cancel
Showing results for 
Search instead for 
Did you mean: 

Temp reporting shows decrease when temp is actually going up

poisonsnak
Level 7
If you are having trouble overclocking, and your system is crashing in spite of pretty low CPU temps (the highest I've seen is 45 in Ryzen Master / HWMonitor), your system is probably running a lot hotter than you think it is.

I have a 1700 non-x and this is how my stability test runs usually go. I've been using AVX linpack through OCCT 4.5.0 as it seems to fail the fastest when my CPU is not stable. Other tests can run for 30 min or even hours when this fails after a few minutes:
- temp starts out showing 15 C
- temp climbs to 45 C over a couple of minutes, fans speed up
- temp slowly drops to 35 C over the next 10 min or so, fans slow down
- system crashes

Thinking I had lots of voltage headroom because the temps were so low, I turned it up but if anything I was just crashing sooner. I stopped at 1.45V as it seems like a pretty safe max.

Then I started looking at the data and realizing it made no sense. Why is the CPU getting colder as the fans slow down and the test runs on? It should be getting hotter. Was it throttling? HWMonitor showed the CPU power, voltage, and clock were rock solid and the temps were so low that couldn't be it.

So I borrowed a FLIR infrared camera from work and used it to watch the temperature on the back of the motherboard, right underneath the CPU. Imagine my surprise when the temp increased steadily up to 75C (so the CPU on the other side must have been hotter yet) as the test went on, even though the software readings all showed it dropping, down to 35C.

Normally I hate software fan control programs but the motherboard fan control clearly could not work if it didn't know if the CPU was getting hotter or colder. So I installed Corsair Link (I am using an H110i). The pump/block has a temp sensor in it and two fan headers so it can control its fans based on the water block temp. Sure enough the H110i shows the temperature steadily increasing just like the FLIR did. Now at least once the CPU got hot, the fans would speed up accordingly, and now I've also realized I don't have nearly as much voltage headroom as I thought.

I just wanted to post this as a PSA as I've seen people running really high voltages (1.5 - 1.6) and looking at their temps thinking they are okay since their temps are low. I don't know if it's the motherboard or the CPU but the temp reporting is not very good, I know about the 20C offset thing but for one, I am running a non-X which doesn't have that issue, and for another, even if there was an offset the temp should not be going DOWN as the test runs on. Something is clearly wrong.
4,319 Views
9 REPLIES 9

Waza88
Level 7
poisonsnak wrote:
If you are having trouble overclocking, and your system is crashing in spite of pretty low CPU temps (the highest I've seen is 45 in Ryzen Master / HWMonitor), your system is probably running a lot hotter than you think it is.

I have a 1700 non-x and this is how my stability test runs usually go. I've been using AVX linpack through OCCT 4.5.0 as it seems to fail the fastest when my CPU is not stable. Other tests can run for 30 min or even hours when this fails after a few minutes:
- temp starts out showing 15 C
- temp climbs to 45 C over a couple of minutes, fans speed up
- temp slowly drops to 35 C over the next 10 min or so, fans slow down
- system crashes

Thinking I had lots of voltage headroom because the temps were so low, I turned it up but if anything I was just crashing sooner. I stopped at 1.45V as it seems like a pretty safe max.

Then I started looking at the data and realizing it made no sense. Why is the CPU getting colder as the fans slow down and the test runs on? It should be getting hotter. Was it throttling? HWMonitor showed the CPU power, voltage, and clock were rock solid and the temps were so low that couldn't be it.

So I borrowed a FLIR infrared camera from work and used it to watch the temperature on the back of the motherboard, right underneath the CPU. Imagine my surprise when the temp increased steadily up to 75C (so the CPU on the other side must have been hotter yet) as the test went on, even though the software readings all showed it dropping, down to 35C.

Normally I hate software fan control programs but the motherboard fan control clearly could not work if it didn't know if the CPU was getting hotter or colder. So I installed Corsair Link (I am using an H110i). The pump/block has a temp sensor in it and two fan headers so it can control its fans based on the water block temp. Sure enough the H110i shows the temperature steadily increasing just like the FLIR did. Now at least once the CPU got hot, the fans would speed up accordingly, and now I've also realized I don't have nearly as much voltage headroom as I thought.

I just wanted to post this as a PSA as I've seen people running really high voltages (1.5 - 1.6) and looking at their temps thinking they are okay since their temps are low. I don't know if it's the motherboard or the CPU but the temp reporting is not very good, I know about the 20C offset thing but for one, I am running a non-X which doesn't have that issue, and for another, even if there was an offset the temp should not be going DOWN as the test runs on. Something is clearly wrong.


I just found the SenseMI Skew setting in the bios of C6H and turned it off. Now My CPU temps are reported 20c higher. This will make more sense when tuning fan control.
I think that the Motherboard doesn't properly set the offset skew for R7 1700. 1700x and 1800x have +20c skew in place from the factory and this setting is fixing it. What ends up happening with R7 1700 is that temps are reported too low unless this setting is set to disabled.

I think not having Corsair Link installed could have been your issue.

I am using a H115i, and without Corsair link installed the pump and the fans that are connected to it only operate at their default speeds. The software is needed to control pump and fans over the USB connection to the pump.

eight-bit wrote:
I think not having Corsair Link installed could have been your issue.

I am using a H115i, and without Corsair link installed the pump and the fans that are connected to it only operate at their default speeds. The software is needed to control pump and fans over the USB connection to the pump.


I briefly tried that Link software and removed it almost immediately. My fans & pump work just fine in response to temperature changes. When I still had my OC @ 4.0Ghz it would spin up as temps increased right along the fan curve I specified and via HWInfo I saw the pump speed increasing in addition to the fans speeding up. There was no mistaking the fans as it sounded like a jet engine when the proc was going full bore.

I run @ 3.8Ghz now and my temps rarely exceed 40C under load, and even under full BOINC load they very rarely see 50C. At these temps the H115i barely needs to increase at all to control the temps. The vast majority of the time my system is barely making any noise, and when it is most of it is from my old 7950.

MysticPixie wrote:
I briefly tried that Link software and removed it almost immediately. My fans & pump work just fine in response to temperature changes. When I still had my OC @ 4.0Ghz it would spin up as temps increased right along the fan curve I specified and via HWInfo I saw the pump speed increasing in addition to the fans speeding up. There was no mistaking the fans as it sounded like a jet engine when the proc was going full bore.

I run @ 3.8Ghz now and my temps rarely exceed 40C under load, and even under full BOINC load they very rarely see 50C. At these temps the H115i barely needs to increase at all to control the temps. The vast majority of the time my system is barely making any noise, and when it is most of it is from my old 7950.


Did you disable Sense MI Skew?
Without that disabled, I get lower than actual temps on my 1700 in Windows.*

alexjames wrote:
Did you disable Sense MI Skew?
Without that disabled, I get lower than actual temps on my 1700 in Windows.*


I'd have to go into my BIOS to check but I don't recall fiddling with that setting. Everything reports the same temperatures though (or within a couple of degrees anyway), my radiator is cool to the touch and the air coming out of it is fairly cool as well.

The PCH on the board gets quite a bit hotter. Need to rig up a fan for it. Proc stays very chill though. I could hit it with a temp gun but just given the overall performance and ambient feel of the case I'm pretty sure the readings I'm getting are accurate.

MysticPixie wrote:
I'd have to go into my BIOS to check but I don't recall fiddling with that setting. Everything reports the same temperatures though (or within a couple of degrees anyway), my radiator is cool to the touch and the air coming out of it is fairly cool as well.

The PCH on the board gets quite a bit hotter. Need to rig up a fan for it. Proc stays very chill though. I could hit it with a temp gun but just given the overall performance and ambient feel of the case I'm pretty sure the readings I'm getting are accurate.


Fair enough. How hot does the PCH get?

My old MSI Z97 Gaming 3 got to around 90-95C overclocked hahaha.
The CH6 PCH hits around 68C max for me.

alexjames wrote:
Fair enough. How hot does the PCH get?

My old MSI Z97 Gaming 3 got to around 90-95C overclocked hahaha.
The CH6 PCH hits around 68C max for me.


That's about where mine hangs our. I'm just sitting here watching Netflix and checking a couple of websites (though I was playing some SC2 about 20 minutes ago), and it's sitting at 67C. Meanwhile my CPU is sitting at just over 19C.

Is the PCH the glowy ROG emblem right next to the SATA ports?

My problem is there isn't much air flow in my case just given its design. It's designed with lots of custom liquid in mind. http://imgur.com/a/whyiJ

BaneSilvermoon
Level 7
My temperatures were all out of whack with the early BIOS versions. If I over-clocked and left SenseMI on, my processor idle temp was often in the negative range. Turning off SenseMI fixed that on those BIOS. And in the more recent versions I've left it on and my temperatures are normal now.

Decoman
Level 7
I find it odd that OP has the area "below" the cpu on the backside of the mobo to be hotter.

With my Crosshair VI Hero board, it is the opposite. Using a cheap laser thermometer, with soc temp at 70 I measure 60 deg C at the top of the metal backplate, and something like 30 deg C at the lower part of the metal plate. All the tiny components in that hole in the backplate are no hotter than 60 deg C iirc.

Btw, unsure if relevant, but I have "sense skew" enabled and set to 272 in the bios.
I am using a passive cpu cooler, so I have an interest in trying to keep things less hot at stock speeds.

I wonder if maybe the differences in heat here on the backside might be indicative of an improper cpu cooler contact.