Too generic.

On later AGP cards you can modify this in software/bios. On earlier you'll have to look up datasheet for controller used, figure out how voltage is configured and modify that.

IMO not worth it if you have to mod hardware, just ensure good cooling...

More cooling / better heat extraction is what you should focus on. I don't think that lowering core voltage by a few 0.1s if volts that I expect you'd get before instability kicks in will make a huge difference in temperature.

In any case, if you still want to mess around- find the core voltage regulator on the card and bypass it with an adjustable one 😊

Highly depends on the card. As others have written.

There is no adapter for undervolting, as the voltages coming from the AGP port (apart from VDDQ) as well from external inputs are used to feed regulators for the needed voltages. Any change of the input voltages (within margin) should not have any influence on things like the GPU voltage.

If you want to know the actual power consumption of some cards, you might wanna check out the measurements I provide here https://tehsiggi.github.io/agp-power-monitor/

Undervolting the GPU would probably have to go hand in hand with underclocking. So not sure if it is worth it. It's imho much more important to properly cool the card and give it some good airflow.

I really don’t see the point in this. First, the amount of undervolting you can achieve without sacrificing performance is meaningless for the lifespan. Second, these used cards have most of their hours already behind them from the era when they were purchased and originally used. The hours they get in retro use are often insignificant, especially with those who have several different systems. Third, graphics chip degradation is only one failure mode and it is IMO one of the least likely. Capacitors fail, memory chips are prone to failure, BGA solder balls break either under the graphics chip or memory etc.

While theoretically decreasing the graphics chip temp you decrease the severity of he heat cycle for BGA joints, it is pretty much meaningless. Like I said, these cards have most of their power cycles already behind them. Shaving couple of degrees off from the temps is also not going to change anything. If those joints are going to crack, the process has already started.

Like others have said, the best preventive thing you can make is to ensure that cards have good airflow and if necessary, improve cooling. Some cards such as radeon 9700 and 9800 have atrocious heatsinks and fans, so if you happen to have a working card, it would be a good idea to change the original cooler to something more effective.

You can undervolt reference GeForce 6800LE/6800/6800GT/6800 Ultra and some Radeon X1950 Pro with digital VRMs.

I’ve noticed something over the years that might be useful for those of us running older hardware. If you drop both voltage and clock speeds by around 15%, the difference in stability and longevity is huge.

I’ve been in CGI rendering for 15 years, and my GPUs/CPUs often run non-stop for days at a time. From direct experience, I can tell you: reducing power by 15% keeps the system much cooler and way more stable. The result? The hardware lifespan increases in ways that are almost impossible to predict—it just lasts far longer than expected.

That’s why I believe the same principle applies to retro PCs. Stability changes everything, and sometimes running hardware at “just under full power” can be the key to keeping it alive for decades.

Well, that is always true, but what's the deal for old hardware here?
Take the Radeon 9X00 series for example. You can not undervolt them without a hardware mod. I'd rather go leaving those cards as they are instead of performing a hardware modification. Reducing the clock alone does not make a huge difference from my testing. And I measure the *exact* power consumption of the cards on all power rails.
Those cards have been on the edge with where they run, however keeping things cooler can also just be achieved by better cooling. This will not reduce the overall power dissipation of course, but drastically reduce the amount of thermal stress on components.

I agree with what’s been said, but I think one important point is often overlooked: stability. The reason I say that reducing system power by about 15% increases stability (and durability) is because it’s not just about the GPU or CPU. The reduced current draw affects every component in the circuit—resistors, capacitors, transistors, everything. Lowering the stress just a bit eases the load on the entire system.

In fact, many failures are caused not only by heat, but also by these less visible factors in the background. A 15% performance drop is usually hardly noticeable in practice, but for retro hardware—especially those cards that were pushed close to their limits in design—it can actually give them “room to breathe.”

That’s why what I’m really interested in is undervolting directly through the BIOS.

There are only a few cards that support it, as mentioned by the serpent rider.
The rest of AGP is basically out of the picture with regard to bios/software based undervolting.

You talk about degradation of parts. Most degradation is the result of heat, which is caused by current flow. Capacitors have an internal resistance, which in combination with the current pushed into/out of them cause heat inside the capacitor which degrades the electrolyte.
Safe-operating-areas for semiconductors are very dependent on temperature.
Most schematics I have seen so far are well over-spec'd in terms of Mosfets etc used.

A 15% performance drop with retro hardware can mean the difference between 34fps and 40fps, which is imho significant.

I still stand by my previous statement: It's probably easier and more practical to improve cooling of those cards, than relying on undervolting.

Let’s clarify a few points:

You must lower the core clock first, then reduce the voltage. If you do it the other way around, the card will of course fail.

Another point: 30 FPS won’t suddenly drop — at most you’ll see a drop of roughly 5–8 FPS.

I think you’ve badly misunderstood this 15% reduction. Most cards were pushed to their limits for competition and weren’t designed for long-term use. We retro enthusiasts need to wake up to that.

This topic is important and it’s sad that it’s so little discussed. Retro parts will soon run out. I’ve actually been doing this for years — I opened this thread so other retro fans could benefit from it and think about the issue.

tarik wrote on 2025-09-11, 08:31:

Let’s clarify a few points: […]
Show full quote

Let’s clarify a few points:

You must lower the core clock first, then reduce the voltage. If you do it the other way around, the card will of course fail.

Another point: 30 FPS won’t suddenly drop — at most you’ll see a drop of roughly 5–8 FPS.

I think you’ve badly misunderstood this 15% reduction. Most cards were pushed to their limits for competition and weren’t designed for long-term use. We retro enthusiasts need to wake up to that.

This topic is important and it’s sad that it’s so little discussed. Retro parts will soon run out. I’ve actually been doing this for years — I opened this thread so other retro fans could benefit from it and think about the issue.

Can you give an example for a card that would be considered "pushed to the limit"? Apart from cooling i don't know of any card schematic i have seen in the years for AGP cards where i would have said: Wow.. This is pretty on the edge.

Even ATI official docs state possible voltages for R350 that are higher than what is used in the end.

The topic of preservation is important, but it is also a question of practicality i think.

Underclocking itself, without undervolting is mostly pointless, unless you're willing to half the speed of the GPU. It works only as a short term solution for failing capacitors. The same goes in other way, there's no real harm from overclocking unless you start to overvolt, which was not possible on most AGP cards anyway.

This applies to almost all high-end cards. I’m not writing this for debate, but for awareness.

What I’m talking about here is very important. I’ve been pushing high-end PCs for rendering work for 15 years, and I’m simply sharing what I’ve learned from that experience. Instead of debating, I think we should actually test it.

Stability and temperature are two very different topics, and explaining them in full would take a long time. My only goal here is to help retro enthusiasts extend the lifespan of their hardware.

Please read again, i never stated a 30fps difference anywhere. The difference between 34 and 40 fps. Which is 6fps.

Yes, you’re right, I misunderstood that. My apologies.

Most cards were pushed to their limits for competition and weren’t designed for long-term use.

Majority of old cards actually had pretty hefty headroom, as it should be, that's why overclocking was possible in the first place. Modern GPUs, on the other hand, really push it to the limit.

Absolutely right, I agree with this too. 😀 Maybe it’ll spark someone’s curiosity and they’ll make a video about it—lowering the core clock through the BIOS first, then reducing the voltage.

To minimize potential risks, it might be better to test this on motherboards that have both PCIe and AGP slots, so the BIOS can be reflashed if anything goes wrong. I’m just putting these ideas forward as suggestions. 😀

The Serpent Rider wrote on 2025-09-11, 09:09:

Most cards were pushed to their limits for competition and weren’t designed for long-term use.

Majority of old cards actually had pretty hefty headroom, as it should be, that's why overclocking was possible in the first place. Modern GPUs, on the other hand, really push it to the limit.

I agree.

Modern graphics cards are the component that most frequently fail.

From my trident 9000 to the geforce2 mx still work. The geforce 4, 6, 8 and 200 series I had since, has all failed. I haven't even overclocked them.

Haven't purchased a graphics card since, running on stuff I can pick up for free and keep using them until they fail.

st31276a wrote on 2025-09-11, 09:25:

I agree. […]
Show full quote

The Serpent Rider wrote on 2025-09-11, 09:09:

Most cards were pushed to their limits for competition and weren’t designed for long-term use.

Majority of old cards actually had pretty hefty headroom, as it should be, that's why overclocking was possible in the first place. Modern GPUs, on the other hand, really push it to the limit.

I agree.

Modern graphics cards are the component that most frequently fail.

From my trident 9000 to the geforce2 mx still work. The geforce 4, 6, 8 and 200 series I had since, has all failed. I haven't even overclocked them.

Haven't purchased a graphics card since, running on stuff I can pick up for free and keep using them until they fail.

That’s exactly why I started this topic 😀 to find new ways to keep these cards alive. Cooling is important of course, but instead of focusing only on that, I’m talking about creating new BIOS versions. The goal should be to ease the load on the cards, ignoring small differences like an 8 FPS drop.