Thread: CPU glitches solved

In order to try to fix my quad-core stuttering I went to Windows 7 Ultimate. After a fresh install the stuttering was still present. After looking closely at my case I found that my rear 250mm fan was dying/dead and barely moving. Also found the mobo temp was way too hot and the CPU temp was just under thermal shutdown. This might be why my computer was shutting down at times with no warning.

In order to save my system I ditched air cooling and went with a Corsair H50 water cooling setup. My idle temps are now 35C to 39C down from 50C to 55C on air. My load temps are now 40C to 45C down from near 60C to 65C. Of course these temps are not internal core temps. One program reported my cores on air at around 55C to 56C. Thermal shutdown for AMD Phenom II's is supposed to be 62C although that sounds awful low.

After playing several games that used to glitch I have found that with the new cooling system none of my games glitch or stutter. NBA 2K11 is smooth as silk as well as a host of other games that used to have problems. The underside of the mobo near the CPU has an obvious different holdout or appearance on the substrate surface than the rest of the board. This is very visible from certain angles which indicates to me that the board has gotten a tad bit warm in these areas.

No more air cooling for me ever again. It is simply not enough to cool down a high performance system.

If you are not overclocking, the stock heatsink + fan has to be able to cool your processor, otherwise the product is defective and you can claim warranty from AMD.

Water cooling is really only for extreme overclocking. Even for mild to high overclocking, air cooling on a good heatsink is usually more than sufficient. A good heat-pipe based heatsink makes a lot of difference.

Congratulations on finding a fix, but I have to say, I've had no problems on stock air for a long time unless the far itself, like yours was bad. I think you would have been fine on air.

Anyway, glad to hear you figured it out.

Soma

Originally Posted by VirtualAce

No more air cooling for me ever again. It is simply not enough to cool down a high performance system.

The stock air cooling solution when used with "Cool and Quiet" is generally adequate for most GP uses such as office tasks or networking. But clearly --as I've said before-- it is not adequate to the task for systems that run constant heavy CPU loads; even at stock speeds.

Some things you can do to help it along, even with water cooling...
Enable "Cool and Quiet" both in BIOS and in your OS (install AMD's driver if necessary).
Set your CPU fan to always spin, ranging from 1/2 speed (abt 7 volts) at idle to full speed (12 volts) at 50c (unless your WC setup does this for you).
Keep the radiator and it's exhaust fan spotlessly clean.

Good to hear you solved the problem.

Originally Posted by Mario F.

Glad you solved it. Although frankly that source of the problem would never occur to me.

It's what I was telling you earlier, Mario... These systems will take a certain amount of heat but they will be in trouble long before thermal shutdown occurs.

1 degree below total failure is not a safety margin.

Originally Posted by bithub

Be very careful with condensation on water cooling setups. I ruined a perfectly good system back in college with a water cooled CPU.

That would be leakage... Modern WC setups are sealed and air-free.

You won't get condensation in a system that never goes below room temperature.

Originally Posted by VirtualAce

The air cooling was sufficient at first but there were other factors involved:

• My case is not too far from a wall on the left side thus limiting air flow
• A bookcase was behind the rear of my computer thus limiting exhaust potential and allowing hot air to come back into the case from the side
• One of my 250mm side case fans died on me. The rear one died and this was the fan that was near the CPU.
• The stupid power supply is situated above the CPU and it has a bottom fan that either draws air in (which would be hot air from the CPU) or blows hot air out (which would be onto the CPU). Really ignorant configuration.
• I'm not the typical user - I routinely push my graphics card, CPU load, and memory to their limits via games and graphics programming.
• My video card and memory are both overclocked from the manuf. which generates more heat than usual.
• My memory sits rather close to my CPU and the heat from the heatsinks really heats up the ambient air which is in the same ambient air space as the CPU.

Actually the best airflow, even with water cooling, is to have cool air enter at the bottom front of the case and exit through the upper back. This moves cool air throughout the entire case and creates an always moving airflow inside.

Some things that everyone will tell you not to do, that are actually very helpful... cover over all unused fan openings... really, no fan; get out the duct tape. If you are not using a duct from center-side openings to your cpu fan cover the side opening. It is also not mere neatness to keep wiring and cables out of the main airflow, even SATA cables can create turbulence that can lead to hot spots; tie them off behind the major airflow paths. The goal is to create a smooth upwards airflow, moving front to back in the case.... Think of it as mechanically assisted convection. And yes, all fans should operate in exhaust mode... that big fan on the front blowing air in probably does 1/4 what the smaller fan on the back blowing air out does.

I cannot verify that all my random shutdowns were due to thermal issues but I'm sure some of them were.

It sure sounds that way...

One question: When you removed the cooler from the CPU did you examine the termal grease to see if the surfaces were well mated (i.e. a uniform thin layer)? Of late I've seen several forced air coolers where if you take a die bar across their bottoms, they are anything but flat. In one case (as an experiment) I grabbed a file and draw-filed the bottom of the heatsink into a nice even diamond pattern, applied new grease, reinstalled and got a temperature reduction of nearly 6 degrees.

As I write this I'm defragging the drive with Diskeeper 2011 and downloading a couple of games from Steam. My CPU temp is 35C and my mobo is 46C. Those are pretty good temps.

Very respectable temperatures.

As for your concern about other causes of spontaneous shutdowns... If, by chance, it was a bad connection you may well have cured the problem accidentally as you re-worked your setup installing the new cooler. I shouldn't worry about it, until it does it again....

Originally Posted by Mario F.

I won't deny that is my own answer to that question. But an unsatisfying one, hence me asking if there's something else.

Sadly most of life is like that. Unless you can settle for the shallow, slow-mindedness that is the meet of the masses, you are doomed to live a life of quiet disappointment. Your only escape is to routinely think beyond the rest, satisfying yourself in intellectual pursuits of a very personal nature while you struggle to not make the rest of the world feel totally stupid in your presence.

Why at least isn't the delta offered as an option in these programs. I've been looking for it -- maybe not hard enough -- and there's only one program that offers this as an option (more below).

Because it's really not any better information than the raw temperature. Think of how it's derived... Take a manufacturer's speck that is NOT guaranteed, get the reading from a sensor that we both know is accurate at only 1 temperature and subtract... "Distance to TjMax" is no more accurate than the input data it is derived from.


I've messed with realtemp ... same thing, relying on sensors that are accurate at only 1 temperature.

Basically all these readings are derived from leakage current through a PN diode that is designed to go into thermal cascade just below the thermal limits of the chip. When it goes cascade, it triggers the thermal throttling... Below that temperature it leaks small amounts of current that are roughly analogous to the temperature it's operating at... Nobody guarantees accuracy at any other than the cascade temperature and I shouldn't be at all surprised to discover that is on a tolerance of plus/minus 5 degrees.

Originally Posted by Mario F.

edit: I swear, as soon as I have some free time I'm going to learn how to read these values and roll out my own processor thermal condition tool.

Placing you squarely into the realm of a hundred other guys with the same ideas...

It's part of the SMBuss, worked through an a/d coverter to give you an arbitrary number that represents some point along the scale from "cold" (minimal diode leakage) to "too hot" (diode cascade). The problem is that cascade "failure" of the kind exhibited by PN Diodes is not linear, it's logarythmic; well sort of... that is to say that for every increase of temperature the diode leakage increases by some amount, the closer you get to cascade the more rapidly the leakage increases... the temperature increase might be linear but the numbers coming off that A/D converter will not be.

As you correctly point out... even the manufacturers are telling us this is not a reliable metric and should be taken, always, with a huge dose of skepticism. The only point at which the PN leakage current is reliable as an accurate standard is when the diode heats up enough to become a short, which triggers the thermal safeties.

It's very informative to take an ordinary PN diode (1n914 will do) measure it's reverse current as you heat it up... we are talking about trying to derive large information from miniscule events... accuracy is not a well known feature in such a process.

Still, if you can crack the problem I'm sure there's a few million self-declared experts who would thank you for it.... but you would, in the end, have to present it in pablum form with some kind of silly bargraph to get them to understand.