Awesome! Thank you ChrisK!

The soldering skill level seems slightly higher than what I'd normally do because of the SMD stuff, but maybe we can adapt this to some other components, using your idea as a base?

Either way, this is a great idea to increase the longevity of these cards. They may survive for decades with the stock regulator, but a lot of them tend to start turning brown after a lot of years of use due to heat. This will prevent that. 😀

Very nice job. That looks amazing! 😀

I added your project to the vrm projects thread. 😀

Re: vrm module project thread roundup, share ideas, make new designs

I have an aversion against switching components, mildly said.
I rather prefer calm, less efficient linear components.

Congratulations for your success, though. 🙂👍

Sphere478 wrote on 2023-01-14, 21:20:

Very nice job. That looks amazing! 😀

I added your project to the vrm projects thread. 😀

Re: vrm module project thread roundup, share ideas, make new designs

Didn't know there was one. Thanks! 😁

Jo22 wrote on 2023-01-14, 22:15:

I have an aversion against switching components, mildly said.
I rather prefer calm, less efficient linear components.

Congratulations for your success, though. 🙂👍

I don't like modifying old hardware and making it non-original in general.
But that one sucked. It got extremely hot and beeing a THT part that was "pressed" against the PCB with a (at the point I bought the card missing) screw wasn't the perfect solution.
The thermal paste between the heatsink and the PCB began to spread as soon as it got on temperature. Wasn't nice at all.
There are better solutions, e.g. choosing a SMT LDO just like Diamond did on their V550 PCI:
https://www.vgamuseum.info/index.php/componen … nvidia-riva-tnt
So, my solution may be reasonable for THT LDOs only.

Thanks for the like 😀

Hey!
In Voodoo3 2000/3000 PCI I replaced EZ1585 with APW7120+2SK3918. Now this unit does not heat up.
The PCB "hand made" with Sprint Layout 5

In Voodoo3 2000/3000 AGP I replaced EZ1580/EZ1582 with APW7120+2SK3918.
It’s not that all of this was absolutely necessary, but once you start doing it, it’s hard to stop:)
The PCB "hand made" with Sprint Layout 5

Wow, I'm really impressed by the designs people are coming up with for these.

Just curious... are there any slightly more efficient replacements for the old parts that will work without having to do so much custom work? Even a moderate reduction in heat output would help a lot of these old cards.

Not with linear regulators. Those just "eliminate" the difference between input and output voltage by "generation" of a specific resistance and therefore heat.
With the same difference from input to output there's always the almost same amount of heat generated. For an output voltage of let's say 2.5V and an input voltage of 5V you'd waste half of the energy. So the best case for using a linear regulator is when this voltage difference in minimal (also depends on the regulator type used, which needs a minimal amount of dropout voltage itself to work correctly).

The only way to mitigate this is by using a switching regulator. With these you can reach efficiencies of at least 70-80%, best in class even 90% and more so that only a little amount of energy is transferred into heat. Sure you know most of this already.

This is also the way industry has gone from no regulators at all for 5V 486 CPUs (remember: 5V came directly from the AT power supply), linear regulators for 3.3V 486 and first Pentiums (AT PSU didn't have 3.3V rails), and later on switching regulators for higher power/lower voltage CPUs (Pentium MMX/K6-2/...).
The pinnacle of this development was reached when mainboard designers had to switch to 12V-input regulators for CPU voltage generation because of the immense currents that were needed from the 5V rails when generating ~70W @ 1.xxV for the latest Athlon XPs. With 12V you had less current on the rail for the same power.
Just think about generating a Vcore of ~1V for a 130W CPU. That is 130A of output current and 130W/12V@95% efficiency = 11.4A input current.

There are drop-in replacments for linear regulators (RECOM does build those for example) but I'm not sure if they can output enough current. I doubt they can but am not up to date with the current lineup companies offer. Historically they were mostly suited for general applications where the original linear regulator had may be 500mA or 1A of output current capability.

ChrisK wrote on 2025-01-23, 07:45:

Not with linear regulators. Those just "eliminate" the difference between input and output voltage by "generation" of a specific […]
Show full quote

Not with linear regulators. Those just "eliminate" the difference between input and output voltage by "generation" of a specific resistance and therefore heat.
With the same difference from input to output there's always the almost same amount of heat generated. For an output voltage of let's say 2.5V and an input voltage of 5V you'd waste half of the energy. So the best case for using a linear regulator is when this voltage difference in minimal (also depends on the regulator type used, which needs a minimal amount of dropout voltage itself to work correctly).

The only way to mitigate this is by using a switching regulator. With these you can reach efficiencies of at least 70-80%, best in class even 90% and more so that only a little amount of energy is transferred into heat. Sure you know most of this already.

This is also the way industry has gone from no regulators at all for 5V 486 CPUs (remember: 5V came directly from the AT power supply), linear regulators for 3.3V 486 and first Pentiums (AT PSU didn't have 3.3V rails), and later on switching regulators for higher power/lower voltage CPUs (Pentium MMX/K6-2/...).
The pinnacle of this development was reached when mainboard designers had to switch to 12V-input regulators for CPU voltage generation because of the immense currents that were needed from the 5V rails when generating ~70W @ 1.xxV for the latest Athlon XPs. With 12V you had less current on the rail for the same power.
Just think about generating a Vcore of ~1V for a 130W CPU. That is 130A of output current and 130W/12V@95% efficiency = 11.4A input current.

There are drop-in replacments for linear regulators (RECOM does build those for example) but I'm not sure if they can output enough current. I doubt they can but am not up to date with the current lineup companies offer. Historically they were mostly suited for general applications where the original linear regulator had may be 500mA or 1A of output current capability.

Interesting! Thank you for the explanation.

I get now why the PCI cards tend to have the ridiculously hot running 3.3v regulators. The card is relying on the 5v supply from the PCI slot and using the regulator to drop it down to 3.3v, so there is a lot of waste heat.

What voltage does the EZ1580/EZ1582 regulator output? Looking at the datasheet it says that if there isn't a voltage specified (3.3, 2.5, etc.) then it is adjustable.

Also, I'm sure you can tell from my questions I have no electronics background, so forgive me if this doesn't make sense... but could a PCI Voodoo3 simply have a wire jumped from the (unused) 3.3v supply pin on the PCI slot to bypass that absurdly hot 3.3v regulator? Obviously it would impact compatibility with boards that weren't properly wired for 3.3v PCI cards, but it seems like it would help a lot to reduce the card's thermals. On the SGRAM models the PCB under that regulator is usually the first area to turn brown.

In line with that, seeing all of these regulator designs is making me imagine an add-in card that efficiently creates the voltages that these old hot-running cards expect. If only there were a quick way to connect an external supply to a card without having to desolder a bunch of components. Maybe just a small "adapter" that solders in place of the regulator and connects to a universal voltage regulator card or something. Could be a pretty neat thing to increase longevity of these little space heaters. 😀

Hmm, something happened to my reply from last Friday ... I'll try to replicate ...

Ozzuneoj wrote on 2025-01-23, 20:48:

I get now why the PCI cards tend to have the ridiculously hot running 3.3v regulators. The card is relying on the 5v supply from the PCI slot and using the regulator to drop it down to 3.3v, so there is a lot of waste heat.

Right!

Ozzuneoj wrote on 2025-01-23, 20:48:

What voltage does the EZ1580/EZ1582 regulator output? Looking at the datasheet it says that if there isn't a voltage specified (3.3, 2.5, etc.) then it is adjustable.

I'd copy that.
Sometimes there's 3.3V written on one regulator. So the other one needs to be adjustable (see below).

Ozzuneoj wrote on 2025-01-23, 20:48:

Also, I'm sure you can tell from my questions I have no electronics background, so forgive me if this doesn't make sense...

All good! You're welcome!

Ozzuneoj wrote on 2025-01-23, 20:48:

but could a PCI Voodoo3 simply have a wire jumped from the (unused) 3.3v supply pin on the PCI slot to bypass that absurdly hot 3.3v regulator? Obviously it would impact compatibility with boards that weren't properly wired for 3.3v PCI cards, but it seems like it would help a lot to reduce the card's thermals. On the SGRAM models the PCB under that regulator is usually the first area to turn brown.

Well, if the voltage required by the graphics chip matches that of your actual power supply (which directly supplies the PCI slots afaik), then theoretically yes.
BUT, the power supply's voltage tolerance is relatively high (5% on the 3.3V rail) and so it is very likely to not fit and making the graphics card potentially unstable.
Also imagine the linear regulator beeing set to 3.3V and your power supply only giving 3.2V then the linear regulator would feed into that rail and additionally supply your whole system. You would at least have to cut the output from the linear regulator from the 3.3V.

Further on this would only work if there'd only be one voltage needed on the graphics card. This may hold true for older cards.
A Voodoo 3 already needs at least two voltages, Vcore (for the graphics chip itself) and Vio (for AGP, RAM). Vcore is between 2.2V and 2.76V depending on the particular chip / speed grade used (Voodoo 3 2000 / 3000 / 3500). Overall a simple patch wire wouldn't work here.
I just had a look at some Voodoo 3 cards and if you look carefully you'll find two regulators on the PCI versions and only one on the AGP versions.
I'd say the two on the PCI versions generate 2.20-2.76V for the core and the other 3.3V for the memory and I/O. Very likely both out of 5V from the PCI port.
On the AGP version there's only Vcore needed as the 3.3V already comes from the AGP port (which itself is regulated on the mainboard and not directly taken from the power supply).

Ozzuneoj wrote on 2025-01-23, 20:48:

In line with that, seeing all of these regulator designs is making me imagine an add-in card that efficiently creates the voltages that these old hot-running cards expect. If only there were a quick way to connect an external supply to a card without having to desolder a bunch of components. Maybe just a small "adapter" that solders in place of the regulator and connects to a universal voltage regulator card or something. Could be a pretty neat thing to increase longevity of these little space heaters. 😀

I think the cleanest solution would be a very small PCB with a switching regulator fitting the footprint and original output voltage of the linear regulator. You would still have to remove or by any means disable the linear regulator.
Since there surely are different linear regulator pinouts there'd also be different PCBs for switching regulators.