperature is measured by an actual sensor inside the socket. 3. The CPU temperature on (modern?) AMD CPUs is not measured, but calculated by an algorithm that is reliable only above ~45 C. The question is, does the socket temperature sensor provide any meaningful information - under load or otherwise? I've seen experienced overclockers telling how they could push their system further by putting an extra fan on the backside of their motherboard to cool the socket. Did it also lower the CPU temperature, or does the temperature of the socket directly affect how, for example, the surface components on the backside of the motherboard work?

I don't understand how you came to some of your conclusions, since IMO the information from AMD does match what you said is what AMD Overdrive said is safe.

This is the quote from the AMD source:

"... As long as your core temperature has not exceeded the high side of the 60 degree mark for extended periods of time you should be ok. 62 degrees holds a generous safety net to begin with."

Your quote about AMD Overdrive:

"... So, AMD Overdrive says that the core/package temperature can safely be ~70 C, but the guy from AMD claims this is too high for extended periods."

I don't see these statements as contradictory, or at worst not by very much. The "high side of the 60 degree mark", would be 67C - 69C. A temperature of approximately 70C is what, 65C - 75C? If a 10C range is too large, call it 68C - 73C, a 5C range. IMO, either of my AMD Overdrive approximate ranges overlaps the range given by the AMD source.

Exceeding the high side of the 60 degree mark is a rather open ended statement. If 70C was a hard limit that would trigger thermal protection being activated, you would know it. Many current Intel processors are known to begin thermal throttling at 100C. That temperature is stored in the CPU and can be read by monitoring programs. Other Intel processors have that temperature set to 105C. Intel over clocking enthusiasts have seen their processors thermal throttle at 100C, I have myself once. Unless AMD does things differently, they ought to have a maximum temperature that triggers thermal throttling too.

Thermal throttling of a processor is not simply based upon temperature. It is also based upon an average power usage of the processor over a defined time period, and the processor temperature, as well as the current processor power usage and temperature. A processor could thermal throttle at say 40C, if its power usage is very low. That indicates the processor's cooling is poor or failing.

The AMD source is likely being more conservative than what was programmed into AMD Overdrive. Intel always goes with a conservative specification for all processor parameters.

I don't agree with two thing you listed as things we know:

2. The socket temperature is measured by an actual sensor inside the socket.
3. The CPU temperature on (modern?) AMD CPUs is not measured, but calculated by an algorithm that is reliable only above ~45 C.

IMO, the AMD source did not describe the processor and socket temperature sensors well.

All processors measure temperature with a thermal diode which is calibrated to allow a specific voltage or current to flow (or other electrical parameter, such as a resistance) at a specific temperature. If the diode can be calibrated well, an increment of lower voltage (for example) can accurately "measure" another temperature. If 1V = 100C, and 0V = 0C, then 0.9V = 90C, 0.5V = 50C, etc. That is was is meant by the temperature being calculated. My example is simplified, but conveys the basic concept.

All the temperature readings for Intel processors are calculated, since they use thermal diodes. All electronic thermometers using thermal diodes such as in stoves and refrigerators have the temperature readout calculated in one way or another.

Normally, the thermal diode is calibrated to be the most accurate where it needs to be, at the dangerous temperature that needs to be detected. If it is less accurate at lower temperatures, that is not a problem since they are not as dangerous. Intel at one time had thermal diodes in their chipsets that had a lowest temperature reading of 50C. Any temperature below that was not worth measuring for safety purposes, so why bother with a more expensive, highly calibrated thermal diode. AMD seems to have the same philosophy regarding processor temperatures.

If the socket temperature measurement on an AMD board is provided by a large, discrete thermal diode, it still works by the same principal. All electronic thermometers do. Otherwise, what is used, a glass thermometer with mercury in it, or a bi-metallic strip as used in thermostats?

Your final question, what is the purpose of the socket temperature reading, is a good one that remains unanswered.

The internal thermal diodes that Intel currently uses seem to show accurate temperatures down to room ambient temperature. We see the same thing with video card GPU chip temperatures. They may have used those as a competitive advantage. Mother board manufactures might have come up with the socket temperature on AMD boards as a way to give us more than AMD provided with their processors. While the processor temperature readings may be simply eye candy, we like them, I certainly do. The sub-ambient CPU temperatures that can be calculated from an AMD processors thermal diode are obviously not real, but some users like to believe they are true.

&¤#@@! Managed to click the wrong mouse button by accident and lost a long(ish) post. Oh well, here goes again...

parsec wrote: I don't understand how you came to some of your conclusions, since IMO the information from AMD does match what you said is what AMD Overdrive said is safe. This is the quote from the AMD source: "... As long as your core temperature has not exceeded the high side of the 60 degree mark for extended periods of time you should be ok. 62 degrees holds a generous safety net to begin with." Your quote about AMD Overdrive: "... So, AMD Overdrive says that the core/package temperature can safely be ~70 C, but the guy from AMD claims this is too high for extended periods." I don't see these statements as contradictory, or at worst not by very much. The "high side of the 60 degree mark", would be 67C - 69C. A temperature of approximately 70C is what, 65C - 75C? If a 10C range is too large, call it 68C - 73C, a 5C range. IMO, either of my AMD Overdrive approximate ranges overlaps the range given by the AMD source.

Let's for argument's sake say it's 68 C - 73 C, since that's probably closer to the truth (I'll double check the value next time I run P95). The key thing is I would expect this temperature to be safe to use 24/7 if I was only getting my info from AMD Overdrive, whereas the statement from AMD would imply that I could be damaging my system if I ran the CPU at these temperatures 24/7. At the very least I feel the answer from AMD contains some mixed messages. It is a good point, though, that whoever gave the answer may have only erred on the side of caution. I would do so too, if I was giving advice to a customer.

I don't agree with two thing you listed as things we know: 2. The socket temperature is measured by an actual sensor inside the socket. 3. The CPU temperature on (modern?) AMD CPUs is not measured, but calculated by an algorithm that is reliable only above ~45 C. IMO, the AMD source did not describe the processor and socket temperature sensors well.

Apologies for being unclear here. I wasn't basing these on the answer I quoted, but instead several sources I've come across. I did, however, get things confused here and appreciate you correcting me. Naturally there needs to be some sort of sensor to provide data that can be turned into a value representing temperature. What I was trying to say is that (AFAIK) the temperature obtained with the help of the socket sensor is, at least in principle, an actual physical temperature. The readings it gives may not be accurate in all temperature ranges or when using it to determine the CPU temperature, but they should still at least sometimes corresponds to temperatures that could be measured from the socket directly by some other method. The CPU temperature on the other hand is not - in princpile - a physical temperature that could be measured anywhere on the system, but instead a value that's relative to the maximum operating temperature of the CPU. This does not mean the value can't sometimes correspond reasonably well to actual temperature values on the die, but this is not really the information the value is designed to provide. This is why AMD Overdrive tells the temperature margin and not the CPU temperature. Of course it doesn't really matter if we use something like HWMonitor and it tells us that the CPU temperature is 55 C, while the actual physical temperature might be something else; all we users need to know is that the value 55 C is a safe value.

The internal thermal diodes that Intel currently uses seem to show accurate temperatures down to room ambient temperature. We see the same thing with video card GPU chip temperatures. They may have used those as a competitive advantage. Mother board manufactures might have come up with the socket temperature on AMD boards as a way to give us more than AMD provided with their processors. While the processor temperature readings may be simply eye candy, we like them, I certainly do. The sub-ambient CPU temperatures that can be calculated from an AMD processors thermal diode are obviously not real, but some users like to believe they are true.

I almost wish my system would show sub-ambient temperatures. The fact that both the socket temp and CPU temp seem entirely reasonable when idle makes the readings feel more trustworthy. ;)


Edited by PetrolHead - 05 Jan 2016 at 9:32pm