if you ran a 3570 at around 100-110C for 12 hours (constant throttle under load)...

[Vectronic]

Depends on the cooler, ambient temperature, and air-flow.

Typically, I would say about 30C above idle temp.

 

[AtenRa]

Its not an issue for the CPU since it was allowed to throttle as needed.

The chip is designed to last for years under load at 105C. Keeping it cooler than 105C will allow it to last even longer as well as use less power (less leakage).

AMD's chips are not designed to operate anywhere near those temperatures. I mention this only to point out that there is no rule of thumb other than knowing what the manufacturer determined to be safe. It varies every chip generation, even within the same node.

Where OC'ers get into significant degradation is when they over-volt and get their processor hot.

Running at TJmax with stock settings is not a concern, but don't do that at 4.5GHz with 1.3V D:

Yeap, if we raise the voltage and/or frequency we should lower the temperature to compensate (Black’s equation).


But it also depends on the interconnects used(materials and node process/shrinkage), different metals exhibit different characteristics in electromigration, for example copper is more resistance than Aluminum. Also, metal alloy’s have higher electromigration resistance like Aluminum+Copper and Copper+Palladium than their host metals.

 

[AtenRa]

What's a normal max temp for a 3570 under 80%+ load?

For the 3rd generation Intel Core processors, Maximum TJmax (temp at the core) is 105c, the maximum Tcase(CPU temp) is 67,4c at default voltage and frequency.

 

[Idontcare]

But it also depends on the interconnects used(materials and node process/shrinkage), different metals exhibit different characteristics in electromigration, for example copper is more resistance than Aluminum. Also, metal alloy’s have higher electromigration resistance like Aluminum+Copper and Copper+Palladium than their host metals.

Today's processors are voltage limited for degradation that occurs in the FEOL (xtors) more so than the BEOL (wire interconnects).

Electromigration is still a concern, but your CPU will long be dead from degradation to the transistor from over-voltage before the wires get a chance to degrade nowadays.

For the 3rd generation Intel Core processors, Maximum TJmax (temp at the core) is 105c, the maximum Tcase(CPU temp) is 67,4c at default voltage and frequency.

Tcase has nothing to do with the CPU temp though, it is a temperature measurement that isn't physically located in the CPU, it is the temperature in the surface of the outside of the IHS.

For example with all mobiles chips, which come without an IHS as they are delidded from the start, Tcase has no definition.

All that matters is the junction temperature, in the silicon of the CPU. No one has ever damaged their IHS by exceeding Tcase, but damaging your processor by exceeding TJmax is a real possibility.

The reason Intel (and AMD) have to specify a Tcase is to prevent cheapskate OEMs and box-sellers from pairing an underperforming HSF and poor airflow cases (not enough fans, etc) with their respective CPUs.

By requiring the HSF to have high enough thermal conductivity as to keep TCase below the spec value while the CPU itself can hit TJmax, the CPU makers force the system integrators to spend money in building the box so as to avoid putting the CPU into thermal throttle under routine usage.

 

[AtenRa]

Today's processors are voltage limited for degradation that occurs in the FEOL (xtors) more so than the BEOL (wire interconnects).

Electromigration is still a concern, but your CPU will long be dead from degradation to the transistor from over-voltage before the wires get a chance to degrade nowadays.

Ohh, i thought that with smaller interconnects (due to smaller process) would accelerate the EM effects.


Do you have any link/s for transistor defects and reliability??

Tcase has nothing to do with the CPU temp though, it is a temperature measurement that isn't physically located in the CPU, it is the temperature in the surface of the outside of the IHS.

For example with all mobiles chips, which come without an IHS as they are delidded from the start, Tcase has no definition.

All that matters is the junction temperature, in the silicon of the CPU. No one has ever damaged their IHS by exceeding Tcase, but damaging your processor by exceeding TJmax is a real possibility.

The reason Intel (and AMD) have to specify a Tcase is to prevent cheapskate OEMs and box-sellers from pairing an underperforming HSF and poor airflow cases (not enough fans, etc) with their respective CPUs.

By requiring the HSF to have high enough thermal conductivity as to keep TCase below the spec value while the CPU itself can hit TJmax, the CPU makers force the system integrators to spend money in building the box so as to avoid putting the CPU into thermal throttle under routine usage.

You are correct, I should have been more thorough and specify that the Tcase is in the IHS. I just used “CPU” for the entire CPU package(PGA+Die+HIS) trying to differentiate it from the CPU die(and the core/s).

 

[Idontcare]

Ohh, i thought that with smaller interconnects (due to smaller process) would accelerate the EM effects.

They do, but the xtors degrade even faster, so in the race to see who is the weakest link it will always be the xtors with today's modern nodes.

Shrink the interconnect, yes EM gets worse, but shrink the xtors at the same time and they get even worse still. So they get the prize, if you can call it that.

Do you have any link/s for transistor defects and reliability??

None offhand. The info in the public domain tends to lag behind the leading edge industry knowledge-bank by a good 8-10yrs in my estimation.

I remember when Yoshio Nishi and Robert Doering (both ex-TI'ers, one a VP) wrote the Handbook of Semiconductor Manufacturing Technology and it was about 8yrs out of date (already) when it went to print.

Not because those guys don't know the latest and greatest, but they are bound by all sorts of NDA's and so forth from actually putting the latest and greatest into the public domain. But they can update the public domain so it doesn't fall any farther behind than the 8-10 yr lag that corporate allows for IP and trade secret leakage.

 

[Puppies04]

^ wrong. The stuff the chip is made of, starts to melt around those temps (I believe).

I stopped reading your post at this point. This is all kinds of fail.

 

[Turbonium]

Depends on the cooler, ambient temperature, and air-flow.

Typically, I would say about 30C above idle temp.

Stock Intel cooler.

My idle temps are about 38-42C.

My load temps around around 75-80C. Sometimes even slightly more. I just thought that was a bit high.

Note: this is what the temps are being reported as by the Intel motherboard/temp utility.

 

[AtenRa]

They do, but the xtors degrade even faster, so in the race to see who is the weakest link it will always be the xtors with today's modern nodes.

Shrink the interconnect, yes EM gets worse, but shrink the xtors at the same time and they get even worse still. So they get the prize, if you can call it that.


None offhand. The info in the public domain tends to lag behind the leading edge industry knowledge-bank by a good 8-10yrs in my estimation.

I remember when Yoshio Nishi and Robert Doering (both ex-TI'ers, one a VP) wrote the Handbook of Semiconductor Manufacturing Technology and it was about 8yrs out of date (already) when it went to print.

Not because those guys don't know the latest and greatest, but they are bound by all sorts of NDA's and so forth from actually putting the latest and greatest into the public domain. But they can update the public domain so it doesn't fall any farther behind than the 8-10 yr lag that corporate allows for IP and trade secret leakage.

Ok got it thanks,

I have forgotten I have the Handbook of Semiconductor Manufacturing Technology, i will read the reliability section, thanks for remind me

 

[ShintaiDK]

Stock Intel cooler.

My idle temps are about 38-42C.

My load temps around around 75-80C. Sometimes even slightly more. I just thought that was a bit high.

Note: this is what the temps are being reported as by the Intel motherboard/temp utility.

Perfectly fine. A large headroom for Tjmax at 105C.

 

[Turbonium]

Just a minor update: my idle temps are around 48°C nowadays (room temp is about 26.5°C). Load temps (say while gaming) are around 70-75°C.

Again, I know it's perfectly within spec, but I was wondering if this is a bit on the high side (even for a stock Intel cooler; it's almost completely free of dust by the way, as I clean it regularly).

 

[ShintaiDK]

Just a minor update: my idle temps are around 48°C nowadays (room temp is about 26.5°C). Load temps (say while gaming) are around 70-75°C.

Again, I know it's perfectly within spec, but I was wondering if this is a bit on the high side (even for a stock Intel cooler; it's almost completely free of dust by the way, as I clean it regularly).

Its not.

 

[Turbonium]

Well it's official now in my mind: stock Intel coolers are mediocre.

Experience is the best teacher at times.

 

[BSim500]

What's a normal max temp for a 3570 under 80%+ load?

I have an i5-3570 (non-K) with a mild 4GHz OC & slight -0.05v undervolt. On a CoolerMaster 212 EVO and depending on the time of the year (summer vs winter), it typically idle's at 29-34c, is 48-54c heavy gaming / 4-thread video editing, and hits 55-59c on 100% 4T Prime / Intel Burn Test max load. And all that's with 2x very slow 700-800rpm 120mm fans (CPU & 1x case fan) optimised for 10-15db silent operation. For many games it barely goes above 50c and fans never spin faster than 850rpm. At more normal 1000-1500rpm fan speeds, even Prime doesn't get it over 55c. Best CPU & cooler combo I ever bought. The only possible way it could get anywhere near 100c is if the heatsink somehow fell off.

Stock coolers are generally rubbish and can easily run 20-30c hotter. If it fits your case, buy a CM 212 EVO or similar class aftermarket cooler. Even the smaller 92mm TX3 is still a big step up. I don't think I've used a "stock" heatsink in about 8 years now.

 

[Turbonium]

Stock coolers are generally rubbish and can easily run 20-30c hotter. If it fits your case, buy a CM 212 EVO or similar class aftermarket cooler. Even the smaller 92mm TX3 is still a big step up. I don't think I've used a "stock" heatsink in about 8 years now.

I would, but at this point, I'd rather not mess with what's already working. While it's running a bit hotter than I'd like, it's still well within spec. I just can't justify spending the money, but more importantly, spending the time and effort to swap out the cooler, all while running the risk of damaging something.

Also: given recent experiences, I've grown wary of pushpin designs, and would rather use a backplate option if at all possible, which would involve taking out the mobo, which is a huge task.

For my next build though, I will strongly consider an aftermarket cooler.