A new undervolt, this time with an already somewhat old processor, Athlon 64 3000+ 939 and a GA-K8NSC-939 motherboard.
It is a pity that at the time I did not know about these details, or they did not interest me, the undervolt in this case is quite important and also surprising since the standard voltage is 1.4v, and it is perfectly stable at 1,150v (in the menu of the BIOS and on the mosfets with a multimeter, although the system monitor does not show that) a difference of 250mv, it starts at 1.1v but it is not totally stable.
The temperatures are also impressive with a simple heatsink as seen in the photograph.
I wonder what is the relationship between the voltage and the power dissipated by the processor, can someone explain how it works.

Well I tried it with a few things and often went too low for stock speeds but I vaguely remember that a k6-2+ was so cold at 1.3V/150MHz that it could run even without a heatsink and a Palomino 1700+ was stable at 1100MHz at a similar voltage (it was actually a messed up mod for overclocking 😁)

How about Athlon XP Barton working at top Thundebird speed with almost 1v?

Another curiosity, to what extent will undervolting extend the life of the hardware or reduce it, is it worth it?

Less heat = more life for temperature sensitive components like capacitors.

if I'm not mistaken, the fine control of the voltage of the motherboards CPU started in the Socket 462/370 era and in the case of GPUs I don't know, maybe around the Nvidia 8000 series era or soon after.

Radeon X1800/1900 had tweakable in realtime digital VRM. Although it was undocumented feature, which was discovered by third-party authors of tweakers.
First slider-style voltage control for Nvidia was implemented in GeForce GTX295 Revision A (the brick one) and some cream-of-the-crop AIB GTX285, followed by GTX 4xx series. Before that, some Nvidia cards had adjustable GPU voltage control via BIOS modding. Although they do not allow memory voltage changes up to this day, while some AMD cards allow it. Like Radeon 7970, for example.

alvaro84 wrote on 2022-11-27, 19:59:

Well I tried it with a few things and often went too low for stock speeds but I vaguely remember that a k6-2+ was so cold at 1.3V/150MHz that it could run even without a heatsink and a Palomino 1700+ was stable at 1100MHz at a similar voltage (it was actually a messed up mod for overclocking 😁)

Curious, I think that not many Socket S7 motherboards allow down to 1.3v, it would be interesting to know at what voltage the processor could maintain its standard frequency.
It is something like the opposite of overclock, to overclock you often have to raise the voltage because this is the opposite.

The Serpent Rider wrote on 2022-11-27, 20:11:

How about Athlon XP Barton working at top Thundebird speed with almost 1v? […]
Show full quote

How about Athlon XP Barton working at top Thundebird speed with almost 1v?

Athlon XP Barton low voltage.png

Another curiosity, to what extent will undervolting extend the life of the hardware or reduce it, is it worth it?

Less heat = more life for temperature sensitive components like capacitors.

if I'm not mistaken, the fine control of the voltage of the motherboards CPU started in the Socket 462/370 era and in the case of GPUs I don't know, maybe around the Nvidia 8000 series era or soon after.

Radeon X1800/1900 had tweakable in realtime digital VRM. Although it was undocumented feature, which was discovered by third-party authors of tweakers.
First slider-style voltage control for Nvidia was implemented in GeForce GTX295 Revision A (the brick one) and some cream-of-the-crop AIB GTX285, followed by GTX 4xx series. Before that, some Nvidia cards had adjustable GPU voltage control via BIOS modding. Although they do not allow memory voltage changes up to this day, while some AMD cards allow it. Like Radeon 7970, for example.

Interesting, because precisely my next plan involves a Semprom 3300+ socket 462 that I'm sure I have but haven't found yet and if not an Ahtlon XP 3200+ that is also somewhere.
My goal is to find the minimum voltages while maintaining the original frequencies. In the case of the Athlon XP it will be interesting because they seem to have a bad reputation for consuming a lot of energy and dissipating a lot of heat; although I never had any problem with a socket 462 processor.

In the case of the Core duo E6700 that I mentioned in the first post, it turns out that the motherboard, a P5L 1394, cannot be set to a voltage lower than 1.2v in the BIOS.

In the case of the Athlon XP it will be interesting because they seem to have a bad reputation for consuming a lot of energy and dissipating a lot of heat

Thunderbird and especially late Palomino cores were in fact dissipating ungodly amount of heat for that time.

I did an undervolting test with Pentium 2 and 3 here:
50 to 133MHz FSB on a BX Mainboard

The Serpent Rider wrote on 2022-11-27, 21:41:

Thunderbird and especially late Palomino cores were in fact dissipating ungodly amount of heat for that time.

Indeed. And that is also because these Athlon motherboards did not work well with CPU idling, so these were supplied with a BIOS which never enabled CPU halt and such. In that case the processor ran at 100% load, all the time.
Using a later Athlon motherboard, like the Asus A7V600, together with some tweaks, cuts the power usage to modest levels.

Hoping wrote on 2022-11-26, 18:28:

A new undervolt

25C with Prime running? are you living in a freezer?

Hoping wrote on 2022-11-27, 21:18:

Curious, I think that not many Socket S7 motherboards allow down to 1.3v, it would be interesting to know at what voltage the processor could maintain its standard frequency.
It is something like the opposite of overclock, to overclock you often have to raise the voltage because this is the opposite.

Not many indeed. I have met only one, GA-5AX and I just had to try.

Ok, I'm disappointed, I got two broken 462 motherboards for this experiment, an A7N8X Deluxe 2.0 and an A7N8X-X 2.0, I understood that they are the best with nForce 2 chipset, so far I only repaired the A7N8X-X and it works perfectly, stable at 11x200 and 1.7v in the mosfets according to my multimeter.
But my great disappointment is in the BIOS since they hardly have the possibility of undervolting, only overvolting, the minimum voltage that it allows for the CPU is 1,650v, which, although it is not exact, is very high for what I want, I expected at least to be able to testing 1.3v and 1.1v would be great.
I think I have seen modified BIOSes for these motherboards but on some page in German and I have not been able to find it.
Any ideas of what motherboard I could try to find that could be used for this experiment, that is, have a wide voltage control.

Hoping wrote on 2022-12-21, 11:46:

I expected at least to be able to testing 1.3v and 1.1v would be great.

I figure it is unlikely that motherboard designers add such undervolting functionality. Because it can easily result in a non-working or unstable system, resulting in time/money-wasting troubleshooting support calls.
Maybe you can modify the CPU itself?...
Back in the day I bought the Sempron 3000+ over its similarly-clocked Athlon variant, both being full featured Barton-cores. The only difference was that the Sempron had a slightly lower voltage request compared to the Athlon.

I repeat: The way to reduce power consumption of the original Athlon is to get it to work with CPU Halt / Stop-grant instructions.

gerwin wrote on 2022-12-21, 17:08:

I figure it is unlikely that motherboard designers add such undervolting functionality. Because it can easily result in a non-wo […]
Show full quote

Hoping wrote on 2022-12-21, 11:46:

I expected at least to be able to testing 1.3v and 1.1v would be great.

I figure it is unlikely that motherboard designers add such undervolting functionality. Because it can easily result in a non-working or unstable system, resulting in time/money-wasting troubleshooting support calls.
Maybe you can modify the CPU itself?...
Back in the day I bought the Sempron 3000+ over its similarly-clocked Athlon variant, both being full featured Barton-cores. The only difference was that the Sempron had a slightly lower voltage request compared to the Athlon.

I repeat: The way to reduce power consumption of the original Athlon is to get it to work with CPU Halt / Stop-grant instructions.

In this line of thought, I believe that if they do not implement a system that allows reducing the voltage, it is because that is not interesting for overclocking, and overclocking is what sells, if it allows a voltage of 1,850v to be established for the CPU, which in case of having the same error of 0.050v that I have seen would be 1.9v for the processor, and that can make the system unstable as well as reduce the voltage and in the long term what will happen does not need an explanation.
Since it didn't allow me to lower the voltage, I tried to overclock and the stable limit at 1.7v is 2400mhz, at 2500mhz it restarts just when it reaches the desktop, so it could be stable only at 1.75v, but that's not my goal in this case.
With these tests, I think the processor could work at its standard frequency with 1.5v, and it's a shame not to be able to test it.
Needless to say, I don't have any kind of temperature or power consumption problems with this processor nor have I ever had them with any Athlon XP, even the latest Thunderbirds never gave me any problems.
And where the hell is my Sempron 3300+...... I've got tired of searching it:)

Hoping wrote on 2022-12-21, 19:46:

Needless to say, I don't have any kind of temperature or power consumption problems with this processor nor have I ever had them with any Athlon XP, even the latest Thunderbirds never gave me any problems.

I am confused
- Why do you want to undervolt then?
- How do you define a problem with power consumption? Lets say, I would not be surprised if in total all the worlds k7 Athlons wasted more power (in kWh or Joules) doing nothing, then all Core-Duos used for doing something.

Back in the day I had two k7 Athlon systems. One day I measured the Wattage with such a power socket meter. Both with and without Halt/Stop-Grant. I remember thinking: what a waste.

Here are some old related snippets:
Socket-A AMD Athlon/Sempron system

But my power usage readings are not there, I suppose I still have them somewhere at home.

Admittedly, my usable overclocking experience comes from modern tech ( 2011 and onward ) because I didn't bother overclocking my stuff when I was younger. I didn't know much about it, never could afford better cooling solutions, and was rather scared of frying my stuff. Most current tech has overvoltage and thermal protections making it much harder to accidentally break your stuff.

I imagine many users already know this, but the reason you can usually undervolt is because stock voltage is set high in the interest of extra stability to support the widest range of chip quality. How far you can undervolt depends on your given chip quality; poor quality chips, and those that barely passed QC typically need more voltage to operate correctly, while pristine silicon is often stable at much lower voltage. Also as silicon lithography shrinks, so does the necessary voltage to allow the semiconductor work ( which also can mean it gets more sensitive to overvolt damage ). During the AM3 and FM2 days, AMD stock voltage was so high you could often reach a significant CPU overclock using less than stock voltage.

As for the question of whether undervolting can damage anything, that depends. Theoretically, yes, you can undervolt an IC to the point where it starts misbehaving and allowing signals to go down wrong paths where things get fried. Realistically, that's not a concern since you'd likely see system instability long before you get to that point. I think in most cases the CPU/GPU has a fairly isolated voltage supply that isn't shared by much else, so I wouldn't worry too much about accidentally damaging something else. Of course, the older the tech, the less likely that will be the case. And then it can also be whether you even have access to some voltage regulation settings ( EG, if the only way you can affect your GPU voltage is by changing the whole AGP voltage, yeah, you can risk collateral damage and stability ).

While a lot of manufacturers have made overclocking-specific models for a long time, remember a lot is also going to be restricted by manufacturing of the time. Fine-tuned voltage regulation circuitry like on today's motherboards likely would have been prohibitively expensive 20 years ago. Early BIOS had limited functions and controls because the ROM capacity was small ( my P2B mboard BIOS is 256KB, my Z68 mboard is 8MB , last gen motherboards I reviewed had 16MB firmware, and the X670E I now have on my test bench is 32MB ). And if you can't fit the controls into the software, then there are only so many jumpers you can put on a motherboard before manufacturing costs are too high or usability is too low.

gerwin wrote on 2022-12-21, 19:57:

I am confused
- Why do you want to undervolt then?
- How do you define a problem with power consumption? Lets say, I would not be surprised if in total all the worlds k7 Athlons wasted more power (in kWh or Joules) doing nothing, then all Core-Duos used for doing something.

This curiosity began with the case that I mentioned in the first post about the 486DX4 100mhz that turned out to be stable at 2.7v, even at 2.64v it was stable, but I left it at 2.7v. And so I thought it would be interesting to know the voltage range that older processors could have, when no one was interested in this, I'm really curious about everything, CPUs, RAM, etc., what range they had when few thought about these things.
When I talk about power or consumption of the K7s that I have had, I mean that it is common in the forum to read that they are unstable, that they need an "X" heatsink and a "Y" power supply, I mean that I never had any of those problems.
Regarding the RAM memory, at the moment I was surprised that the DDR3 RAM modules that I tested are stable at 1.30v and even if they are DDR3 1.5v, although unfortunately I do not have many samples at the moment in terms of RAM memory.

Jaron wrote on 2022-12-21, 21:18:

Admittedly, my usable overclocking experience comes from modern tech ( 2011 and onward ) because I didn't bother overclocking my […]
Show full quote

Admittedly, my usable overclocking experience comes from modern tech ( 2011 and onward ) because I didn't bother overclocking my stuff when I was younger. I didn't know much about it, never could afford better cooling solutions, and was rather scared of frying my stuff. Most current tech has overvoltage and thermal protections making it much harder to accidentally break your stuff.

I imagine many users already know this, but the reason you can usually undervolt is because stock voltage is set high in the interest of extra stability to support the widest range of chip quality. How far you can undervolt depends on your given chip quality; poor quality chips, and those that barely passed QC typically need more voltage to operate correctly, while pristine silicon is often stable at much lower voltage. Also as silicon lithography shrinks, so does the necessary voltage to allow the semiconductor work ( which also can mean it gets more sensitive to overvolt damage ). During the AM3 and FM2 days, AMD stock voltage was so high you could often reach a significant CPU overclock using less than stock voltage.

As for the question of whether undervolting can damage anything, that depends. Theoretically, yes, you can undervolt an IC to the point where it starts misbehaving and allowing signals to go down wrong paths where things get fried. Realistically, that's not a concern since you'd likely see system instability long before you get to that point. I think in most cases the CPU/GPU has a fairly isolated voltage supply that isn't shared by much else, so I wouldn't worry too much about accidentally damaging something else. Of course, the older the tech, the less likely that will be the case. And then it can also be whether you even have access to some voltage regulation settings ( EG, if the only way you can affect your GPU voltage is by changing the whole AGP voltage, yeah, you can risk collateral damage and stability ).

While a lot of manufacturers have made overclocking-specific models for a long time, remember a lot is also going to be restricted by manufacturing of the time. Fine-tuned voltage regulation circuitry like on today's motherboards likely would have been prohibitively expensive 20 years ago. Early BIOS had limited functions and controls because the ROM capacity was small ( my P2B mboard BIOS is 256KB, my Z68 mboard is 8MB , last gen motherboards I reviewed had 16MB firmware, and the X670E I now have on my test bench is 32MB ). And if you can't fit the controls into the software, then there are only so many jumpers you can put on a motherboard before manufacturing costs are too high or usability is too low.

Maybe I'll do overvolt mod to do an undervolt instead, overvolt mod where more comon. 😉

Undervolting for me is usually about heat and power consumption as well. When you're paying your utility bill, especially if funds are tight or your electricity is quite expensive, you tend to be more mindful of what appliances you leave turned on. Less heat coming from the computer means less heat being dumped into your room, which can be quite noticeable on warm summer nights when the door is closed so you don't disturb the sleeping family. It's also less heat for the air conditioner to deal with ( which means AC uses less electricity too ). Less heat means you can sometimes get by with a less expensive cooling solution, or that the fans don't need to spin so loudly. Years back I tested a liquid-cooled i7-6950X overclocked to 4.3 GHz all core. If I recall correctly, under Prime95, the CPU would draw more than double its 140W TDP.

As for your DDR3 stability, don't forget the DDR3L 1.35V standard. I'm sure plenty of memory dies were borderline 1.35V stable but they went into 1.5V sticks instead. Also, lots of motherboards cheat the RAM voltage by 0.1V or more, especially the DDR3 era. You may have it set to 1.3V, but it may be as high as 1.45V. Most of the time they do this so they can market a board as having a particularly high official RAM overclock support.