Overlocking A6-3670K And OCCT

[Locknload]

Hello guys, new guy here. My name is Erick.

I come to you seeking advice and answers that I haven't been able to find regarding a couple of subjects.

First of all. I own an AMD A6-3670K that runs stock at 2.7GHZ.
I wanted to try overclocking it since I started playing FSX, and it's a CPU intensive simulator, so I read that increasing your clock would increase FPS.

I ran ATI's autotune that comes with it's Catalyst Control Center, the program took the CPU up to 3.5GHZ before getting a blue screen. I promptly restarted my computer, and set the new clock to 3.2 Ghz, I haven't had a crash since. I didn't make any stress test whatsoever, but it's been 2 days, and everything seems good.

I inquired about raising my clock, and how raising the voltage could lead me to clocking to a higher speed, and I was recommended to run OCCT to check for current stability, temperature, etc..etc...

However, upon opening OCCT the system information/cpu tab showed that my current clock speed was 2.7Ghz which is stock, and that my original was 3.2. Core Temp shows that my current speed fluctuates from 2.4 to 2.7 to 3.2. And CPUZ shows it constantly at 3.2

When I ran OCCT for 20 minutes first, the max speed was stuck at 2.7003mhz, and the temperature was 63.50. I really don't understand what exactly is going on, and what it means when OCCT shows my current speed at 2.7 I thought I had overclocked this thing?

What should I do?

Thanks for your time, please ask for any information you need about my system, I will provide screenshots if needed.

 

[Automata]

I would trust whatever CPU-z is reporting as your speed to be correct. If it says 3.2, then you are at 3.2.

For overclocking, you really need to do this properly. That means going full manual and working your way up. Letting software do it for you is prone to instability (as you've seen) and it doesn't teach you how the system functions.

Here is the basic testing procedure that I do:
1) At stock speeds, lower the voltage
2) Test for stability, if it passes, go to 1. If it fails, go to 3.
Once I know how low I can go, that gives me an idea how it will respond to overclocking.
3) Up the voltage slightly. Bump the multi up one or two values.
4) Test for stability. If it fails and temperatures are within safe limits, go to 3. If it passes, increase the multi and test again.

If you make large jumps when overclocking, you lose your bearings on how the system is responding. You could certainly give it a whole bunch of voltage and then start upping the multiplier in large amounts, but you have no idea where your instability line is. That makes finding your final overclock harder than it needs to be. Once you learn how this works, you can start cutting corners, but that is for another time.

 

[Locknload]

Thank you Thideras.
A couple of question.
After I lower the voltage at stock speed (2.7ghz) should I run the test in OCCT?
For how long? The test has a default of 1 hour tests in unlimited settings so the test stops if there is an error.
What is a safe temperature 60 70 Celsius?

Thanks for your time!

 

[Automata]

Yes, if you make any changes to the frequency or voltage (or timings, in the case of memory), you should always test for stability. How long you run it is a function* of how stable you want your system to be. If you use it for non-important tasks and crashing isn't very important, a couple hours would probably be enough. Personally, I need my system to be 100% stable because I'm running virtual machines, compiling code, and generally just need my system to work all the time. Because of this, I run my testing for 24+ hours.

I don't know the specifics on Llano, but it says the max operating temperature is 72c. I would keep it below 65c.


*There is no way to know if a computer is 100% stable short of running stability tests until the computer dies. We can say it is mathematically insignificant by testing for longer periods. The longer you run a stress test (and the harder the stress test is on your hardware), the less chance you will run into problems later on.

 

[Locknload]

Yes, if you make any changes to the frequency or voltage (or timings, in the case of memory), you should always test for stability. How long you run it is a function* of how stable you want your system to be. If you use it for non-important tasks and crashing isn't very important, a couple hours would probably be enough. Personally, I need my system to be 100% stable because I'm running virtual machines, compiling code, and generally just need my system to work all the time. Because of this, I run my testing for 24+ hours.

I don't know the specifics on Llano, but it says the max operating temperature is 72c. I would keep it below 65c.


*There is no way to know if a computer is 100% stable short of running stability tests until the computer dies. We can say it is mathematically insignificant by testing for longer periods. The longer you run a stress test (and the harder the stress test is on your hardware), the less chance you will run into problems later on.

Where did you find the information about Llanos temperature range?
I have never been able to find a trusted source, and the information that came with my CPU did not say anything, and the internet is full of people saying that Llano runs cool, others say it runs hot.

I don't know if the fact that mine is an unlocked 4 core 100w cpu changes it's temperature spectrum. I ran it with AMD'S overclocking program for two days, and the temperature under full load reached 68. Then again.. it has been pretty hot here recently.

 

[Automata]

This is actually the only place I see it listed: http://www.cpu-world.com/CPUs/K10/AMD-A6-Series A6-3670.html

The number of cores won't change the maximum temperature it can handle. It will change how fast you can run it or how much voltage you can push.

 

[Locknload]

This is actually the only place I see it listed: http://www.cpu-world.com/CPUs/K10/AMD-A6-Series A6-3670.html

The number of cores won't change the maximum temperature it can handle. It will change how fast you can run it or how much voltage you can push.

Thank you, I a running OCCT right now at stock speed and reduce the voltage .025 values. The average temperature at full load is 65 Celsius.

Idle it goes down to 25 Celsius, it is a big change in temperature, I get the feeling something else is wrong, maybe the heatsink is not properly installed, bad temperature readings.. I don't know.

 

[Locknload]

Alright. I got the temperatures down by A LOT.
I took the heatsink apart, cleaned both fans, took dust out of the radiator, applied new thermal paste, gave the computer a start up run to warm up the TIM. Checked Idle temperatures, 17 Celsius idle. 25 while browsing the internet.

And Running OCCT, with standard clock at full load is giving me a huge difference.
I am getting 55 to 58 Celsius under load. Previously I would get 65 at stock speed. I'm assuming dust, and most likely incorrectly applied thermal paste were big factors in the systems temperature readings.

I will start lowering the voltage until the test finds an error, and later start bumping up the multiplier when I find out how much I can raise my voltage.

 

[Automata]

Idle temperatures you have are inaccurate are not possible. 22c is 72f. It can't be below ambient temperature.

 

[Locknload]

Idle temperatures you have are inaccurate are not possible. 22c is 72f. It can't be below ambient temperature.

Well.. these are the programs that support temperature monitoring..
Are all of these wrong readings and could my cpu have a faulty temperature sensor?

http://i49.tinypic.com/25us9yq.png

 

[Automata]

The readings are wrong. If it is the same as the Intel chips, the farther away the chip is from the highest allowable temperature, the more inaccurate it is. It doesn't mean anything is wrong, just that you can't trust temperature while not under load.

 

[Locknload]

The readings are wrong. If it is the same as the Intel chips, the farther away the chip is from the highest allowable temperature, the more inaccurate it is. It doesn't mean anything is wrong, just that you can't trust temperature while not under load.

I understand what you mean now.
Besides that, I think the temperatures under full load are an improvement over what they were before, I will attempt to start manually overclocking soon.