Underclocking for lower temps.

[Daniel Byun]

Recently while gaming, I have noticed that my CPU is handling EXTREMELY high temps. My CPU is a AMD A10 6800k running at 4.1 GHz stock, cooled by the stock AMD cooler (Saving up for a new aftermarket cooler). While gaming my CPU was around 90-110 degrees, idle it was 50-80 degrees even though my case has more than enough airflow. My case is a NZXT Phantom 410 with 5 fans. Lowest it ever got was around 45 degrees when my screen saver/AMD Zero core kicked in. I've tried everything from reapplying the thermal paste. But when I went into AMD Overdrive and under-clocked my CPU from 4.1 Ghz to 3.2 Ghz, my CPU has never gotten over 65 degrees. Even during gaming my CPU stands about 60 degrees. The lower I under-clock my CPU the lower the temps get. At 2.0 Ghz the CPU ran about 45-50 degrees constant. Is this normal for a AMD APU? I'm quite concerned.

 

[theadler]

completely normal for any cpu. Lower power, less heat.

 

[natedawg72]

The A10-6800k's maximum operating temperature is supposedly 74 C. That should be read as "the temperature under which it is guaranteed to work" rather than maximum possible.

However, I'm 95% certain that 110 Celsius is beyond the A10-6800K shut off temperature, or whatever you would call it. If your APU was actually hitting that temperature it should have shut down your PC. The idle temperatures are far too high as well, your supposed idle temps are more like what your gaming temps should be. I'd bet that whatever is reporting the temperatures is inaccurate. I have an A10-5800K (which is technically the same chip as the 6800K I believe) that I could pull out, but I don't have the stock CPU cooler. My cooler is only made to handle 65W, so if anything I should see higher temps than you. If you want me to see what temps I get with my cooler I'd be happy to help.

The stuff you are talking about clockspeeds vs temperature is perfectly normal. Changing voltage has a similar effect, if you lower voltage & clock speeds simultaneously you will see even lower temperatures. Voltage will determine stability though, so make sure you know what you are doing if you change that.

 

[hunter315]

Power consumption(heat generation) scales linearly with clock speed, so halving the clock speed should result in about half as much heat getting generated. If you want to effectively reduce temps while still being clocked reasonably high try messing with the voltage, you can generally undervolt a bit. Power consumption goes with the square of the voltage so if you drop the voltage to 90% of its current level your temps should decrease to about 81% of the current level. Note that reducing the voltage may impact the stability of the CPU at stock clocks.

 

[Daniel Byun]

The A10-6800k's maximum operating temperature is supposedly 74 C. That should be read as "the temperature under which it is guaranteed to work" rather than maximum possible.

However, I'm 95% certain that 110 Celsius is beyond the A10-6800K shut off temperature, or whatever you would call it. If your APU was actually hitting that temperature it should have shut down your PC. The idle temperatures are far too high as well, your supposed idle temps are more like what your gaming temps should be. I'd bet that whatever is reporting the temperatures is inaccurate. I have an A10-5800K (which is technically the same chip as the 6800K I believe) that I could pull out, but I don't have the stock CPU cooler. My cooler is only made to handle 65W, so if anything I should see higher temps than you. If you want me to see what temps I get with my cooler I'd be happy to help.

The stuff you are talking about clockspeeds vs temperature is perfectly normal. Changing voltage has a similar effect, if you lower voltage & clock speeds simultaneously you will see even lower temperatures. Voltage will determine stability though, so make sure you know what you are doing if you change that.

I was also wondering, after underclocking my CPU to 3.2 Ghz my CPU is running at around never under 45 degrees and never over 65. Did I get my CPU to a stable temperature at a stable clock? I'm also using CPU-Z's HWMonitor to monitor my temps.

 

[leeb2013]

heat increases with clock speed and voltage, so reducing either will reduce temperatures. Your CPU seems excessively hot though, I'm surprised it's not damaged or shut down at 110C and 50-80C at idle is ridiculous even with the stock cooler. I think you have some other problem there.

 

[Daniel Byun]

heat increases with clock speed and voltage, so reducing either will reduce temperatures. Your CPU seems excessively hot though, I'm surprised it's not damaged or shut down at 110C and 50-80C at idle is ridiculous even with the stock cooler. I think you have some other problem there.

Do you think theres a problem with HWMonitor by CPU-Z? I've heard from some people and reviews online that it sometimes show inaccurate temps.

 

[leeb2013]

I'd heard that's possible with some AMD cpus, can you try another monitor? What does your BIOS say about temps?

 

[randomizer]

With an underclock of that much you could easily undervolt as well for even better temperatures. Cutting performance isn't really a good solution to the problem in the first place though.

 

[Daniel Byun]

With an underclock of that much you could easily undervolt as well for even better temperatures. Cutting performance isn't really a good solution to the problem in the first place though.

Any suggestions to avoid extreme heat?

 

[randomizer]

Any suggestions to avoid extreme heat?

Assuming that the temperature readings are relatively accurate and you've not overclocked the thing to the moon, stupidly high temps nearly always indicate poor contact between the CPU heatspreader and the heatsink base.

 

[Daniel Byun]

Any suggestions to avoid extreme heat?

Assuming that the temperature readings are relatively accurate and you've not overclocked the thing to the moon, stupidly high temps nearly always indicate poor contact between the CPU heatspreader and the heatsink base.

I have checked 2-3 times and reapplied my thermal paste as well as removed all the dust from the heat sink.

 

[randomizer]

I have checked 2-3 times and reapplied my thermal paste as well as removed all the dust from the heat sink.

In spite of your previous reseat attempts it may still not be on properly. However, AMD heatsinks are usually a lot easier to get right than Intel ones due to Intel's use of crummy push pin design (although I've actually had few issues with them even though they are garbage) so it could well be something else. I'm not familiar enough with temperature monitoring of AMD CPUs to give any more advice.

Out of curiosity, you are reading temps in Celsius/Centigrade, right?

 

[Daniel Byun]

I have checked 2-3 times and reapplied my thermal paste as well as removed all the dust from the heat sink.

In spite of your previous reseat attempts it may still not be on properly. However, AMD heatsinks are usually a lot easier to get right than Intel ones due to Intel's use of crummy push pin design (although I've actually had few issues with them even though they are garbage) so it could well be something else. I'm not familiar enough with temperature monitoring of AMD CPUs to give any more advice.

Out of curiosity, you are reading temps in Celsius/Centigrade, right?

Yep. Im measuring these temps in Celsius via CPU-Z's HWMonitor.

 

[randomizer]

That's a shame. If it was in Fahrenheit then this would be a non-issue. Check out Core Temp and see if it gives the same readings.

 

[natedawg72]

Randomizer is taking you on the right track. Your issue is in software, or the temp probes are faulty. As i said earlier, it isn't possible for an a10-6800k to hit 110c. It would have shutdown before it could hit that temperature.

I believe hw monitor shows the a10-6800ks temperature as "TMPN2". What one are you reading?

Many people (me included) have had trouble getting accurate temp readings from an APU. The issue lies in the unusual layout of the sesnsors on the APU, combined with the sensors having names that don't make much sense making it hard to figure out which sensor is actually reporting CPU temp.

 

[Daniel Byun]

Randomizer is taking you on the right track. Your issue is in software, or the temp probes are faulty. As i said earlier, it isn't possible for an a10-6800k to hit 110c. It would have shutdown before it could hit that temperature.

I believe hw monitor shows the a10-6800ks temperature as "TMPN2". What one are you reading?

Many people (me included) have had trouble getting accurate temp readings from an APU. The issue lies in the unusual layout of the sesnsors on the APU, combined with the sensors having names that don't make much sense making it hard to figure out which sensor is actually reporting CPU temp.

Sorry but there is no such thing as TMPN2 on my HWMonitor and I'm reading my temps from my CPU: AMD A10 6800k --> Temperatures --> Package.

 

[natedawg72]

Randomizer is taking you on the right track. Your issue is in software, or the temp probes are faulty. As i said earlier, it isn't possible for an a10-6800k to hit 110c. It would have shutdown before it could hit that temperature.

I believe hw monitor shows the a10-6800ks temperature as "TMPN2". What one are you reading?

Many people (me included) have had trouble getting accurate temp readings from an APU. The issue lies in the unusual layout of the sesnsors on the APU, combined with the sensors having names that don't make much sense making it hard to figure out which sensor is actually reporting CPU temp.

Sorry but there is no such thing as TMPN2 on my HWMonitor and I'm reading my temps from my CPU: AMD A10 6800k --> Temperatures --> Package.

Package temperature is not accurate (at least, not on the A10 5800/6800k), and its not the same as the core temperatures. Sorry for the late reply. Are you still looking into this? I'll help you out in finding the right readings.

 

[Daniel Byun]

Randomizer is taking you on the right track. Your issue is in software, or the temp probes are faulty. As i said earlier, it isn't possible for an a10-6800k to hit 110c. It would have shutdown before it could hit that temperature.

I believe hw monitor shows the a10-6800ks temperature as "TMPN2". What one are you reading?

Many people (me included) have had trouble getting accurate temp readings from an APU. The issue lies in the unusual layout of the sesnsors on the APU, combined with the sensors having names that don't make much sense making it hard to figure out which sensor is actually reporting CPU temp.

Sorry but there is no such thing as TMPN2 on my HWMonitor and I'm reading my temps from my CPU: AMD A10 6800k --> Temperatures --> Package.

Package temperature is not accurate (at least, not on the A10 5800/6800k), and its not the same as the core temperatures. Sorry for the late reply. Are you still looking into this? I'll help you out in finding the right readings.

I'm looking for a backplate less aftermarket cooler because my mobo has a vrm on its back near the cpu which won't allow a backplated cpu cooler to stick on.

 

[LucoTF]

Nate is correct about package temps being inaccurate for the A-series APUs. AMD Overdrive's 'thermal margin' is ok but it only works under load and doesn't give accurate idle temps.

Personally I use the cpu socket temp on the motherboard with HWinfo64.

 

[Daniel Byun]

Nate is correct about package temps being inaccurate for the A-series APUs. AMD Overdrive's 'thermal margin' is ok but it only works under load and doesn't give accurate idle temps.

Personally I use the cpu socket temp on the motherboard with HWinfo64.

I now use a different software made by NZXT called CAM. The same temps also measure if I leave it at stock clocks (4.1 Ghz). Thats is why I underclocked my CPU to 3.2Ghz. It has now never gone over 65 degrees and it stays 45-60 constant.

 

[natedawg72]

Nate is correct about package temps being inaccurate for the A-series APUs. AMD Overdrive's 'thermal margin' is ok but it only works under load and doesn't give accurate idle temps.

Personally I use the cpu socket temp on the motherboard with HWinfo64.

I now use a different software made by NZXT called CAM. The same temps also measure if I leave it at stock clocks (4.1 Ghz). Thats is why I underclocked my CPU to 3.2Ghz. It has now never gone over 65 degrees and it stays 45-60 constant.

If you are happy with it, that is fine, but I can't help but feel this isn't necessary. I have an A10-5800k (same chip as the A10-6800K: it is a rebrand) that I have underclocked to 3Ghz and did some serious undervolting, but that was because it is in an extremely small case with a CPU cooler rated for only 65 watts compared to the A10-5800/6800k's typical 100-110 watts in stressful tasks. Not to mention that the power supply is only 120w max peak, so that PC gave me quite a bit of experience with managing Trinity/Richland temperatures and power.

If you are using a cooler equal to or better than the stock heatsink you should be able to run at default speeds (4.1Ghz) with reasonable temperaturse (<75C, at worst). I still believe that your issue was not with your chip or cooling but with the temperature reading software.

 

[natedawg72]

I'm gonna do a double post. Figured it would be easier to add this way, as it may be a long post.

I ran a number of temperature monitoring programs and then took a screen shot of idle and full CPU load.
Programs:
CAM - temps
HWMonitor - temps
EasyTune6 - Gigabyte motherboard only software, used for temps
LinX - Stress test

Here are the images. I've color coded the important points on the idle picture, stuff is in the same place in the load picture.
idle: http://imgur.com/kd1qqU3
load: http://imgur.com/Zf53zlR

color codes:
Blue: Package Temperatures
Red: TMPIN2. This one may be specific to my motherboard, and gigabyte reports it as CPU temp. From my understanding, this is not real temperatures but instead a relative temperature, which is why the idle temperatures are reported colder than the ambient temperature in my room. What I've heard is that you should not let this exceed 70C.
Green: CPU Voltage.
Black: GPU temperatures (according to HWMonitor).

Here are my observations and what I take from this.

Package temperatures are reported too high to be core temperatures. 3Ghz at the voltage I am using should not idle at 50C, and I know for a fact my heatsink has good contact. Since my case is open, I even touched the heatsink during 100% load and I'd say it was only warm. It's not exactly scientific, but I'd say the heatsink would've been hotter to the touch if it was dissipating the heat from a 62C quad core processor. Another thing to note - the fan, which is controlled by the motherboard and is at default settings, didn't ramp up as far as it should have if the cores were actually at 62C.

TMPIN2 reports temperatures too low, as it suggests the CPU is colder than my room temperature during idle (which is, of course, impossible).

GPU temperatures seem more, erm, true to life than Package and TMPIN2. Despite the fact that the GPU was not stress tested, it still reported idle and load temperature changes since it is on the same die as the CPU.

Now here is what I want you to take a look at: There is something in common with all 3 of these temperature readings. All of them (Even GPU!) rose by the same exact amount during load (9C). GPU temperature was likely a coincidence since it is the neighbor of the CPU, and would likely be +/- 1 or 2 degrees different if I repeated it. But Package and TPMIN2 both seem to be reporting CPU temperature, but they are offset. Which one is most accurate? Honestly, both are probably inaccurate when it comes to real temperature. GPU temperature seems to be closest to what I would expect for real temperature, so I would use Package/TMPIN2 to monitor change in temperature, and check GPU temperature every once in awhile to get an idea of real temperature.

Conclusion: APU sensors accurately report CHANGE IN TEMPERATURE, however, actual temperature appears to be incorrect.

I hope this helps.

 

[Daniel Byun]

I'm gonna do a double post. Figured it would be easier to add this way, as it may be a long post.

I ran a number of temperature monitoring programs and then took a screen shot of idle and full CPU load.
Programs:
CAM - temps
HWMonitor - temps
EasyTune6 - Gigabyte motherboard only software, used for temps
LinX - Stress test

Here are the images. I've color coded the important points on the idle picture, stuff is in the same place in the load picture.
idle: http://imgur.com/kd1qqU3
load: http://imgur.com/Zf53zlR

color codes:
Blue: Package Temperatures
Red: TMPIN2. This one may be specific to my motherboard, and gigabyte reports it as CPU temp. From my understanding, this is not real temperatures but instead a relative temperature, which is why the idle temperatures are reported colder than the ambient temperature in my room. What I've heard is that you should not let this exceed 70C.
Green: CPU Voltage.
Black: GPU temperatures (according to HWMonitor).

Here are my observations and what I take from this.

Package temperatures are reported too high to be core temperatures. 3Ghz at the voltage I am using should not idle at 50C, and I know for a fact my heatsink has good contact. Since my case is open, I even touched the heatsink during 100% load and I'd say it was only warm. It's not exactly scientific, but I'd say the heatsink would've been hotter to the touch if it was dissipating the heat from a 62C quad core processor. Another thing to note - the fan, which is controlled by the motherboard and is at default settings, didn't ramp up as far as it should have if the cores were actually at 62C.

TMPIN2 reports temperatures too low, as it suggests the CPU is colder than my room temperature during idle (which is, of course, impossible).

GPU temperatures seem more, erm, true to life than Package and TMPIN2. Despite the fact that the GPU was not stress tested, it still reported idle and load temperature changes since it is on the same die as the CPU.

Now here is what I want you to take a look at: There is something in common with all 3 of these temperature readings. All of them (Even GPU!) rose by the same exact amount during load (9C). GPU temperature was likely a coincidence since it is the neighbor of the CPU, and would likely be +/- 1 or 2 degrees different if I repeated it. But Package and TPMIN2 both seem to be reporting CPU temperature, but they are offset. Which one is most accurate? Honestly, both are probably inaccurate when it comes to real temperature. GPU temperature seems to be closest to what I would expect for real temperature, so I would use Package/TMPIN2 to monitor change in temperature, and check GPU temperature every once in awhile to get an idea of real temperature.

Conclusion: APU sensors accurately report CHANGE IN TEMPERATURE, however, actual temperature appears to be incorrect.

I hope this helps.

The program I now use is Asrock A-Tuning which came with the mobo. I find it the most accurate and trustworthy