Death of an FX-8350

[jvroig]

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.

Not for Piledriver in general. Just for Vishera (83xx line). Piledriver in Trinity, whether mobile or desktop, is specced properly. Off the top of my head, the 5800K, for example, has a max operating temp of 74C, while voltage ranges were specified as 0.825-1.475V. All nice and tidy as expected.

I don't want to rag on AMD (it's not like Intel needs help from me to further beat down AMD or anything ), so what I'm about to say is purely my educated guess on the matter based on all facts available to me (which is simply what AMD elected to publish). But seeing how the Phenom II line was properly specced (and by "properly specced" I mean the relevant electrical and thermal specs were published), and the mobile APU's and the desktop APU's were/are also properly specced, and the only line that actually didn't get to be specced properly was the big core line (Bulldozer 81xx and Vishera 83xx), I think they had to 'relax' their specs for the big core SKU's in order to be able to actually "bin" a significant enough quantity of SKU's for their target top-bin products. In other words, had they opted to follow a strict spec as they did in the Phenom II lines and as they continue to do in the APU lines from Llano to Trinity to Richland, they probably saw they'd never be able to produce as much of their target top-end SKU (the 8-cores) because most would be running out of spec and would have to be binned lower. IDC's testimony that even AMD support and support people from mobo makers (not sure if he means ASUS or MSI, or both) couldn't produce a spec or official thermal and electrical guideline from AMD lends a little bit of credence to this. To avoid producing clearly out of spec SKU's, they trashed the spec so there's nothing to breach. (Of course, internally, there still has to be a spec, it just wasn't published for public consumption and didn't make it to their client-facing support teams.)

I'll say this now so I don't have to take too much flak from those who would take offense at me suggesting that the 83xx line doesn't have a published spec because most of them would otherwise be clearly out of spec: this is merely the conclusion I formed in my mind as the answer to the question posed by member bononos. I do not mean this to be gospel, and if you have any other alternative theories, they are just as valid as mine, since the validity of these theories are effectively zero unless some of us manage to actually get hold of internal operational documents from AMD. For example, you could propose an alternative hypothesis that states AMD merely forgot to update the website portion where the 8xxx lines reside, because they only have 1 intern left for website duty, and he had to prioritize the more mainstream APU products. That's fine, I'm not going to waste my time and yours by arguing the validity of such an alternative hypothesis. Only AMD people can really answer this question, so your guess is as good as mine as to why only the non-APU big core line since Bulldozer was left not properly specced.

 

[jvroig]

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.

Yeah, the idle temps are definitely off because they are below ambient, but what I noticed is that there are only 3 steps from what is clearly "wrong" to what is expected. I can't check now since this guy kicked the bucket, but if I remember correctly, the below ambient temps were always 12C, then 15C, then 22C, then it will be ~30C (I consider 30C "correct" because that's not below ambient, a couple of degrees higher already). It was always like that. And it didn't matter what I used - Core Temp, HWMonitor, or the lm-sensors in Linux - they show comparable temps, and they all have the same hops from idle to loaded - starts at 12C, then 15C, then 22C, then 30+ and from there it increments by 1, no more of the fixed arbitrary hops.

So at idle, I dismiss the actual figure (12C) as imaginary and consider it 28-29C (normally 1-2C above ambient). The loaded temps seem reasonable and comparable to the 1090T and 965, so I take the loaded temps at face value.

That's simply how I made sense of the numbers, because for me, in this particular setup, only the really idle, unloaded temps looked off, while at load the temp monitors even in linux emit temp readings comparable to the Deneb/Thuban with considerably less power consumption (c/o kill-a-watt).

If yours produce crazy low readings even at load... I don't know, sorry.

 

[USER8000]

I suppose you could always use an infrared thermometer to get a reading??

Shame they are so expensive!!

 

[AnitaPeterson]

Yeah, the idle temps are definitely off because they are below ambient, but what I noticed is that there are only 3 steps from what is clearly "wrong" to what is expected. I can't check now since this guy kicked the bucket, but if I remember correctly, the below ambient temps were always 12C, then 15C, then 22C, then it will be ~30C (I consider 30C "correct" because that's not below ambient, a couple of degrees higher already). It was always like that. And it didn't matter what I used - Core Temp, HWMonitor, or the lm-sensors in Linux - they show comparable temps, and they all have the same hops from idle to loaded - starts at 12C, then 15C, then 22C, then 30+ and from there it increments by 1, no more of the fixed arbitrary hops.

So at idle, I dismiss the actual figure (12C) as imaginary and consider it 28-29C (normally 1-2C above ambient). The loaded temps seem reasonable and comparable to the 1090T and 965, so I take the loaded temps at face value.

That's simply how I made sense of the numbers, because for me, in this particular setup, only the really idle, unloaded temps looked off, while at load the temp monitors even in linux emit temp readings comparable to the Deneb/Thuban with considerably less power consumption (c/o kill-a-watt).

If yours produce crazy low readings even at load... I don't know, sorry.

My "idle" temperatures (as reported by CoreTemp) hover around 10C, even if the ambient's at 22-23... Under load, I see the temperature shoot into the 30s and occasionally 40s... I have a Corsair H60 water cooler, with the fan set on low (about 1,000 rpm)...

 

[Idontcare]

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

Yeah not happening with ambient air or water-cooling if the water is cooled by the ambient air.

My 8350 claimed it was idling at 8C in my 22C room.

 

[tolis626]

Yeah not happening with ambient air or water-cooling if the water is cooled by the ambient air.

My 8350 claimed it was idling at 8C in my 22C room.

Well,when I wrote what I wrote about reported temps being lower than ambient,I didn't think we had to do with such big discrepancies.However,it isn't too rare to have a 1-3 degrees discrepancy.One usually monitors room (Ambient) temperature from a point somewhat far from the PC itself.One example of this happening is in a room which is directly exposed to sunlight.If the PC is in a space shielded by the sun,the CPU idle temps can actually be lower than the room's temperature.And this is more common (Altough stil rather rare) with water cooling that it is with air cooling.

 

[bononos]

Not for Piledriver in general. Just for Vishera (83xx line). Piledriver in Trinity, whether mobile or desktop, is specced properly. Off the top of my head, the 5800K, for example, has a max operating temp of 74C, while voltage ranges were specified as 0.825-1.475V. All nice and tidy as expected.

I don't want to rag on AMD (it's not like Intel needs help from me to further beat down AMD or anything ), so what I'm about to say is purely my educated guess on the matter based on all facts available to me (which is simply what AMD elected to publish). But seeing how the Phenom II line was properly specced (and by "properly specced" I mean the relevant electrical and thermal specs were published), and the mobile APU's and the desktop APU's were/are also properly specced, and the only line that actually didn't get to be specced properly was the big core line (Bulldozer 81xx and Vishera 83xx), I think they had to 'relax' their specs for the big core SKU's in order to be able to actually "bin" a significant enough quantity of SKU's for their target top-bin products.
.........

I (and others before me) figured out just as much and I was only being slyly circumspect

I thought AMD did put out max temps for the 8 core bulldozer 8150 (61C) but omitted the info for piledrivers:
http://www.cpu-world.com/CPUs/Bulldozer/TYPE-FX-Series.html
http://products.amd.com/en-us/DesktopCPUResult.aspx

 

[bononos]

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.

No you shouldn't add 10C to the coretemp values since the the reported temps are only useful when the cpu is loaded. The idle temps are just off.

 

[Idontcare]

I suppose you could always use an infrared thermometer to get a reading??

Shame they are so expensive!!

You mean something like this?

For my FX-8350 (which was lapped, as was the stock HSF, so great thermal contact) the HSF temperatures at load were around 10C lower than the reported CPU temperatures when using CoreTemp.

Well,when I wrote what I wrote about reported temps being lower than ambient,I didn't think we had to do with such big discrepancies.However,it isn't too rare to have a 1-3 degrees discrepancy.One usually monitors room (Ambient) temperature from a point somewhat far from the PC itself.One example of this happening is in a room which is directly exposed to sunlight.If the PC is in a space shielded by the sun,the CPU idle temps can actually be lower than the room's temperature.And this is more common (Altough stil rather rare) with water cooling that it is with air cooling.

I measure ambient air temperature just upstream of the air-intake for the cooler in question, here is an example: (note the "remote temperature sensor")

The problem is the error in the on-die temperature sensor itself. Not that this isn't common, RealTemp went to great pains years ago to develop a calibration procedure for estimating the error and correcting for it when using RealTemp (but it only works with Intel CPUs )

 

[jvroig]

I thought AMD did put out max temps for the 8 core bulldozer 8150 (61C) but omitted the info for piledrivers:

Indeed. The fact that BD had max temps while Vishera had nothing was something I already touched upon in the OP, and the existence of the BD max temps did factor into my rationale for choosing to continue with the test. (It was a data point I considered while making the decision, but ultimately discarded in favor of using a more complete data point from a Piledriver-based SKU)

But take note of how I defined "properly specced" to mean both the max operating temps AND operating voltages, not just one or the other. The max safe temperature isn't a constant value that applies to EVERY scenario. It actually depends on what max voltage a particular max safe temp was specced at. It's kind of like how power supplies, when trying to look better than they actually are, boast "80+" efficiency, but they "forgot" to tell you that the measurement was taken while the unit was at a cold 20C, instead of while it was at 40-50C, because at those high (but common @ load) temps, the efficiency tanks to 73.14159265%. So too with the max temps spec. Particularly if you are overclocking and looking for a reasonable guideline to what is 'safe' (i.e., what you can more or less reliably assume the chip can take as per its designed thermal and electrical characteristics) you really need to know both to make a more accurate assumption. They've always specced both as expected, (it's impossible to not have that data internally, they must have ran through a truckload of PVT (Process-Voltage-Temperature) corners during the design stage, so it's merely a choice to publish or not), so voltage ranges disappearing from BD was weird, and then voltage ranges AND max temps disappearing from Vishera is even weirder, leading me to conclude as I did. I could be wrong, I could be right, doesn't really matter as it won't change the facts.

 

[AnitaPeterson]

Wouldn't it be great if someone from AMD would demonstrate their commitment to us enthusiasts, by coming to these forums and providing some clear advice and information concerning their products?

Oh, wait... http://forums.anandtech.com/showthread.php?t=2339379

 

[jvroig]

Wouldn't it be great if someone from AMD would demonstrate their commitment to us enthusiasts, by coming to these forums and providing some clear advice and information concerning their products?

Oh, wait... http://forums.anandtech.com/showthread.php?t=2339379

Good idea!

He said he wasn't much of a hardware expert these days, so I'd suggest that when you post in his thread to make a query about your problems with your FX 6300, use small words and short phrases. I'm kidding, of course, but it would probably help him know what to ask the people around him if you explained clearly your difficulties and the scenario in general. Go for it!

 

[aigomorla]

Yeah not happening with ambient air or water-cooling if the water is cooled by the ambient air.

U mean its not gonna happen unless u put energy back into the system.

These guys would fall under your not happening definition partially..

Expanding on IDC's Test bench.
When i was testing i had 2 AC's.. 1 window unit.. and 1 portable to ensure the ambients in my room were within 1-2C though out the entire test.

I used Watercooling because it was easier to control set ambients though a radiator then it is though an air heat sink.

I also think i had about 10 temp problems scattered everywhere though out the bench logged via crystalfontz software.

Ugh... i dont see myself getting back in that type of testing anymore...