Death of an FX-8350

[jvroig]

Has AMD published the temperature offset yet? If not, then you have no idea what temperature your CPU is running at.

Still no official specs from AMD, temps and voltage.

Although whatever the actual temps are (and they don't matter since no max temps are published anyway), since hwmonitor, core temp, the BIOS, and the linux sensors (lm-sensors) show the same temps (only ~1-2C variance max), then we can just assume offset to be any constant value, since if I read, for example, 70C, then a similar rig using similar sensor-reading software, will also read 70C under similar conditions. We will both say "70C", since that is the readout, regardless of whether a 70C readout means an actual temperature of 50, 60, 70, 80, or 90C.

So all 8350 temps cited by forumers the world over are comparable to each other, although it won't necessarily be telling us any info about actual temps. So when I say, for example, that "76C killed my processor", all other enthusiasts can still use that as a data point, even though none of us know what the temp offset it, because all our temp reading gizmos show the same temp values under the same conditions. Whatever the temp offset is, they will then, just as an example, avoid the 76C readout and limit temps to not go near as high as that. (Of course, this is my assumption, that all mature boards and temp-reading software do actually produce comparable readouts of the temps, based only on my own experience with two Piledriver SKU's, one mobile and one desktop, on Windows and Linux)

As for using the 76C readout and the 100C max temp spec on the A4 4300m upon which I based my decision to let the test continue, the Phenom II behavior + the 4300m temp measurements of HWMonitor more or less allowed me to guess (again, without published info, everything we can do is guess as smart as we can) that the figure HWMonitor reads out is the same figure that the AMD specs say. For example, 62C was specced for the Phenom II chips. When temps do breach 62C according to HWMonitor, the CPU starts failing stress tests (be it IBT or Prime), so we see a strong correlation there, regardless of whether "62C" is an actual temperature (as in the silicon is actually 62C hot) or 62C only refers to a readout in the sensors that refers to the max safe limit, which could actually mean an actual temp higher or lower than 62C.

Anyway, just guesses. No published info on anything, so most everything is just guessing based on all other provided info that is even tangentially related to OC'ing an 8350.

 

[jvroig]

404 monumental mess-up not found

Thanks for the assist, Phil

My FX-8350 does not throttle either. Goes all the way to 83C before the computer simply shuts down.

Hmmm... that's interesting. Could you check or do you remember if it has anything to do with HPC Mode being on or enabled? (Do you even have an HPC Mode on your BIOS? I'm not even sure if it is MSI only, or a generic feature of every provider, and I know yours is an ASUS). With HPC Mode disabled, I think I actually see it throttle a little every now and then (but only for a second or two before resuming max clocks) under P95 L-FFT, and that was the behavior I wanted to "cure" by turning on HPC Mode (because temps were still very good, 55-56C only). However, I thought I was only turning off some sort of max core clocks limiter, not actually the entire temp throttling safety mechanism. No matter how much I googled, there simply wasn't any official documentation about HPC mode in FX boards, so yet again no way to actually make an informed decision based on official info there. Based on the remainder of your posts though, I guess now I know what exactly HPC Mode does (more on that below).

I even asked AMD support and they said there are no internal documents that even they can look at to give me guidance, unofficial or otherwise.

Ah, terrible. That's very unfortunate.

When I contacted ASUS support (my mobo maker) they said AMD gave them no specs on operating temperature. It is completely up to the mobo makers as to whether or not the CPU starts throttling at any given temperature - the throttling is not because of temperature but because of current draw.

If you see an AMD PD-based CPU throttle at 60C it is because it is already hitting the mobo makers internally defined current limit. That's it, nothing more. If another mobo maker allows more current before hitting the limit then your PD-based CPU might go to 70C before being throttled.

This is all very unfortunate and a bit surprising. Based on this, what HPC Mode probably did was simply turn off the current draw limiter, so any sort of throttling will likewise be disabled.

This does answer why Durvelle and Erenhardt were sure of 2 different temp throttling points. Their respective Piledriver rigs most likely had different mobos with different throttling limits preset.




At the end of the day, I still want a Piledriver-based CPU to play with, and I don't think the 4300m will cut it (no frequency controls and only 1 module), so I'll probably get another Piledriver SKU. Perhaps a 6300 is in the future for this rig (again shelving the 1090T), or I'll let the 1090T live out its life and just get another rig based on Richland/Trinity, or maybe just Kaveri if I can wait for it.

 

[USER8000]

Actually the earlier Bulldozer based CPUs had their maximum temperatures published by AMD.

Here are relevant webpages:

http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=
http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=
http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=
http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=
http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=

Depending on the model,it was either 61C,61.1C or 70C.

Also,could it be the motherboard which is the source of the problems?? MSI ones are known to the most problematic ones especially with AMD AM3+ CPUs.

 

[jvroig]

Actually the earlier Bulldozer based CPUs had their maximum temperatures published by AMD :

http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=
http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=
http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=
http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=
http://products.amd.com/en-us/Deskt...2=&f3=&f4=&f5=&f6=&f7=&f8=&f9=&f10=&f11=&f12=

Depending on the model,it was either 61.1C or 70C.

Indeed, in fact I touched on that in the "Long Story" portion of the OP (where a lot more details are present, compared to the "Short Story" version, which is why it is longer), and I quote:

Here's what I was thinking. AMD doesn't publish max temps AND vcore ranges for this part. If I go by the FX-8150 (direct ancestor), max temp should only be 61C (but at what range of operating voltages? AMD doesn't publish). But Trinity mobile SKU's (like the A4-4300m I benched in VC&G) sporting the same Piledriver cores have a much higher limit (sites like CPUBoss and CPU World specify 100C, despite a vcore range of up to 1.35V). So, the same old Piledriver cores can take 1.35V and 100C? Well, if so, I'm still relatively ok (1.396V, 76C).

 

[USER8000]

At least going from other people I know who overclocked AMD CPUs,they tend to keep to 60C to 65C at most and preferably under 60C when overclocking.

Having said that some of the stress test programmes you use,might be not representative of what you run,so you might never ever see such temperatures although YMMV.

 

[Erenhardt]

This does answer why Durvelle and Erenhardt were sure of 2 different temp throttling points. Their respective Piledriver rigs most likely had different mobos with different throttling limits preset.

My mobo is cheap asus mATX: M5A78LM-LX up to 95W CPU.

 

[OatisCampbell]

I would try and RMA it -- yeah the fine print ofc reads about how oc'ing and out of spec usage blah blah blah. But this chip is MARKETED to hit like 5 GHz (http://sites.amd.com/us/promo/products/Pages/amdfx.aspx) check out the webpage. It drones on and on about oc'ing it: No premium to pay. No code to input. Just unrestrained, overclockable power right away. In fact *ALL THREE* bullet points note OC, OC, OC, unlocked, unlocked unlocked.

So if AMD advertises the chip in this fashion, and also has tech issues with the chip shutting down at certain limits, AND often has poor temp sensors that make it difficult to determine what temp you're running at AND doesn't note max voltage / temps and what not... then I would RMA away. You are using the chip in the manner designed and the bs fineprint is just that. Chips should be advanced enuff in this day and age to throttle down to prevent all but the most insane situations. Plus if this is one of the only times someone had an FX chip die it sounds like it could have been a bunk chip.

I've got a hankering that perhaps the current FX 8350 chips are a little shittier than the ones they HAD been making as the higher binned ones are used as 9xxx's. Same thing happened when intel released the 2700k -- scooped up all the nice 2600ks and called them the 2700k. So if the OC'ability of your chip went down the drain as they wanted to squeeze more money out of consumers then god forbid you get a replacement.

Rationalize it anyway you like, but to RMA this chip (because it was over volted) is theft.

The fact AMD does some binning and the fact the multiplier is unlocked does not mean "Why, they MAKE these chips to run at 5GHz and any voltage- every single one!"

A better car analogy would have putting a supercharger on your new Challenger SRT8 to force more air and fuel into the cylinders, then RMAing it "because it's marketed to be a fast car, and runs on air and fuel, doesn't it?".

 

[Idontcare]

This is all very unfortunate and a bit surprising. Based on this, what HPC Mode probably did was simply turn off the current draw limiter, so any sort of throttling will likewise be disabled.

This does answer why Durvelle and Erenhardt were sure of 2 different temp throttling points. Their respective Piledriver rigs most likely had different mobos with different throttling limits preset.

That is exactly the situation.

From the mobo maker's perspective, you choosing to use their product to invalidate your AMD warranty is not their concern or problem.

Had they labeled the BIOS option "AMD Warranty Invalidator" instead of "HPC" then you'd probably be a tad more reserved in electing to enable that "feature" in the BIOS

 

[Idontcare]

Rationalize it anyway you like, but to RMA this chip (because it was over volted) is theft.

The fact AMD does some binning and the fact the multiplier is unlocked does not mean "Why, they MAKE these chips to run at 5GHz and any voltage- every single one!"

A better car analogy would have putting a supercharger on your new Challenger SRT8 to force more air and fuel into the cylinders, then RMAing it "because it's marketed to be a fast car, and runs on air and fuel, doesn't it?".

This.

There is no moral or ethical justification for warranty fraud.

There is, however, plenty of unethical and amoral justification for warranty fraud to be found.

 

[jvroig]

Had they labeled the BIOS option "AMD Warranty Invalidator" instead of "HPC" then you'd probably be a tad more reserved in electing to enable that "feature" in the BIOS

Haha, yeah, hmmm... not so much in that case, although I can't guarantee I wouldn't try it for the heck of it anyway (I call the writ of boys will be boys /Simpsons)

Well, at least you didn't buy an FX 9590 and have this happen!

Hey Steve! Yeah, haha, that would be a bit more stressful, I imagine.

Hey, you still owe me that review of your v2 MSI 990FXA-GD80 when you get that rig up and running!

 

[Magic Carpet]

I'd still send the chip back to AMD to find out the reason, "why my experience has been so drastically different to the others". AMD, out of good will, might send a replacement. No fraud involved whatsoever

 

[aigomorla]

We are able to disable all the protecting instructions and put the CPUs straight into hell. Your imagination is your limit...

And we are also allowed to do that on Intels too.
My board starts flashing yellow and red numbers as u get closer to hell.

I'm a bit confused by your combined statements.

First, you tell me you've seen guys pump higher volts AND temps, drawing the conclusion I just had a lousy chip because it shouldn't have died. (Very possible, of course)
Most Definitely Yes... Your actually still in comfort zone from my understanding

But then you end with your own experience that 8xxx chips should have a max temp of 60C only, which I breached very cleanly with a 76C max temp (that I remember; I wasn't paying attention to it since I was working on a laptop beside it; it could very well have been at 80C at the specific time it died).

First off What temp program were you using to get temps on an AMD?
Second my understanding was AMD has a very temperamental diode?
Im betting it was more cooking at 90C....

I'm just not sure if you are telling me "well, that's unlucky, most forumers torture it for >3 hours with greater volts and greater temps, so it shouldn't have died" or if you are telling me "well, at least you know it's your fault, because you really shouldn't have exceeded 60C"

No... you should of paid more attention to your stress test instead of playing on your laptop... this is your fault and error, and could of been avoided...
If it makes u feel any better... ive done the same thing more then once... yes i still haven't fully learned from that mistake...

Either statement is perfectly applicable, though. I'm just not sure which one to use when I tell you "yep, you're right "

Please read my bolds as reply... now u see why people usually dont leave their machines unattended for at least the first 30 min during a stress test.... ITS NOT OCD!!!

I would also RMA the CPU cause there is nothing wrong with doing so. At least i don't see.

he cooked his cpu... so he should RMA it?

I'd still send the chip back to AMD to find out the reason, "why my experience has been so drastically different to the others". AMD, out of good will, might send a replacement. No fraud involved whatsoever

If this was Intel... possibly... and no im not kidding...
Intel LOVES Dead CPU's as long as no physical damage can be seen on the cpu.
They want to see what the weakest link is, and improve it each revision.

 

[jvroig]

No... you should of paid more attention to your stress test instead of playing on your laptop... this is your fault and error, and could of been avoided...

Aigo, please, read more of the posts, particularly the "Long Story" section in the OP

I've been through this before with member PPB, and it's all laid out in the "Long Story" section in the OP so I won't elaborate too much again here, but to recap:

Everything was done on purpose (except the death), there was no accident, it was a guess as "well, this is still safe, should be safe" based on pertinent data I accumulated in the absence of AMD's official guidelines, to wit: (a quote from OP)

Using a Thermaltake Frio OCK at max fan with the Tuniq, I'd be getting max load temps (Prime95 Large FFT) of ~57-59, HPC Mode on. I tried that same profile on the TG-2, and temps soared to over 70C, settling at around 74-76C.

Here's what I was thinking. AMD doesn't publish max temps AND vcore ranges for this part. If I go by the FX-8150 (direct ancestor), max temp should only be 61C (but at what range of operating voltages? AMD doesn't publish). But Trinity mobile SKU's (like the A4-4300m I benched in VC&G) sporting the same Piledriver cores have a much higher limit (sites like CPUBoss and CPU World specify 100C, despite a vcore range of up to 1.35V). So, the same old Piledriver cores can take 1.35V and 100C? Well, if so, I'm still relatively ok (1.396V, 76C).

So, I just let it run continuously as I continued working on my laptop.

Again, this was no accident, it's not like I would have stopped the test IF ONLY I saw it was at 70 and higher. In fact, I did see the temps, and then made the conscious decision to let the test continue, based on the rationale quoted above, which came from the OP


EDIT FOR CLARITY:
I think the simplest way to put it, to avoid any confusion on whether I bothered to monitor it or not, or if it was an accident or not, is this:
- I saw the temps as the test began and until they settled.
- I recognized they were high-ish
- I had a choice - stop the test, or let it continue. Where do I base this decision?
- AMD doesn't publish any relevant specs for this particular SKU regarding operating voltages and temps, so no help here.
- Having no official reference, I can either use data from direct ancestor (8150, temps only, no voltage spec), or from Trinity mobile SKU's that have
supposedly the exact same Piledriver cores as this SKU I am torturing, and these have both the voltage AND temp specced.
- I decided to go with the specs of an SKU that also has a Piledriver core (since this one specs both the voltage AND the temp)
- Therefore, despite the temps settling at 74-76C, I let it continue anyway since this is still acceptable under Trinity mobile specs.
- Amazingly, unlike Phenom II X4 and X6 chips that would error out or BSOD the moment you cross 62C, this one just kept on doing ok for over 2 hours. This furthered my belief/guess that basing it on Trinity specs is the right course of action, because based on Phenom II experience, if I really breached the max operating temps already, this would have BSOD'd already.
- Not amazing, it died shortly after. It was no accident, though. It was the result of a conscious decision based on the limited data available at the time.


Your point about the thermal diode is fair though. I use HWmonitor and Core Temp in Windows, and it seems to align with BIOS temps, and also with the lm-sensors readout in Linux. So whatever they are reading, they are all reading it correctly. But, as Phynaz posted earlier, we don't really know what the readouts mean, only that the readouts are uniform and comparable to each other. For all we know, they could in fact be cooking at 90C as you suggest. We've got no official specs from AMD, so we really can't say anything definitive here.

 

[aigomorla]

Again, this was no accident, it's not like I would have stopped the test IF ONLY I saw it was at 70 and higher. In fact, I did see the temps, and then made the conscious decision to let the test continue, based on the rationale quoted above, which came from the OP

and knowing its you, im fairly sure you did it on purpose..
I was more stating it on the people who told you to RMA it... lol...
I apologize for the confusion.

its like when i went poping E8500's...
I litterally was given a bin to kill off, and you should of seen people as i showed screen shots of 1.5vcore on these puppies...

The massive massive hate i was thrown at, because they were E8500's all doing 4.5ghz+.

I think i ended up giving IDC one of them to play/test/kill...

Those were fun days... very very fun days....

 

[jvroig]

its like when i went poping E8500's...
I litterally was given a bin to kill off, and you should of seen people as i showed screen shots of 1.5vcore on these puppies...

You know, that's actually a good idea... I could get a handful of cheap Vishera quads, purposely try to kill them, and then see what temps and voltage they die at. It seems this mobo can take quite a beating anyway (over 2 years old now, so far has survived a highly overclocked 1090T and outlived an 8350 in HPC mode), so might as well kill off a few cheap CPU's in an attempt to produce more data points to accompany this one.

EDIT: I just checked my favorite retailer's pricelist. Holy crap, it makes almost no sense to buy a quad-core Vishera (FX 4300) when a hex-core (FX 6300) is only ~$5 more expensive, and both are 95W TDP chips. Whether for killing or actual usage, the 6300 looks like a no-brainer buy.

 

[Rvenger]

This.

There is no moral or ethical justification for warranty fraud.

There is, however, plenty of unethical and amoral justification for warranty fraud to be found.

Its not theft when you tell them the truth. I RMA’d cpus and telling tech support I overclocked the CPU. They still replaced it. AMD has incredible customer service and trust me, they want your business and support more than ever now.

 

[Torn Mind]

Having a warranty for chips that were "popped" due to some overclocking or overvolting creates a moral hazard situation in which people will do more reckless things with their chips because they won't feel most of the burn(downtime is still a cost) financially. And of course, there could be "haters" who'd buy chips just to kill them so they can drain the "enemy's" pockets.

 

[bononos]

I'm not sure which of you is more accurate, but the real problem isn't whether it is 61C or 70C. The real issue (in my mind, anyway) is that this info is not published by AMD. The FX-8350 neither has the range of operating voltages nor the max operating temperature published (unlike, say, the Phenom II parts). You guys could argue 'til the cows come home, but without an official published reference from AMD it wouldn't really matter, would it?
......

Thats bizzare isn't it. First AMD doesn't report temps properly. And on top of that they don't publish max operating temps for many of their models (or is it just for the piledriver?). Its almost like its some big secret.

 

[monkeydelmagico]

Thats bizzare isn't it. First AMD doesn't report temps properly. And on top of that they don't publish max operating temps for many of their models (or is it just for the piledriver?). Its almost like its some big secret.

If the variability is high I can understand why they may not have published. Tolerances all over the map = ridicule.


Personally I think OP got a bum chip. It happens. I appreciate the fact that he decided to post up his results despite the inevitable armchair quarterbacking.

 

[AnitaPeterson]

Actually....

As an AMD FX-6300 user myself, I have a question... I read the entire thread and i still can't figure out something here:

How DO you guys measure the CPU temperature, anyway? Through BIOS? I have CoreTemp, but its readings are less than reliable (usually showing below ambient temperature). Some people (here and elsewhere) say that you should add 10 degrees Celsius to any CoreTemp value shown in windows - but that still seems awfully arbitrary to me.

 

[tweakboy]

Just RMA it, and get something temporary for now. gl

 

[tolis626]

Actually....

As an AMD FX-6300 user myself, I have a question... I read the entire thread and i still can't figure out something here:

How DO you guys measure the CPU temperature, anyway? Through BIOS? I have CoreTemp, but its readings are less than reliable (usually showing below ambient temperature). Some people (here and elsewhere) say that you should add 10 degrees Celsius to any CoreTemp value shown in windows - but that still seems awfully arbitrary to me.

AFAIK the temps reported by CoreTemp and HWinfo are in line with what the bios reports.I guess they're reading from the same sensors.The fact that the reported temperatures can be lower than ambient is kinda weird,but it doesn't necessarily mean they are wrong.What kind of cooler are you using?Are you overclocking?Perhaps you're just running cool and the ambient temperatures are kinda high.I've seen it happen during summer months...

 

[ClearlyWeary]

I would try and RMA it -- yeah the fine print ofc reads about how oc'ing and out of spec usage blah blah blah. But this chip is MARKETED to hit like 5 GHz (http://sites.amd.com/us/promo/products/Pages/amdfx.aspx) check out the webpage. It drones on and on about oc'ing it: No premium to pay. No code to input. Just unrestrained, overclockable power right away. In fact *ALL THREE* bullet points note OC, OC, OC, unlocked, unlocked unlocked.

So if AMD advertises the chip in this fashion, and also has tech issues with the chip shutting down at certain limits, AND often has poor temp sensors that make it difficult to determine what temp you're running at AND doesn't note max voltage / temps and what not... then I would RMA away. You are using the chip in the manner designed and the bs fineprint is just that. Chips should be advanced enuff in this day and age to throttle down to prevent all but the most insane situations. Plus if this is one of the only times someone had an FX chip die it sounds like it could have been a bunk chip.

I've got a hankering that perhaps the current FX 8350 chips are a little shittier than the ones they HAD been making as the higher binned ones are used as 9xxx's. Same thing happened when intel released the 2700k -- scooped up all the nice 2600ks and called them the 2700k. So if the OC'ability of your chip went down the drain as they wanted to squeeze more money out of consumers then god forbid you get a replacement.

I'm inclined to agree with this train of thought. Now if you were trying to RMA it because it couldn't overclock to whatever speed that would be one thing but I honestly think for the chip to fail at the meager speeds you were using something had to be actually wrong with the chip itself.. Hence why a RMA should be valid.

My experience with my 8350 is that you can't run it to hot. My computer simply shuts off if it hits that wall. It doesn't throttle back, simply turns off.

 

[SlowSpyder]

Hey Steve! Yeah, haha, that would be a bit more stressful, I imagine.

Hey, you still owe me that review of your v2 MSI 990FXA-GD80 when you get that rig up and running!

Absolutely! I started running a few benches on my Thuban to get some baseline numbers. I have a little more hardware on the way, then its time to get the new rig up and running. Soon I hope!

 

[Phynaz]

The fact that the reported temperatures can be lower than ambient is kinda weird,but it doesn't necessarily mean they are wrong.

Of course it does. How do you air cool something below ambient?