FX-8350 temps look fine... and then spike up to 255° C for no apparent reason

[Dankk]

Just built a new machine recently, with one of the new FX-8350 CPUs. Decided to do some stress testing with Prime95. I had two monitoring programs open; HWiNFO64 (sensor only), and Core Temp.

Under full load with my Hyper 212 EVO cooler, temps slowly crawl upward, until they reach 56° C. The temperature will stay at 56, until suddenly after a few minutes, when it starts spiking up into the hundreds and going crazy. This was the reading I got from Core Temp, and from HWiNFO64 under the "AMD 10h+ CPU Thermal Sensor".

Of course, that's silly and inaccurate. If it were literally running that hot, my computer would've shut down/caught on fire/exploded pretty quickly, right?

I'm not sure what I'm asking for here... I guess I just need some guidance as to how to accurately view CPU temperature. What is the best temp monitoring program for an AMD FX-8350? HWiNFO64 appears to have more than one sensor. Is the "Fintek F71889A/879A" sensor more accurate than the "AMD 10h+ CPU Thermal Sensor"? Which sensor should I be looking at?

Should I be worried?

For the record, my PC still runs smoothly and stably while running the stress test. My CPU fan is pleasantly quiet too.

 

[2is]

Sounds like a bad sensor to me

 

[guskline]

What motherboard?

 

[Dankk]

What motherboard?

MSI 970A-G46

http://www.newegg.com/Product/Produc...82E16813130637

 

[Idontcare]

Of course, that's silly and inaccurate. If it were literally running that hot, my computer would've shut down/caught on fire/exploded pretty quickly, right?

Yeah you'd know right quick if your CPU's silicon actually hit 255°C. It would let the blue smoke out for starters

Primary concern for you to have is what does the CPU's power-controls think is going on, as well as the mobo BIOS?

Because that is where the thermal-throttling decisions are being made, and if those decision centers are convinced the core really is getting that hot then the chip will be throttled and halted regardless how cool it really is.

You should run some benches and compare the results to other published results for the same clockspeed just to confirm you aren't being robbed of performance because of the spurious temperature reading.

 

[FwFred]

255 is 0xFF in hexadecimal... a common value (all 1's) if HW registers return invalid data.

 

[Dankk]

255 is 0xFF in hexadecimal... a common value (all 1's) if HW registers return invalid data.

I noticed this too. 255 is the greatest number representable in 8 bits.

Primary concern for you to have is what does the CPU's power-controls think is going on, as well as the mobo BIOS?

Because that is where the thermal-throttling decisions are being made, and if those decision centers are convinced the core really is getting that hot then the chip will be throttled and halted regardless how cool it really is.

You should run some benches and compare the results to other published results for the same clockspeed just to confirm you aren't being robbed of performance because of the spurious temperature reading.

Unfortunately, it does look like there's some throttling going on. In my unscientific tests involving Prime95 and Core Temp, every time the temperature spikes up to 255°, my CPU clocks briefly decrease from 4000Mhz to 1400 Mhz (and back to 4000Mhz when the sensor corrects itself).

It's interesting to watch how the temp monitoring software works. The CPU temperature will be sitting at a stellar 55° or 56° under full load, and when the sensor freaks out, it will instantaneously jump up to 255°. However, this only last for a few seconds, and when the sensor "corrects" itself, the temperature reading slowly drops one-by-one back to the actual value, rather than spiking back down.

So when the temperature shoots up, it does it instantly; but when the temperature shoots down, the number slowly and gradually decreases. I'm not sure if this is significant, but I thought it was worth mentioning.

So how serious is this problem for actual real-world applications? My PC is 99% used for gaming, and I'm pretty sure there aren't any games that stress the CPU like Prime95 does. At the very most, I'll be rendering HD video in Adobe Media Encoder, and I haven't actually tested that yet on my new PC so I'm not sure what kind of load that puts on the processor compared to Prime95.

Is this the kind of thing that could be fixed in a BIOS update? Some of the Newegg reviews (which I admittedly should've read before purchasing) indicate that this motherboard has some common temperature reporting issues. It's still a fairly new board, and I would imagine MSI will continue to release BIOS updates for it on a regular basis.

Any information would be helpful. I'd like to know what my options are, or how this will actually affect me.

 

[inf64]

Can you verify this happens with other stress-test programs apart from Prime95?

 

[Dankk]

Can you verify this happens with other stress-test programs apart from Prime95?

I did a short 15-minute CPU test with OCCT in Large Data Set mode. I sat and watched intently, looking for temperature spikes and throttling. Surely enough, ~10 minutes into the test, this happens.

Looking at the Frequencies and Temperature graphs, you can see where it screwed up. OCCT automatically stops the test when it exceeds a certain temperature, which is why you see everything dropping back down at the tail ends of the graphs.

This does not look good.

 

[inf64]

Have you tried to update your BIOS?

 

[Dankk]

Have you tried to update your BIOS?

I have indeed flashed my motherboard's BIOS to the latest version. It's one of the first things I did when I set up my PC.

Do you think this is the kind of problem that could be fixed by MSI in the form of a BIOS update? Or is it strictly a hardware problem? I'm hoping it's not the latter; having to RMA new hardware is a major bummer.

If it's a possible BIOS fix, then that makes me hopeful. MSI appears to regularly support and update this motherboard with new BIOS versions; they release them as little as a few days apart from each other.

 

[inf64]

It could be a BIOS issue. Try using some previous version,maybe the latest one you use has some problem with temp reading. As a worst case scenario,it could be an issue within a CPU,but it would be a first of this kind I have heard of when it comes to BD/PD.

 

[Durvelle27]

its a bug as my FX 4100 use to say that when running prime95 until i changed the temp monitoring software.

 

[CKXP]

I have an FX-8320, and the same motherboard, and I also experienced the same exact issue. Try the recommendations from the link below. Right now I have my 8320 @4.6ghz with no more throttling issues.

http://www.overclock.net/t/1202751/amd-fx-cpu-throttling-fixes-please-sticky

btw, how do I load a image?

 

[Idontcare]

its a bug as my FX 4100 use to say that when running prime95 until i changed the temp monitoring software.

Changing your 3rd party temperature monitoring application won't change the fact the system is forcibly throttling itself when the erroneous temperature value is entered into the thermal throttling algorithm and decision loop.

btw, how do I load a image?

You can't load images to Anandtech but you can embed them in your posts provided they are hosted elsewhere on another web server (photobucket, etc).
Use the insert image icon, it looks like

, to embed images. You may have to go to "Advanced" post editor to see that icon.

 

[Durvelle27]

Changing your 3rd party temperature monitoring application won't change the fact the system is forcibly throttling itself when the erroneous temperature value is entered into the thermal throttling algorithm and decision loop.


You can't load images to Anandtech but you can embed them in your posts provided they are hosted elsewhere on another web server (photobucket, etc).
Use the insert image icon, it looks like

, to embed images. You may have to go to "Advanced" post editor to see that icon.

except mines didn't throttle

 

[Idontcare]

except mines didn't throttle

That's great (for you), but that means your situation is not like the OP's and your post telling the OP it is a bug that can be resolved by simply changing the temp monitoring program is not going to help him one bit.

The OP has already reported thermal throttling is occurring...

Unfortunately, it does look like there's some throttling going on. In my unscientific tests involving Prime95 and Core Temp, every time the temperature spikes up to 255°, my CPU clocks briefly decrease from 4000Mhz to 1400 Mhz (and back to 4000Mhz when the sensor corrects itself).

...and your response/solution is that "it is a bug, change your temp monitoring program"

its a bug as my FX 4100 use to say that when running prime95 until i changed the temp monitoring software.

 

[Durvelle27]

no i suggest getting the latest BIOs avaible for your motherboard and that should fix it

 

[CKXP]

From what I have read, it's not a temperature issue causing these FX processors to throttle but due more to the power draw of Bulldozer/Piledriver and a limitation of some motherboards to cope the excessive power draw that these chips are capable of when stressed, causing the boards VRM's to overheat, in return causing these chips to throttle(we that's one plausible theory that gets tossed around.)

This occurred while playing BF3. I use HWmonitor to check my temps, and they maxed out at 48c while playing a few rounds. Also unlike the OP, this occurred while my 8320 was oc'd to 4.6ghz, my temps were fine, but yet my 8320 started throttling. In the link I posted earlier, I disabled some of the power saving features, and now this issue no longer occurs for me.

 

[Dankk]

I have an FX-8320, and the same motherboard, and I also experienced the same exact issue. Try the recommendations from the link below. Right now I have my 8320 @4.6ghz with no more throttling issues.

http://www.overclock.net/t/1202751/amd-fx-cpu-throttling-fixes-please-sticky

From what I have read, it's not a temperature issue causing these FX processors to throttle but due more to the power draw of Bulldozer/Piledriver and a limitation of some motherboards to cope the excessive power draw that these chips are capable of when stressed, causing the boards VRM's to overheat, in return causing these chips to throttle(we that's one plausible theory that gets tossed around.)

This occurred while playing BF3. I use HWmonitor to check my temps, and they maxed out at 48c while playing a few rounds. Also unlike the OP, this occurred while my 8320 was oc'd to 4.6ghz, my temps were fine, but yet my 8320 started throttling. In the link I posted earlier, I disabled some of the power saving features, and now this issue no longer occurs for me.

Wow, what are the chances of running into someone on this forum with the exact same motherboard, having the exact same problem? Thank you for the link. Come to think of it, I know I left a lot of those CPU power settings to "Auto" in the BIOS without doing any fine-tuning. I'm at work right now but I'll definitely take a look at those when I get home.

I guess my only concern is: Is it really necessary to disable *all* of those settings? Couldn't I just disable the "CPU Thermal Throttle" option and call it good, or is there something about those other options (such as Core C6 State or C1E) that are problematic? Keep in mind, I'm not overclocking.

Although... now I am curious about overclocking just a wee little.

This is off-topic, but does the FX-8350 have a little bit of headroom for increasing clock multiplier without increasing voltage? I remember my older Phenom II 965 essentially had a "free" 200Mhz that I could overclock it by, just by increasing the clock multiplier and not messing with anything else. It was the highest I could go while my power-saving features still worked (mainly Cool 'n' Quiet). It's small, but it's something.

Does the FX-8350 have something like this? Should I just increase the clocks a tiny bit at a time and experiment?

 

[inf64]

@ LessThanDan

Yes, there is some headroom on stock vcore. You can expect to reach at least the Turbo cock on all cores(so 4.2Ghz). With good cooling you may reach even a little bit over 4.2Ghz,it depends on the CPU specimen. But to get over 4.4/4.5Ghz you will need to increase the Vcore (not sure how that will work out with the motherboard issues you are having).

 

[Dankk]

Well, I followed the suggestions on this page (both the BIOS fix and the AMD Overdrive fix):

http://www.overclock.net/t/1202751/amd-fx-cpu-throttling-fixes-please-sticky

Nothing worked. My processor is still reportedly being throttled as a result of temperature spikes. I'm a bit frustrated.

CKXP, can you go into detail on the exact steps you took to fix your problem? In my motherboard, the throttling feature is not called "CPU Thermal Throttle"; rather, the option is titled "CPU Smart Protection" (and disabling it still does nothing). Maybe there are some minor differences specific to this model of MSI motherboard I need to take into account. How exactly did you get yours to not throttle?

 

[CKXP]

Well, I followed the suggestions on this page (both the BIOS fix and the AMD Overdrive fix):

http://www.overclock.net/t/1202751/amd-fx-cpu-throttling-fixes-please-sticky

Nothing worked. My processor is still reportedly being throttled as a result of temperature spikes. I'm a bit frustrated.

CKXP, can you go into detail on the exact steps you took to fix your problem? In my motherboard, the throttling feature is not called "CPU Thermal Throttle"; rather, the option is titled "CPU Smart Protection" (and disabling it still does nothing). Maybe there are some minor differences specific to this model of MSI motherboard I need to take into account. How exactly did you get yours to not throttle?

I took some pictures of my bios settings, hope this helps OP. IMHO I believe your issue maybe more thermal related than a power draw issue, since your throttling occurs while running at stock. One more note, even though it's not shown in any of the pics, I have my NB voltage set 1.3v instead of [auto] by default, this made a difference with stability.

 

[Dankk]

I know it's been a while, but I'm resurrecting this thread. I've found out exactly what the problem is, right out of MSI's mouth.

Basically: The 970A-G46 has an inherent problem with FX-8350 CPUs, involving an overheating MOSFET/VRM. It can't handle the power draw of the FX-8350 so it throttles processor speed by tricking the temeprature sensor into reporting 255 C.

Also, I'm a moron for not doing research on this board before buying it.

Here's an email from MSI:

Hi Dan,

According to RD engineers from our HQ,

1. AMD has claimed that the FX-8350 to be 125W. However, during our internal testing, when the usage of CPU reaches 100% through Prime 95 for a while, the power consumption can exceed 125W and reaches 140W. With such condition on the 970A-G46, the high amount of power draw also causes the MOSFET to exceed its spec and will overheat.

Thus, to prevent such any potential hazard to the MOSFET, 970A-G46’s BIOS will trick the processor that its temperature is 225 degrees which will then allow the CPU to throttle. This is aimed to make the system stable in this condition.

2. We had tested using 3DMark and it did not cause throttling. Throttling only happens when the loading on a 125W CPU usage is heavy by the use of heavy burn-in tools such as Prime 95/OCCT, and such testing methods are not standard usage scenario or practical.

Please also keep in mind that this throttling behavior keeps system stable under such heavy-loading condition.

This is more of AMD’s issue on the FX-8350 because the TDP was rated lower than its actual value. http://forums.anandtech.com/showthread.php?t=2289809&page=2

Currently, only the 990FXA-GD65/GD80 can handle up to 140W TDP, see below.

http://www.msi.com/product/mb/990FXA-GD65.html#?div=CPUSupport

If you really need 140W boards such as 990FXA-GD65/GD80 then we can offer you a courtesy step-up option by paying for the differences.

Please note that your current board will need to come in through RMA service, you are responsible for 1-way shipping sending the board to us, and the replacement board will be factory-rectified board but not brand new.
The only accessory that will come with the replacement board is the I/O shield.

990FXA-GD65 - $40
990FXA-GD80 - $70

Let me know if you have any questions, thank you.

Sincerely,
MSI Service & Support Division

So basically, if I really want to, I can send in my board for a better one, and I'll only have to pay the difference. Not sure if it's worth the trouble though... very rarely do I actually use applications that make my CPU load go up to 100% (Adobe Media Encoder is the only thing I can think of).

So while the issue doesn't pertain to me a whole lot, I'm just kind of bothered by the principle of it all. A motherboard should be able to properly handle 100% CPU load for more than a few minutes. But this board can't.

I'm kind of torn. I don't *need* a better motherboard, but it would still be nice... :hmm:

 

[cytg111]

Sounds like a bad sensor to me

Yea, something just FF'ed that memory location, being bad hardware(sensor) or bad software, who knows.