Practical underclocking and 45nm K10 lower limits

[zir_blazer]

For those that have an overall insight of general performance requeriments, is probable that many will agree with me in the fact that general purpose performance improvements aren't the same that they were 5 years ago or more. Basically, while a decade or so upgrading a computer was day and night, these days tangible performance can go only so far, and beyond that, is mostly requiered just for gaming or other performance intensive applications that aren't really of common usage. While some people (Mostly on this forum, considering that is ground of so-called extreme overclockers) are always power hungry no matter if they archieve it using practical and efficient methods or the inverse (And possibily have little means of capitalize it on some useful way that isn't just for ePenis comparision purposes), others like me that are conservative enough would possibily prefer to know how things are before nominal Frequency and Voltages and not beyond, that is where I am aiming at. The fact that most people can happily live with low performance can be easily proven due to the overall sucess of Netbooks and Intel Atom Processors, that just brings performance that was mainstream around 5 years ago yet for many purposes are still useful (Though I think Intel aimed much lower than they should with them). Besides, underclocking/undervolting is something available to everyone because is making your Hardware effort less, for overclocking you actually need to expend more money to make those overclocks stable, you got some chances of damaging it, etc.
Now, we are supposed to be Hardware enthusiasts, but for me, that doesn't means that the only think that I should care about is if a Thuban PH-E0 can reach 4 GHz or not (That is far from efficient), but more about architecture scaling as a whole (Maybe like this or this). So far, the K10 made using 45nm are very capable running at low Frequencies and Voltages (Too bad that BIOSes doesn't helps to reach them) consuming and dissipating very little power, and possibily being able to work finely under passive cooling. So I'm attemping to document what my Processor is capable of when underclocking while still having sufficient performance for common tasks when I'm not being exigent, then higher but efficient settings (Like, nominal Frequency but at the lowest possible stable Voltage).


My machine is the following:

Processor: AMD Athlon II X4 620 (ADX620WFK42GI / CACYC AC 0937DPMW)
CPU: 200 MHz * 13.0x = 2.6 GHz @1.325V IMC: 200 MHz * 10.0x = 2.0 GHz @ 1.175V HT = 200 MHz * 10.0x = 2.0 GHz
Heatsink: Retail one for this model, with Thermal Pad
Motherboard: ASUS M4A785TD-V EVO (AMD 785G (Radeon 4200, Sideport 128 MB DDR-III 1333 MHz)/SB710), PCB Rev. 1.01G / BIOS 0604
BIOS 2002
Memory Modules: 2x GSkill Ripjaws DDR-III / 2 GB / 1333 MHz (F3-10666CL8-2GBRM)
JEDEC #2 518 MHz 7-7-7-19-26 @ 1.5V
JEDEC #3 592 MHz 8-8-8-22-30 @ 1.5V
JEDEC #4 666 MHz 9-9-9-24-33 @ 1.5V
XMP-1334 667 MHz 8-8-8-27-32 2T @ 1.5V
Timmings CL-tRCD-tRP-tRAS-tRC-CR
Power Supply: Cooler Master Extreme Power Plus 460W RS-460-PMSR-A3

It doesn't POST if I enable ACC and Unleashing Mode, so while my Processor is a Deneb RB-C2, the Cache L3 seems to be damaged. My machine boots with the following settings:
Base Clock 200 MHz
CPU Multiplier 5x (1 GHz), if I recall correctly, it doesn't POST at 4x
IMC Multiplier 6x (1.2 GHz), if I recall correctly, it BSOD at WXP loading at 4x or 5x
HT Multiplier 6x (1.2 GHz), if I recall correctly, it BSOD at WXP loading at 4x or 5x
RAM setted in BIOS 400 MHz (800 MHz effective), automatic Timmings
GPU at default (500 MHz)
Sideport at default (1333 MHz)

All Voltages are default. The BIOS allows to undervolt the CPU and IMC by 0.3V relative to the Processor nominal Voltages, that means that I could boot with the CPU @ 1.025V as the bare minimum (Though at 1 GHz, the CPU part requieres just 0.8V). However, the BIOS Voltage settings works like some sort of offset, so k10stat display nominal Voltages though it is actually reciving less. I suppose than that is because the BIOS manages the VRM directly to supply a lower Voltage but doesn't make the Processor registers that manage this (Current Power State I suppose) aware.
While I have on the machine 24/7, it could be several days on Idle at pretty low Voltages with no issues, then sometimes greeting me with a few random resets here and there when using it, so it forces me to do to do stability testing. I started doing so using wPrime 2.03 with a single 1024M run, however, later I switched to OCCT 3.1.0 and figured out that in the OCCT test the machine resetted instantly on one setting that wPrime didn't presented issues, so I consider it more stressing. I also figured out later than OCCT Linpack is even more stressing than OCCT, at least for the IMC, so I could have to retest some settings.
So far, I have documented these settings:


CPU: 200 MHz * 13.0x = 2.6 GHz @ 1.125V IMC 200 MHz * 6.0x = 1.2 GHz @ 0.875V HT = 200 MHz * 6.0x = 1.2 GHz
25°C Idle / 40°C Full Load (CoreTemp 0.99.6) wPrime 2.03 1024M (497,484s)
JEDEC #1? 400 MHz 6-6-6-15-20 1T @ 1.5V Dual / Unganged

CPU 2.6 GHz @ 1.100V resets barely after setting it on k10stat



CPU: 200 MHz * 5.0x = 1.0 GHz @ 0.800V IMC 200 MHz * 6.0x = 1.2 GHz @ 0.875V HT = 200 MHz * 6.0x = 1.2 GHz
23°C Idle / 27°C Full Load (CoreTemp 0.99.6) wPrime 2.03 1024M (1290,781s)
JEDEC #1? 400 MHz 6-6-6-15-20 1T @ 1.5V Dual / Unganged




CPU: 200 MHz * 5.0x = 1.0 GHz @ 0.775V IMC 200 MHz * 6.0x = 1.2 GHz @ 0.875V HT = 200 MHz * 6.0x = 1.2 GHz
23°C Idle / 27°C Full Load (CoreTemp 0.99.6) wPrime 2.03 1024M (1290,015s)
JEDEC #1? 400 MHz 6-6-6-15-20 1T @ 1.5V Dual / Unganged

CPU 1.0 GHz @ 0.775V resets almost instantly with OCCT




The impressive low Voltage made me being doubtful of it so I attempted to test it with OCCT because it could run for infinite time, but then the machine reseted pretty much as soon as it started the test. I raised the Voltage to 0.8V and leave it all the night running (8 hours plus 1 that I did some Internet browser with it running on the background) with no issues.


CPU: 200 MHz * 5.0x = 1.0 GHz @ 0.800V IMC 200 MHz * 6.0x = 1.2 GHz @ 0.875V HT = 200 MHz * 6.0x = 1.2 GHz
23°C Idle / 27°C Full Load (CoreTemp 0.99.6) OCCT 3.1.0 OCCT LDS/Normal (9h 15m 24s)
JEDEC #1? 400 MHz 6-6-6-15-20 1T @ 1.5V Dual / Unganged




However, when I started doing the OCCT Linpack tests, it stopped after reporting an error, and happened two times in a row. So I raised the IMC Voltage from 0.875V to 0.9V and all was fine for the 2 hours or so that I let the test run (Though it had some Internet browsing mixed while doing the test). As a side note, until now I usually had the IMC at 0.85V that can explain the earlier random resets that weren't related to the rest of the CPU. 0.8V was also possible though the machine freezed as soon as I applied the setting, though the times that it didn't, it was fully usable so I didn't figured its weakness until playing something where it would sooner or later reset.


CPU: 200 MHz * 5.0x = 1.0 GHz @ 0.800V IMC 200 MHz * 6.0x = 1.2 GHz @ 0.900V HT = 200 MHz * 6.0x = 1.2 GHz
23°C Idle / 29°C Full Load (CoreTemp 0.99.6) OCCT 3.1.0 Linpack Max (2h 29m 42s)
JEDEC #1? 400 MHz 6-6-6-15-20 1T @ 1.5V Dual / Unganged


So far these are all my results, though I would have to fully test the first two with both OCCT and Linpack. I did some fast made calculations taking as reference this, and my Processor power consumption should be at 20-25W Full Load or even slighty less.

11.4A / 1.4V = 8.14A/V
8.14A/V / 3.4 GHz = 2.39A/V/GHz
2.39A/V/GHz * 1 GHz = 2.39A/V
2.39A/V * 0.8V = 1.91A
1.85A * 12V = 22.9W

I'm aiming at attemping to find the lowest Voltage settings to run fully stable at 1 GHz, something that I'm convinced that would run finely at even without the Fan, then find the lowest Voltage for its nominal 2.6 GHz if I ever need the performance.
In addition, while k10stat seems to manage pretty well a wide range of CPU Frequencies and Voltages for both the CPU and IMC, it pretty much lacks anything about IMC Frequency as the only thing that it got is a divisor that cuts in half current IMC Frequency (From 1.2 GHz to 600 MHz), so it could be more useful to boot with the IMC at its nominal 2 GHz, then use that divisor to halve it to 1 GHz. This means that I could fully switch Processor settings within Windows from Idle or light usage to full power, though I have to see what I could do about the Memory Modules.


Enough writting for now until I do more testing and can present it in a more... understandable way. Any constructive suggestions? Save yourself the "If it is too fast for you, why you buyed it?" comment because it is a pretty worthless and ignorant statement. More useful ones are welcome though.

[zir_blazer]

Tested with OCCT Linkpack and it was fine for 11 consectuvive hours, with CPU 1 GHz @ 0.8V and IMC 1.2 GHz @ 0.9V, with temperatures varying from 23°C at Idle and rare 29°C peaks at Full Load.

But as no one seems to be paying attention to this Thread, there is no use on keeping it alive.

[wez]

Tested with OCCT Linkpack and it was fine for 11 consectuvive hours, with CPU 1 GHz @ 0.8V and IMC 1.2 GHz @ 0.9V, with temperatures varying from 23°C at Idle and rare 29°C peaks at Full Load.

But as no one seems to be paying attention to this Thread, there is no use on keeping it alive.

I'd be interested in seeing a couple benchmarks at those speeds

[demonkevy666]

wow you even under clocked the Nb that's really extreme
how much less watts does it draw vs the normal Nb speed ?

[radaja]

zir_blazer,wow,this is a great idea.

wow you even under clocked the Nb that's really extreme
how much less watts does it draw vs the normal Nb speed ?

im also curious as to the difference between stock and underclocked
watts and overall power draw.

[zir_blazer]

I'd be interested in seeing a couple benchmarks at those speeds

You tell what ones you like to see. Keep in mind that the less the GPU matters on them, the best, because I'm using the IGP (Radeon 4200), so for gaming is just adequate at very low resolutions and settings or slighty old games.

wow you even under clocked the Nb that's really extreme
how much less watts does it draw vs the normal Nb speed ?

That is a really good question that I'm wondering myself too, but I don't have any means to measure power usage. While there are articles that did an interesing job on CPU power consumption like the one of X-Bit Labs, they didn't analyze anything about the IMC. I'm basically clueless on the power consumption of the IMC in an isolated state, however, that it should consume less than at nominal settings, that's for sure.

[zir_blazer]

I was scared after having looking at undervolting results from the new BL-C3 Propus Core, 2.4 GHz @ 1.0V is total destruction. Remember that my AIIX4 is a Deneb RB-C2 that was superceded quite a bit ago with the Revision RB-C3 ones, but isn't that bad to know that you could finally get the C3 improvements in an AIIX4.

[zir_blazer]

*BUMP*

@wez
Still waiting on how you're wanting me to benchmark this.


The machine is still running flawlessly and rock solid day and night with the last settings, though I didn't do any further testing to see if I can use k10stat to make the 2 GHz to 1 GHz jump of the IMC so I can fully switch from low power to performance, nor lowest possible Voltage for 2.6 GHz (Well, I didn't even set it again). Another interesing thing is knowing if different Motherboard BIOSes Voltage settings controls just the supplied Voltage to the Processor, or also makes sure than the Processor reports the Voltage that it is running at via P-States info, VID, or whatever, because it would also be interesing if the BIOS Voltage and that reported by k10stat worked in sync.

[wez]

Still waiting on how you're wanting me to benchmark this.

A Cinebench run at 1ghz would be cool, just to get an idea of the performance

[zir_blazer]

I mostly don't like Synthethic Benchmarks because they don't reflect real world usage, but as is the only one that you suggested...

[cezar]

I mostly don't like Synthethic Benchmarks because they don't reflect real world usage, but as is the only one that you suggested...

Cinebench has an engine from the real world 3D application: CINEMA 4D.

[demonkevy666]

2C temperature change during load lol

[zir_blazer]

2C temperature change during load lol

Actually the delta is 6°C, because my Idle is 23°C, while Full Load is 29°C, but I didn't let the temperature go lower than 27°C as I was playing games before opening CoreTemp then running CineBench. Also, we are in winter in Argentina, and my Case is also a cheap generic piece of junk with no Fans on it (But similar to what you would expect from a fully passive cooling computer).

Cinebench has an engine from the real world 3D application: CINEMA 4D.

But the scores are pretty much meaningless because they aren't representative of how other things do, or general system usability or responsiveness. It could be useful to compare general system performance against other machine, but it doesn't tell me how much tangible such difference can be for me, as for my feels everyday usage reachs a performance cap even on budget computers.

My Windows XP SP3 boots usually with 16 or 17 Threads, though after stabilizing on Idle it keeps just 15. Its extremely clean considering that the installation is more than one year old, and it could be four years old already if it wasn't because the previous two installations catched an EXE infector virus (My mother executed an infected file in my machine that she transfered on a Pendrive) that forced me to reinstall, and delete tons of things included executable ZIP or RAR format files (The autodecompressable ones).
General applications include Windows Messenger Live 8.5 that is always running, Firefox pretty much always running too but on Idle (With the exception that I browsing like on a Forum with a lot of open Tabs), and also gaming with old games that includes Warcraft 3 (DOTA), zDoom port of Doom with custom WADs, Ultima Online (Classic client on a Freeshard), Ragnarok Online, Freelancer (Not anymore though), Teamspeak or Ventrilo, Dolphin (Nintendo Wii emulator, I usually play Final Fantasy 4: The After), and that pretty much covers everything that I currently do. At 1 GHz, everything seems as fast as in my old A64 (And for most things it wasn't that faster than my old AXP Palomino). Besides, there are chances that for more recent and heavy games, all them should be GPU bottlenecked by the integrated Radeon 4200 before even complaining about the Processor.
With the exception of Dolphin, as I didn't tested it on my old A64, everything runs pretty much the same, though the system feels more responsible under heavy load (Example, if I am on Windows Desktop talking by MSN while Warcraft 3 loads a map on a multiplayer match on the background goes basically unnoticed, while in my A64 it had an slighty freeze when it started to load). The only moment when I actually needed the 2.6 GHz was when running a Spear of Destiny mod on DOSBox (End of Destiny) that forced me to run it on nominal to have it perform on real time.
However, pretty much Hardware components these day are an overkill in performance for common usage, and that is very good for the common consumer. So that is why I aim for being just above the tangible performance cap, and use the extra margings for lowering Frequency and Voltage to aim for lower power consumption and the possibility to run it on passive cooling. There is a massive difference on raw power on what you can do actually do on low power modes compared to nominal or overclocked, yet still they are pretty confortable for most things.

[demonkevy666]

Actually the delta is 6°C, because my Idle is 23°C, while Full Load is 29°C, but I didn't let the temperature go lower than 27°C as I was playing games before opening CoreTemp then running CineBench. Also, we are in winter in Argentina, and my Case is also a cheap generic piece of junk with no Fans on it (But similar to what you would expect from a fully passive cooling computer).



But the scores are pretty much meaningless because they aren't representative of how other things do, or general system usability or responsiveness. It could be useful to compare general system performance against other machine, but it doesn't tell me how much tangible such difference can be for me, as for my feels everyday usage reachs a performance cap even on budget computers.

My Windows XP SP3 boots usually with 16 or 17 Threads, though after stabilizing on Idle it keeps just 15. Its extremely clean considering that the installation is more than one year old, and it could be four years old already if it wasn't because the previous two installations catched an EXE infector virus (My mother executed an infected file in my machine that she transfered on a Pendrive) that forced me to reinstall, and delete tons of things included executable ZIP or RAR format files (The autodecompressable ones).
General applications include Windows Messenger Live 8.5 that is always running, Firefox pretty much always running too but on Idle (With the exception that I browsing like on a Forum with a lot of open Tabs), and also gaming with old games that includes Warcraft 3 (DOTA), zDoom port of Doom with custom WADs, Ultima Online (Classic client on a Freeshard), Ragnarok Online, Freelancer (Not anymore though), Teamspeak or Ventrilo, Dolphin (Nintendo Wii emulator, I usually play Final Fantasy 4: The After), and that pretty much covers everything that I currently do. At 1 GHz, everything seems as fast as in my old A64 (And for most things it wasn't that faster than my old AXP Palomino). Besides, there are chances that for more recent and heavy games, all them should be GPU bottlenecked by the integrated Radeon 4200 before even complaining about the Processor.
With the exception of Dolphin, as I didn't tested it on my old A64, everything runs pretty much the same, though the system feels more responsible under heavy load (Example, if I am on Windows Desktop talking by MSN while Warcraft 3 loads a map on a multiplayer match on the background goes basically unnoticed, while in my A64 it had an slighty freeze when it started to load). The only moment when I actually needed the 2.6 GHz was when running a Spear of Destiny mod on DOSBox (End of Destiny) that forced me to run it on nominal to have it perform on real time.
However, pretty much Hardware components these day are an overkill in performance for common usage, and that is very good for the common consumer. So that is why I aim for being just above the tangible performance cap, and use the extra margings for lowering Frequency and Voltage to aim for lower power consumption and the possibility to run it on passive cooling. There is a massive difference on raw power on what you can do actually do on low power modes compared to nominal or overclocked, yet still they are pretty confortable for most things.

I play freelancer too, online.
dolphin is very cpu clock heavy. (the wind waker is one game also, twilight princesses is gpu memory hog heavy).

[zir_blazer]

I'm doing new tests and having been facing an interesing issue or bug. So far, I was doing test to archieve the lowest stable Voltage for the CPU and IMC, the issue is what I will explain at the end...


Hardware

Processor: AMD Athlon II X4 620 (ADX620WFK42GI / CACYC AC 0937DPMW) Deneb RB-C2
CPU: 200 MHz * 13.0x = 2.6 GHz @1.325V IMC: 200 MHz * 10.0x = 2.0 GHz @ 1.175V HT = 200 MHz * 10.0x = 2.0 GHz
Heatsink: Retail w/Thermal Pad
Motherboard: ASUS M4A785TD-V EVO (AMD 785G (Radeon 4200, Sideport 128 MB DDR-III 1333 MHz)/SB710), PCB Rev. 1.01G / BIOS 0604
BIOS 2005
Memory Modules: 2x GSKILL Ripjaws DDR-III / 2 GB / 1333 MHz (F3-10666CL8-2GBRM)
SPD JEDEC #2 518 MHz 7-7-7-19-26 @ 1.5V
SPD JEDEC #3 592 MHz 8-8-8-22-30 @ 1.5V
SPD JEDEC #4 666 MHz 9-9-9-24-33 @ 1.5V
SPD XMP-1334 667 MHz 8-8-8-27-32 2T @ 1.5V
RAM Timmings CL-tRCD-tRP-tRAS-tRC-CR
Hard Disk: SAMSUNG SpinPoint F3 1 TB (HD103SJ) Rev. A
Firmware 1AJ100E4
DVD-RW: ASUS DRW-2014L1T
Power Supply: COOLER MASTER Extreme Power Plus 460W (RS-460-PMSR-A3)


Software

OS: Windows XP Professional SP3
CPU Temperature: CoreTemp 0.99.6
CPU Settings (VID): K10stat 1.41
FAN Regulation: SpeedFan 4.40





Lower Voltage for CPU w/CPU 2.6 GHz / IMC 2 GHz


CPU: 200 MHz * 13.0x = 2.6 GHz @ 1.3250V IMC 200 MHz * 10.0x = 2.0 GHz @ 1.1750V HT = 200 MHz * 10.0x = 2.0 GHz
JEDEC #4 666 MHz 9-9-9-24-33 1T @ 1.5V Dual / Unganged

OCCTPT 3.1.0 Linpack Max 1h 15m 28°C Idle / 53°C Full Load CPU FAN: 100% 3068 RPM


CPU: 200 MHz * 13.0x = 2.6 GHz @ 1.1500V IMC 200 MHz * 10.0x = 2.0 GHz @ 1.1750V HT = 200 MHz * 10.0x = 2.0 GHz
JEDEC #4 666 MHz 9-9-9-24-33 1T @ 1.5V Dual / Unganged

OCCTPT 3.1.0 OCCT LDS/Normal 10h 30m 24°C Idle / 46°C Full Load CPU FAN: 100% 3096 RPM
OCCTPT 3.1.0 Linpack Max 9h 40m 26°C Idle / 45°C Full Load CPU FAN: 100% 3140 RPM
OCCTPT 3.1.0 Linpack Max 4h 10m 28°C Idle / 66°C Full Load CPU FAN: 0% 1020 RPM


CPU: 200 MHz * 13.0x = 2.6 GHz @ 1.1250V IMC 200 MHz * 10.0x = 2.0 GHz @ 1.1750V HT = 200 MHz * 10.0x = 2.0 GHz
JEDEC #4 666 MHz 9-9-9-24-33 1T @ 1.5V Dual / Unganged

OCCTPT 3.1.0 OCCT LDS/Normal 5m 28°C Idle / 44°C Full Load CPU FAN: 100% 3111 RPM ERROR CORE 2
OCCTPT 3.1.0 Linpack Max 2h 5m 28°C Idle / 44°C Full Load CPU FAN: 100% 3111 RPM



Lower Voltage for IMC w/CPU 2.6 GHz / IMC 2 GHz


CPU: 200 MHz * 13.0x = 2.6 GHz @ 1.1500V IMC 200 MHz * 10.0x = 2.0 GHz @ 1.1250V HT = 200 MHz * 10.0x = 2.0 GHz
JEDEC #4 666 MHz 9-9-9-24-33 1T @ 1.5V Dual / Unganged

OCCTPT 3.1.0 Linpack Max 3h 25°C Idle / 45°C Full Load CPU FAN: 100% 3096 RPM


CPU: 200 MHz * 13.0x = 2.6 GHz @ 1.1500V IMC 200 MHz * 10.0x = 2.0 GHz @ 1.0750V HT = 200 MHz * 10.0x = 2.0 GHz
JEDEC #4 666 MHz 9-9-9-24-33 1T @ 1.5V Dual / Unganged

OCCTPT 3.1.0 Linpack Max 8h 45m 26°C Idle / 45°C Full Load CPU FAN: 100% 3096 RPM


CPU: 200 MHz * 13.0x = 2.6 GHz @ 1.1500V IMC 200 MHz * 10.0x = 2.0 GHz @ 1.0250V HT = 200 MHz * 10.0x = 2.0 GHz
JEDEC #4 666 MHz 9-9-9-24-33 1T @ 1.5V Dual / Unganged

OCCTPT 3.1.0 Linpack Max 6h 30m 25°C Idle / 46°C Full Load CPU FAN: 100% 3111 RPM


CPU: 200 MHz * 13.0x = 2.6 GHz @ 1.1500V IMC 200 MHz * 10.0x = 2.0 GHz @ 1.0000V HT = 200 MHz * 10.0x = 2.0 GHz
JEDEC #4 666 MHz 9-9-9-24-33 1T @ 1.5V Dual / Unganged

OCCTPT 3.1.0 OCCT LDS/Normal 6h 30m 23°C Idle / 45°C Full Load CPU FAN: 100% 3125 RPM CPU TOO HOT



Yes, what you hear. OCCT failed the test after 6h 30m because it says that the CPU is "too hot". According to the attached graphic log, Core 1 hitted 250°C before the test finished, something that is obviously a bug, because the machine didn't freezed, reseted, melted, or anything. The other Cores according to OCCT didn't reported abnormal temperature readings (Though I don't know where they get their readings because K10 Processors are supposed to have just a single temperature diode), just Core 0, and CoreTemp reported normal temperatures too.
In addition, before doing the OCCT stress test I also tested with Linpack two times. Both times reseted during night at what I thinked that were around the six or seven hour mark for no apparent reason. I suppose that the IMC Voltage is too low (At 1.0250V it didn't presented the same weird behaviator that at 1.0000V), but when that happens Linpack is supposed to report error instead of the machine reseting itself. That only happens consistently when you have a critically low Voltage (For example, 0.8V to 0.85V with IMC @ 1.2 GHz that requiered 0.9V to be fully stable) that could cause random crashes instead of just causing detectable errors under stress test that aren't usually critical enough to reset the machine. So I'm basically puzzled about this weird behaviator.

[AmdAtiTom420]

You would think a topic like this would generate more interest among hardware enthusiasts and/or environmentalists.

Part markers give the impression they want to be " green ", but your results prove they are barely trying. Both the environment and your pocket book will benetfit if systems had Xtreme power saving options !

[Tom128]

I've not done this extensive of testing but your results are pretty close to what I have gotten as well. Especially on my HTPC I find undervolting fun and it has a functional side too (lower power consumption and temps).

I've had a Phenom 9600BE, Phenom II 955 and Sempron 140 (unlocked to X2) in two different motherboards in my HTPC. There seems to be an architectural minimum around 0.800v. I've tried even 400Mhz and it seems to crash in that area. On the flip side it will run towards 1Ghz or higher at that voltage. Only exception was my 955BE, it ran at 0.725v stable, but even 0.715v would insta-crash.

Both Intel and AMD seem to set fairly high voltages as a "just incase" but AMD in particular seems to really pick higher idle/load voltages than needed.