How important is that 62* max temp on Phenom IIs?

[maniac5999]

The backstory: I just got a 940, and it's stable at 3.4ghz with 1.43v. The issue that I'm having is that when I'm stress testing it using OCCT it flirts with 60* even though the top of the heat sink (Sunbeam CCF) is barely warm to the touch. This is in a 68*F (20*C) room in the middle of winter, and I don't normally air condition the room in the summer, even though it can hit 95*F (35*C)

Because of this, I want to know just how bad it is to exceed 62*. if it hits 65* or 70* occasionally under load in the summer would that be horrible, or no big deal? I'm asking this because a 62* max temp seems to be VERY low. My old Kuma was listed as fine to 73*, the L335 in my netbook is listed as fine to 95* (they're both 65nm, but even 73* to 62* is a big drop).

Also, I'm wondering if my board is reporting the temperatures as being higher than they really are. I've re-mounted the heat sink a couple times, using varying amounts of thermal paste and it doesn't seem to make a difference, it idles around 40* case open or case closed in a 20* room, when according to Frostytech it should only have a 21* delta to ambient under 150w load. http://www.frostytech.com/articleview.cfm?articleid=2271&page=5 Interestingly with my old Kuma, the temperatures seemed to be reported high as well, but I ran into the voltage wall long before the heat wall, so I didn't really pay attention.

What do people think? should I act as if my temperatures are 10* lower than they are? should I just shrug it off? Should i reduce my OC to keep it cooler?

Edit: The plot gets even thicker. Even conservative old AMD says the X2s can go to 70*, and they're the same d#%$ silicon as the X4 http://products.amd.com/en-us/DesktopCPUSideBySide.aspx?id=621&id=560

 

[LoneNinja]

I would have sworn the max temp on the Phenom II X4 processors upon launch was 70C, thought it was the C3 revision that lowered both max safe voltage and max temp.

I use to run my 940 around 65C all the time under full load, but now I've got a hyper 212+ and 3.6Ghz @ 1.45V I never see above 55C.

 

[fffblackmage]

The max temps for Phenom II processors are weird. The max temp for the 550 is 70C while the 955 is 62C. It just seems weird that the quad cores don't have a max temp as high as the dual core, which is the same die anyways.

Just wondering, how much TIM are you applying? Apparently... I was hitting 70C on my unlocked 550BE because I applied too much TIM....

 

[veri745]

As long as it's stable, I wouldn't be too concerned if it surpassed 62C by a little. I wouldn't let it go above 70C though.

It does seem weird that the TCaseMax varies so widely on the PhII's

 

[Idontcare]

The max temps for Phenom II processors are weird. The max temp for the 550 is 70C while the 955 is 62C. It just seems weird that the quad cores don't have a max temp as high as the dual core, which is the same die anyways.

Remember that temperature specs are created for the purpose of ensuring the IC fully functions for the intended lifetime of the device.

Temperature reduces functional lifetime.

Doubling the number of transistors in the IC will reduce the functional lifetime (statistically speaking) by a factor of 2x if all things are held constant.

AMD elects to not hold all things constant by way of reducing the max operating temp of the quad-core chips.

So they double the number of things that can fail but then they reduce the max spec temp so the probability of component failure is reduced.

From AMD's perspective its all about limiting their liability. In-field fails are not good, both reputation-wise as well as warranty-replacement cost-wise.

1X xtor count @70C yields ~ the same lifetime as 2X xtor count @62C for the specific parts AMD sold.

 

[maniac5999]

Just wondering, how much TIM are you applying? Apparently... I was hitting 70C on my unlocked 550BE because I applied too much TIM....

I'm currently using as little as I think I can get away with, I put four of the thinnest lines that I could on the 4 heat pipes, and they're only about 1/2" long. I'll try to re-mount with even less, but I'm suprised that the temp is getting that high, The CCF is one giant hunk of metal, and it barely even seems to get warm. Frostytech said it only yielded a 21* rise over ambient on the lowest setting under a 150w load.

 

[Rhoxed]

can't bring the voltage under 1.43? whats NB volts at?

regardless that seems high, my 940BE barely cracks 60C @ 3.9ghz 1.55V with a xigmatek thors hammer heat sink

what case do you have? is there good airflow? HS fan pulling from VID card? Top Mount PSU?

 

[eternalone]

My Phenom 2 925 X4 95w C2 Has a TDP max of @70C

 

[Smoove910]

I'm currently using as little as I think I can get away with, I put four of the thinnest lines that I could on the 4 heat pipes, and they're only about 1/2" long. I'll try to re-mount with even less, but I'm suprised that the temp is getting that high, The CCF is one giant hunk of metal, and it barely even seems to get warm. Frostytech said it only yielded a 21* rise over ambient on the lowest setting under a 150w load.

So you applied 4 separate 'lines' of TIM on your heatsink? I'm willing to bet that's part of the problem.

Put a single glob in the middle of your heatsink equal to the size of a grain of rice. Install heatsink and be done with it.

 

[maniac5999]

So you applied 4 separate 'lines' of TIM on your heatsink? I'm willing to bet that's part of the problem.

Put a single glob in the middle of your heatsink equal to the size of a grain of rice. Install heatsink and be done with it.

I apologize, but you are incorrect there: http://benchmarkreviews.com/index.p...ask=view&id=170&Itemid=1&limit=1&limitstart=5
Coolers with heat pipes touching the processor lid require a different method because of the grooves between the flattened pipe and the metal backplate. A) you need more TIM to fill in those grooves, and B) you need to apply it differently because they affect the spread pattern.

I pulled off the heat sink and checked, instead of too much, i may have put slightly too little TIM on it initially, it was only covering about 85-90% of the contact surface, re-applying and re-mounting seems to have dropped my temps by 1-2 degrees. I do suspect that my board is reading the temperatures a little high, however. Immediately after booting into windows in a 20*C room it was reporting temperatures in the upper thirties. After shutting it down and letting it sit for half an hour with everything off and the case open, I booted into BIOS, and it was reporting 32*C, and it sat there at that temperature for about two minutes, before starting to climb by a degree every 30 seconds back into the upper thirties.

Would it be safe to say that a cold computer booting straight into BIOS should report ambient temperatures, and that if it doesn't that the temp. sensors would then be off by that many degrees?

 

[Rhoxed]

Would it be safe to say that a cold computer booting straight into BIOS should report ambient temperatures, and that if it doesn't that the temp. sensors would then be off by that many degrees?

no, bios does create a load, and you will sometimes get better idle in windows than bios.

 

[Jovec]

AFAIK AMD only provides one on-die sensor for reading temps, which is why all cores read the same. Any other temperature sensor reading is from the mobo sensor located below the CPU socket (Assuming one is reading the correct TMPIN sensor that varies by mobo make). Depending on which temperature you are reading, you can easily get a 10C difference or more. AMD has never publicly stated which temperature reading needs to stay under 62 (typically, higher for some CPUs as mentioned), and I also think a lot of confusion is due individuals using one temperature or the other, making straight comparisons harder.

I'd appreciate any definitive info regarding AMD temps, or maybe JFAMD could get someone to clear it up.

 

[BD231]

Quite frankly I'm shocked you get it to operate at that temp, but what's more confusing is why you're even close to 60 at that voltage.

Something's not right, you shouldn't be seeing those kinds of temps.

 

[maniac5999]

Quite frankly I'm shocked you get it to operate at that temp, but what's more confusing is why you're even close to 60 at that voltage.

Something's not right, you shouldn't be seeing those kinds of temps.

I truly think that I'm getting an erroneous temperature reading. From what I understand, the on die temp. sensor has an output that the board has to interperet to get the actual temperature. (please tell me if I'm wrong here) To be blunt, I don't trust my board when it comes to temperatures. The Mobo CPU sensor has always read 20* higher than the "Core" sensor which is borderline impossible. (I've been quoting core temps btw)

no, bios does create a load, and you will sometimes get better idle in windows than bios.

I do understand that BIOS (and windows) will create a load, and eventually it'll stabilize at an active idle temperature, where the heat dissipated through the heatsink matches the heat generated. However, what I'd like to know is how quickly the sensor jumps under initial load. Theoretically, the second that I hit the power button, that sensor's very first reading should be the ambient temperature, since the CPU has had zero time to warm up. In a 20* room, the first reading should be 20*, anything different would be an inaccuracy in the sensor, or sensor interperetation software. (and such an error would likely be carried through the entire temp. range) For further corroberation, I put it into an OCCT loop last night when I went to bed, and it errored out around 2am (3.5ghz @1.4625v looks like i'm going to need 1.475v) and the computer went to sleep. Immediately after waking it OCCT reported a core temp of 32 degrees as well.

Is it possible for that sensor to jump 12* in the 30 seconds that it takes to get into BIOS? or the 15 seconds that it takes to wake from sleep? BIOS and the wake operation are quite CPU light from what I understand.

 

[deimos3428]

The max safe temp numbers for Athlon/Phenom II processors appear to be (somewhat loosely) related to the TDP:

125/140W = 62C.
45/65/95W = 70 to 73C
25W = 81C

The sensor in question is the one located under the IHS, not the individual cores. Under load, it's always lower than the temperature of the individual cores. The lower your TDP, the higher your reported temp at the IHS can be and still be safe.

So if a 25W processor reports an IHS temp of 71C, you know the core(s) are reasonably toasty but will be fine. If a 125W processor reports 71C at the IHS, that'd mean the cores are frying. It might not die right away, but definitely a bad thing.

 

[leptooners]

You shouldn't be getting those kinds of temps with any direct contact HSF. If you have a 95W CPU, 40c is the max you should see under full load imo, even if you overclock/volt.

 

[LoneNinja]

You shouldn't be getting those kinds of temps with any direct contact HSF. If you have a 95W CPU, 40c is the max you should see under full load imo, even if you overclock/volt.

940 is a 125W part, and his temps are only 5C higher than mine with similar clocks/voltage to me, and I'm using a hyper 212+.

 

[leptooners]

940 is a 125W part

Ah, my bad.

his temps are only 5C higher than mine with similar clocks/voltage to me, and I'm using a hyper 212+.

Well, I'm about to get a DC HSF for my new rig, and I'll be pissed if I can't get it down to 35c under load. I'm a stickler for superior air cooling. I made my current HSF myself from scratch and it keeps my old ass Pentium D at 27c MAX under full load. That's on air cooling, my friend. I've also made all of my mobo's heatsinks from scratch since the day I started building computers. On a side note, overclocking integrated graphics is usually a horrible idea, because it heats up your north bridge like you wouldn't fucking believe LOL. But, with proper cooling, it is possible (even long-term) without frying anything.

 

[busydude]

We need a sticky regarding how AMD measures the temp's. In my 6+ months of use, I really don't know what my exact CPU temp is, all I can do is take an uneducated guess.

 

[fffblackmage]

We need a sticky regarding how AMD measures the temp's. In my 6+ months of use, I really don't know what my exact CPU temp is, all I can do is take an uneducated guess.

How accurate is the temp readings in any CPU anyways? My impression has always been that temp readings are only approximations.

Also, are the readings temperature at the die or is it from a thermistor or something similar sitting under the cpu on the mobo?

 

[maniac5999]

Another question, is L3 timing tied to the HT Reference clock on the P2s? (if I set the clock to 240 instead of 200 will my L3 response time be 20% faster?) I think this was the case on the Old Phenom 1s, where you'd get higher performance by raising the clock instead of just increasing the CPU and NB multipliers, but I'm not sure if it still is, and if it isn't, then it would be easier for me to just mess with multipliers than to have to mess with all kinds of memory timings and whatnot.

 

[ClickClack]

I'll just add that with a Phenom II 940 there's no reason you need to up the voltage for 3.4 GHz clock. You should be able to have it at 3.4 on stock voltage (1.35 V). That's all I've read anywhere and that's what I usually have my PII 940 on with no issues. Just thought I'd add that in there. Above 3.4 you need to add some volts but yeah..

Just for reference, I have my Phenom 2 940 @ 3.5 GHz, 1.425 V and it's 37 degrees celsius right now. Granted, it's kind of colder than normal here in Texas and my room temp is at about 70 - 72 Fahrenheit. At full load I think I get about 50 - 52 degrees. That's with a Hyper 212+ cooler.

I just bought an i7 950 tonight though so after tomorrow it looks like I will have a new processor to tinker with!

 

[DeathRayLoveMachine]

Doubling the number of transistors in the IC will reduce the functional lifetime (statistically speaking) by a factor of 2x if all things are held constant.

It would be a factor of two if circuit failures were statistically independent of each other, but they're pretty clearly not. Device failures are usually modeled using a Weibull distribution, which assumes that failures are correlated with each other.

I can try to give you an example (the chance of rolling a 6 on a weighted die, after many consecutive trials), but it's boring.

(Manufacturing defects are a whole different story. Transistor count and certain kinds of manufacturing errors ARE directly, linearly proportional.)

 

[Idontcare]

It would be a factor of two if circuit failures were statistically independent of each other, but they're pretty clearly not. Device failures are usually modeled using a Weibull distribution, which assumes that failures are correlated with each other.

I can try to give you an example (the chance of rolling a 6 on a weighted die, after many consecutive trials), but it's boring.

(Manufacturing defects are a whole different story. Transistor count and certain kinds of manufacturing errors ARE directly, linearly proportional.)

Well of course their not 100% decoupled, the temp decrease in max spec temp wasn't 10C either. It's only an 8C drop for a reason.