XP-M frequency potential degrades with time?

[Petr]

Hello!

I would like to ask whether there is something like GNDS on Athlons. My problem is I can no longer run the frequencies / voltages I ran when my Mobile Athlon XP-M 2600+ IQYHA0352MPMW was new (two months ago). I have quite inadequate cooling for it - Arctic Cooling Copper Silent 2 TC (I have it because it is quiet and doesn't cost too much). My mobo is EPoX EP-8KRA2I, have Enermax 353W ATX12V v1.1 PSU and quality RAMs. This mobo has 3-phase voltage regulation with 6 MOSFETs each 45A max. There are 5 capacitors near socket and other bigger behind MOSFETs. In my opinion the mobo has quite good voltage regulation and I never suffered any problems with stability. The only drawback is there isn't 12V Power connector.

I used to run 2400 MHz / 1.65V what was stable (at 1.625V Prime failed after more than five hours). Now I need 1.675V as with 1.65V Prime sometimes fails after random number of hours - can be half an hour or more that four. The same is for lower frequencies - 2300 MHz at 1.575V (was 1.55V) and 2200 MHz at 1.475V (was 1.45V). Although weather was hot in Europe in last weeks, my temps doesn't seem to increase and I have around 65 with Prime. Yes, it is quite hot but it was rock stable and now has sometimes problems. I noticed that maybe running higher RPMs on Enermax rear fan improves stability (?) of CPU at all frequencies. What I noticed when CPU was new was it runs so hot. At 2400 MHz / 1.65V it run as hot as my older Thoroughbred 1700+ B0 DUT3C at 2200 MHz / 1.825V. With Tbred, I could run 2000 MHz / 1.6V with idle temp of 35C and 1000 RPM on CPU cooler. With this mobile, I have in idle 40C at 1400 RPM. Can it be this CPU suffers from excessive leakage and therefore doesn't like voltages that are no problem for regular desktop parts? I noticed here IQYHA should be best overclockers but also run much hotter than other series.

Another thing on this topic... my friend with the same CPU batch (IQYHA0352MPMW, XP-M 2600+) ran 2600 MHz at 1.8V prime stable all the day with his watercooling. He had temperature of around 50C. Now prime fails within few minutes and temperatures seem to have increased to 53C. He tried new PSU - no improvement. He tried cooling MOSFETs - no improvement. His mobo is Abit NF7 v2.


Any suggestions what is happening?

 

[Captain Newbie]

I suggest you read this post on CPU frequency degredation (by Hitechjb1), but to distill it down:

1. Anything that happens over a short period of time (instant, seconds, minutes), I would consider that caused by exterme conditions, above manufacturing stress test voltage and temperature, due to mistakes. It can be voltage or temperature related, e.g.
- large step voltage increase causing abnormal current surge which leads to heat and meltdown in weak spots
- high overvoltage can cause transistor gate oxide breakdown due to high electric field
- bad thermal contact leading to high temperature spot
- even the measured temperature (average over the chip area) is not extreme, but certain spot inside the chip may get damaged due to high voltage and high temperature gradient if sudden excessive voltage is put on.

2. Anything that happens over a medium period of time (weeks, months), I would attribute that to weak spots resulted from design margin issues and/or manufacturing random defects not caught during manufacturing stress tests (such as weak spot accelerated electromigration, leaking currnet heat up) and overvoltage would further speed up the deterioation over weak spots leading to failure.

3. Normal long terms failure expectancy (years), this is caused by the long term electromigration due to voltage and temperature. Higher voltage and higher temperature would reduce the long term failure time of the chips, it is a tradeoff between overclocking (i.e. the MHz gain) and expected failure time. The statistical life expectancy can be calculated from the amount of voltage increase and temperature increase.

 

[hitechjb1]

That link may not work properly. This pointer links to the original post:

What could damage a chip/CPU permanently?


In many case, the frequency degradation may be just due to seasonal effect, i.e. higher room temperature and case temperature and hence reducing the stable CPU frequency.

Impact of higher ambient temperature on CPU clock frequency


The relationship between voltage, frequency and temperature, and effect of overclocking on life expectancy, this link points to some details:

Voltage, temperature and frequency: the basic variables of overclcoking

 

[schismspeak]

I would just say that it just hasn't been adequatley cooled, had you had a cooler that cooled it to just 55°C under load, I don't think you would have any problems, also make certain that it isn't a memory problem, it could be getting too hot also, but I would conclude that your processor slowly got worse over the two months of extreme temperature it was subjected to, or it is just to hot.

My dad's pc runs around 60-63°C load, and i'm not worried because it is only oc'ed about 375mhz and voltage is only .25v over stock. But I think when you start to oc much higher than 400mhz, and/or you overvolt more than .5v you should do your best to keep your temps below 60°, preferably 55° if possible.

I would get a thermalright cooler with a 3000rpm panaflo or something all copper with an 80mm fan running atleast 2900rpm, and you might get your oc back, that few degrees it rose in your home could make the difference between being stable and not. Your processor might not even really be damaged, some cpu's will be fine, but right when they hit their maximum operating temp which is deifferent for every one, they will fail tests.(as you know)
Put a better cooler on it and see if it is any better. That is my advice.

 

[Petr]

Thank for answers but this doesn't fully cover my questions. I agree with inadequate cooling, but on the other hand it is majority of time idle and never touched temps listed for desktops (85C). Besides my friend ran overclocked Palomino more than a year non-stop and all the time calculating SETI at temperature of about 80C+. The rubber in corners on that processor burned, but it still runs without any problem. I do not think temperature itself could damage my XP-M so it can not run frequency / voltage any more. The second CPU with water cooling was cooled perfectly and it happened to it too, now with the same voltage can run about 70 MHz lower. The question is could that have been caused by very high frequency for that architecture? Do some XP-Ms have problems with leakage that can damage them while increasing votlage to 1.65V? And is there a way how to prevent this to happen again?

 

[The Coolest]

Actually the higher the temp, the faster the processes of electromigration will occur, and the faster the processor will fail. Just for the record.
My 1400Tbird was a great clock and used to run at 1630MHz at 1.8v, but then I retired it for a while, and when I put it back in I couldn't even hit 1500MHz stably, and it gradually degraded to the point of me running at stock speed had to raise voltage to 1.8v.

 

[Petr]

I know that but the problem is higher voltage causes higher temperature no matter of what sensor says - I speak about local overheat, what is the biggest problem for reliability. It can affect only thousands of transistors so thermal sensor won't catch it, but still be a problem. Many people are running crazy voltages here but not reporting problems. My problem may well be related to power supply not being able to work as good in higher ambient temperature, who knows. It can also be a defect of the chip itself causing heavy leakage - that would answer why it runs so hot. The cooler I use is absolutely adequate for Barton 2500+@3200+ with slightly increased voltage, why it hardly cools down mobile running only 200 MHz faster but with 1.65V ? Maybe it couldn't have been avoided anyway, even with sub-zero cooling.

Recently, I had Sempron 2800+ (2GHz, 1.6V Tbred B0) and it also ran very hot. Compared to it, Celeron D on Prescott core was moderate hot and Sempron 3100+ on Newcastle core was cool like ice. My older Tbred 1700+ was, compared to this 2800+ Sempron, also very cool. Isn't there something wrong with AMD chips lately?