AMD Phenom II TDP and underclokcing / undervolting

[Mr. Hilarious]

Hi Forum readers!

I have a HTPC (but not deticated - occasional gaming and some tinkering with stuff) setup (OS is Linux / Gentoo) and I'm going to upgrade the CPU (and add 2-4 GBs of RAM). Energy-Power -efficacy is important, because of NOISE. There's just one problem: I can't choose which CPU to upgrade to

Currently it is running on AMD X2 4850e. The temperatures are VERY good on this 45W processor with the aforementioned exhaust fans just next to the Nexus Ninja Mini, even with full load it never gets above 50°C (in the summer maybe just there - now that the sun isn't heating this apartment to the 32°C ambient temperature, I really need to work getting the cores to 46°C).

I have two processors currently in mind (that are available here):

AMD Phenom II X4 910e (2.6GHz, @65W TDP)
AMD Phenom II X6 1055T (2.8GHz, @95W TDP)

The price difference is negligible. I think that with the current cooling solution I'd have no problems cooling a 65W processor, but with a 95W TDP processor, I might run into problems. But, I read one can LOWER the multiplier on a non-BE Phenom II - so, if the cooling is insufficient, I could lower the multiplier on the X6 so that I'd run (when running at maximum speed) on 2.6GHz? If that is the case, I'd go for the extra two cores - or maybe even a 125W Phenom processor?!??

So, any experience with underclocking / undervolting Phenom II's? Is there a lot of TDP to gain, say, undervolting the aforementioned 1055T? Or a 125W one (either a X6 or a higher speed X4)? Do these new processors have any stability issues when undervolting / underclocking?

Cheers!

[Thatguy]

Hi Forum readers!

I have a HTPC (but not deticated - occasional gaming and some tinkering with stuff) setup (OS is Linux / Gentoo) and I'm going to upgrade the CPU (and add 2-4 GBs of RAM). Energy-Power -efficacy is important, because of NOISE. There's just one problem: I can't choose which CPU to upgrade to

Currently it is running on AMD X2 4850e. The temperatures are VERY good on this 45W processor with the aforementioned exhaust fans just next to the Nexus Ninja Mini, even with full load it never gets above 50°C (in the summer maybe just there - now that the sun isn't heating this apartment to the 32°C ambient temperature, I really need to work getting the cores to 46°C).

I have two processors currently in mind (that are available here):

AMD Phenom II X4 910e (2.6GHz, @65W TDP)
AMD Phenom II X6 1055T (2.8GHz, @95W TDP)

The price difference is negligible. I think that with the current cooling solution I'd have no problems cooling a 65W processor, but with a 95W TDP processor, I might run into problems. But, I read one can LOWER the multiplier on a non-BE Phenom II - so, if the cooling is insufficient, I could lower the multiplier on the X6 so that I'd run (when running at maximum speed) on 2.6GHz? If that is the case, I'd go for the extra two cores - or maybe even a 125W Phenom processor?!??

So, any experience with underclocking / undervolting Phenom II's? Is there a lot of TDP to gain, say, undervolting the aforementioned 1055T? Or a 125W one (either a X6 or a higher speed X4)? Do these new processors have any stability issues when undervolting / underclocking?

Cheers!

the stock 1055t cooler is actually pretty nice " if you get a retail box cpu kit" and its not to loud. I have one in my wifes computer in the living room. I put a v8 on my cpu and went the other way with clockspeed. hehe.

anyways, the fan is pwm so if you downclock, I got mine to 2.0ghz at 1.29volts" I only had to run the fan at like 15-20% total.

It was nearly silent.

[droidhacker]

The stock coolers are ok, if you don't mind them being CRAP. There is a high frequency noise that comes from some transistor that is part of the PWM crap... the fan is TRULY low quality junk.

[Thatguy]

The stock coolers are ok, if you don't mind them being CRAP. There is a high frequency noise that comes from some transistor that is part of the PWM crap... the fan is TRULY low quality junk.

its such a piece of crap that I was able to overclock my 1055t from a stock setting of 2.8ghz to 3.9ghz on air. Wow what piece of junk.

[Mr. Hilarious]

the stock 1055t cooler is actually pretty nice " if you get a retail box cpu kit"

The stock coolers are ok, if you don't mind them being CRAP.

its such a piece of crap that I was able to overclock my 1055t from a stock setting of 2.8ghz to 3.9ghz on air.

Heh, I started a small flame war

But seriously, thanks for your input, though you never answered my question: which one is better, the 65TWP processor, or the 95W hexacore underclocked to an equivalent clock speed, in terms of current consumption? Another way to put my question: are there measurements, that approximately shows the TDP in relation to the clock speed? Or TDP vs. power steps graph when using Cool'N' Quiet? Is the curve linear? Or, does the slope get higher at higher clock speed? Remember, more current = more heat = more noise.

After posting the thread, I figured, that of course you can always lower the clock speed... that's what Cool'N'Quiet is all about (and also lowering the voltage, IIRC, although quite conservatively). After booting to Linux, probably just editing the highest frequency step, or limiting the PM governor to the lower ones, should be enough to acquire what I want. IIRC, in Linux editing the power steps is possible, tought I'm not sure.

I believe that the 95W underclocked by 200MhZ (that's their clock speed difference) won't do much difference, so I'm leaning towards the quad core at the moment.

Maybe I'm at the wrong forum. I might get more attention at spcr forums for this question

About stock coolers: in my experience the stock coolers are the minimum (or near minimum) you need to safely use a processor. Never QUIET by my standard (and, quietness is a subjective thing). While it might be that AMD has put some effort on their stock cooler, I doubt that the stock heatsink is better than the Scythe Ninja I currently have (I'm not going to put a dedicated CPU fan in there). I conclude that AMD might have the same heat sink and cooler for their 125W and 95W lines of AM2/3 socket processors (perhaps for simpler mass producing). That might explain why you get good results with overclocking on the stock cooler.

Currently I have two Nexus test winners (don't remember the model) sucking (in terms of pressure, of course, not sucking as in being CRAP ) right next to the Ninja Mini. Look at here to get a picture of my setup. (Same case as in the review ) I think the current fans I have start to sound annoying if another one is running at =>1200RPM (i.e. I hear them without needing to concentrate on them). But that is the case only when I'm doing something CPU intensive, not when listening to classicfm.nl


Cheers!

[Thatguy]

Heh, I started a small flame war

But seriously, thanks for your input, though you never answered my question: which one is better, the 65TWP processor, or the 95W hexacore underclocked to an equivalent clock speed, in terms of current consumption? Another way to put my question: are there measurements, that approximately shows the TDP in relation to the clock speed? Or TDP vs. power steps graph when using Cool'N' Quiet? Is the curve linear? Or, does the slope get higher at higher clock speed? Remember, more current = more heat = more noise.

After posting the thread, I figured, that of course you can always lower the clock speed... that's what Cool'N'Quiet is all about (and also lowering the voltage, IIRC, although quite conservatively). After booting to Linux, probably just editing the highest frequency step, or limiting the PM governor to the lower ones, should be enough to acquire what I want. IIRC, in Linux editing the power steps is possible, tought I'm not sure.

I believe that the 95W underclocked by 200MhZ (that's their clock speed difference) won't do much difference, so I'm leaning towards the quad core at the moment.

Maybe I'm at the wrong forum. I might get more attention at spcr forums for this question

About stock coolers: in my experience the stock coolers are the minimum (or near minimum) you need to safely use a processor. Never QUIET by my standard (and, quietness is a subjective thing). While it might be that AMD has put some effort on their stock cooler, I doubt that the stock heatsink is better than the Scythe Ninja I currently have (I'm not going to put a dedicated CPU fan in there). I conclude that AMD might have the same heat sink and cooler for their 125W and 95W lines of AM2/3 socket processors (perhaps for simpler mass producing). That might explain why you get good results with overclocking on the stock cooler.

Currently I have two Nexus test winners (don't remember the model) sucking (in terms of pressure, of course, not sucking as in being CRAP ) right next to the Ninja Mini. Look at here to get a picture of my setup. (Same case as in the review ) I think the current fans I have start to sound annoying if another one is running at =>1200RPM (i.e. I hear them without needing to concentrate on them). But that is the case only when I'm doing something CPU intensive, not when listening to classicfm.nl


Cheers!

well, tdp is also based heavily on load. More load higher tdp, so its not just voltage and clock speed.

You can downclock a thuban or a phenomII fiarly easily. you could also disable cores " with some bios's" and shed some tdp there as well.

You could " in thoery" passively cool a 95w CPU but it would take a Large heatsink and lots of case ventilation.

The stock AMD3 thuban cooler is fiarly quiet, for what it is. I would hardly call it silent.

[jannis]

I have an Phenom II X6 1055T 125W here. I wanted the 95W-version but it didn't get available (and still isn't as far as I can see).
Using a Noctua NH-C12P SE14 (very large and silent fan) rotating at 408 RPM, the CPU idles at about 30°C. Under full load, the temp goes up to 58°C (the fan is at 900 RPM then).
And that is with default freqs and voltages. Using my MSI 890FXA-GD70 I could well adjust all the voltages but I didn't yet find the time or the need to do so.
Talking about undervolting: Support for "TurboBoost"-enabled processors is on the way, other CPUs are supported. (see: http://linux-phc.org/forum/viewtopic.php?f=13&t=38). I'm in discussion with the phc-k8 developer since he doesn't own a TurboCore-CPU himself.
And of course you can disable all but one core anytime using "/sys/devices/system/cpu/cpuX/online" while the system is running.

[Mr. Hilarious]

well, tdp is also based heavily on load. More load higher tdp, so its not just voltage and clock speed.

Yes, it seems so. I'm leaning towards the energy efficient quad-core.

I found a couple of interesting bits of information:

1) The equation P = CV^2f from the Wikipedia SpeedStep article. I think it roughly applies to all processors. I did some calculating for fun, and for the 910e, I got a capacitance of 16 (not sure of the unit), and for the 1055T (95W version) 16,7 (and, for the 125W version, 20,5 btw). I used the specified maximum frequency and operating voltage.

I even went to the lenght of calculating a theoretical TDP when underclocking for these and some other processors via the equation above (this is of course not correct, but just an excersise and a rough estimate to give some direction. You'll get a linear correlation).

2) While doing this I found some voltage ramps from tomshardware - THIS was exactly what I was lookin for when posting this thread!). But, no tests for 910e (or similar) and no 1055T 95W TDP version there .

But, then I calculated the C for some processors found on the article linked above, and roughly estimated the current when underclocking - and of course, compared the results . My calculations are more optimistic than the real measurements (they don't take the voltage decrease into account), and even more so on lower frequencies. But at >2,0GHz the difference is only in the order of few watts. According to my calculations, it seems that the energy efficient 910e would be much cooler than the 95W 1055T at the same frequency (for example, 88,2W vs. 65W @ 2,6GHz).

I also calculated the C value for all Phenom II X6 processors. With this method, only one has a C value lower than 16 - this is the Phenom II X6 1065T 95W version (at 15,06), but it can't be found on sale anywhere. I guess it COULD run cooler at, say 2,6GHz than the 910e, but also might not, since the correlation is not linear - i.e. C is not constant but only roughly (another matter is, why it is not constant).

Of course, this is just speculation and I might be wrong. Real-world tests would be needed. And I don't think there are a lot of people around to do the power measurements required for this

And, of course every chip is an individual, even in this aspect . If you're lucky, you might get one that runs at a lower voltage and lower current (even at load).

Any comments?

Cheers!

[Thatguy]

alot of p3 chips that were 1.4v were passive air cooled, just saying. I have no idea whay your obessing so much over the tdp. I'd bet the 6 cores could actually be as energy efficient as the 4 cores given the same number of active cores.

Also you could just water cool your rig and that would make it nearly instantly siletn.

Just sayin.

Yes, it seems so. I'm leaning towards the energy efficient quad-core.

I found a couple of interesting bits of information:

1) The equation P = CV^2f from the Wikipedia SpeedStep article. I think it roughly applies to all processors. I did some calculating for fun, and for the 910e, I got a capacitance of 16 (not sure of the unit), and for the 1055T (95W version) 16,7 (and, for the 125W version, 20,5 btw). I used the specified maximum frequency and operating voltage.

I even went to the lenght of calculating a theoretical TDP when underclocking for these and some other processors via the equation above (this is of course not correct, but just an excersise and a rough estimate to give some direction. You'll get a linear correlation).

2) While doing this I found some voltage ramps from tomshardware - THIS was exactly what I was lookin for when posting this thread!). But, no tests for 910e (or similar) and no 1055T 95W TDP version there .

But, then I calculated the C for some processors found on the article linked above, and roughly estimated the current when underclocking - and of course, compared the results . My calculations are more optimistic than the real measurements (they don't take the voltage decrease into account), and even more so on lower frequencies. But at >2,0GHz the difference is only in the order of few watts. According to my calculations, it seems that the energy efficient 910e would be much cooler than the 95W 1055T at the same frequency (for example, 88,2W vs. 65W @ 2,6GHz).

I also calculated the C value for all Phenom II X6 processors. With this method, only one has a C value lower than 16 - this is the Phenom II X6 1065T 95W version (at 15,06), but it can't be found on sale anywhere. I guess it COULD run cooler at, say 2,6GHz than the 910e, but also might not, since the correlation is not linear - i.e. C is not constant but only roughly (another matter is, why it is not constant).

Of course, this is just speculation and I might be wrong. Real-world tests would be needed. And I don't think there are a lot of people around to do the power measurements required for this

And, of course every chip is an individual, even in this aspect . If you're lucky, you might get one that runs at a lower voltage and lower current (even at load).

Any comments?

Cheers!

[crazycheese]

Hello, threadstarter.

I recommend Scythe Big Shuriken if you need both low height and efficient cooling. Yes, it is that amazing and silent. Dont forget hot air should be transfered out of the case and horizontal HTPC cases have very bad air blowout performance.

Also, please read http://phoronix.com/forums/showthrea...s-power-hunger
I recommend either core i3-540 or athlon II e- series. But in the end intel will consume less power in long run, while providing more performance. This is current situation.