Are you implying that Socket A is picky with RAM and voltages or that it isn't?

I'd say SoA in general is very accepting, with the exception of later Asus boards which are utterly inexplicably picky even when boards from other vendors with the same chipsets have no trouble.

For universal AGP you're looking at late P3, early to mid SoA and early P4.

But If you want to test RAM too, it would make sense to choose a board with SDR and DDR SDRAM, and if you want to do anything PCI-related it might be better to avoid Via chipsets. To me that seems to point towards SiS chipsets.

By far the easiest to find with those specs would be the ECS K7S5A. Its PCB is wafer-thin and early BIOS releases were appallingly buggy, but ECS fixed the latter and so long as the caps are good (expect to have to re-cap if it hasn't already been done) it's stable. It was also the fastest single SoA board in most benchmarks at time of release (and the fastest overall in anything not hopelessly skewed to P4).

If you haven't seen it, this thread lists boards and also has some confirmed as working: Socket 423/478 Motherboards with Universal AGP Slot

dionb wrote on 2026-06-21, 21:20:

Are you implying that Socket A is picky with RAM and voltages or that it isn't? […]
Show full quote

Are you implying that Socket A is picky with RAM and voltages or that it isn't?

I'd say SoA in general is very accepting, with the exception of later Asus boards which are utterly inexplicably picky even when boards from other vendors with the same chipsets have no trouble.

For universal AGP you're looking at late P3, early to mid SoA and early P4.

But If you want to test RAM too, it would make sense to choose a board with SDR and DDR SDRAM, and if you want to do anything PCI-related it might be better to avoid Via chipsets. To me that seems to point towards SiS chipsets.

By far the easiest to find with those specs would be the ECS K7S5A. Its PCB is wafer-thin and early BIOS releases were appallingly buggy, but ECS fixed the latter and so long as the caps are good (expect to have to re-cap if it hasn't already been done) it's stable. It was also the fastest single SoA board in most benchmarks at time of release (and the fastest overall in anything not hopelessly skewed to P4).

I loved the k7s5a, but is it true that I would need a behemoth psu when it comes to legacy voltages?

Repo Man11 wrote on 2026-06-21, 22:39:

If you haven't seen it, this thread lists boards and also has some confirmed as working: Socket 423/478 Motherboards with Universal AGP Slot

I own a 645E Max-U for that very thread.
That bastard gives continuous I/o errors with ide channels if AGP is populated….

There are also these things: https://theretroweb.com/motherboards/s/asus-a7a266-rev.-1.10
This one has Universal AGP Pro plus both SD-RAM and DDR1 sockets.
I have one of these but I have to admit that I have very little experience with it, so I don't know how reliable it is.
(And yes, as most SoA boards it also needs a PSU with a strong 5V rail)

To be honest, I don’t care about testing ram.
I need a reliable motherboard with good AGP implementation (universal slot) that won’t me make doubt about psu voltages or caps about to fail or whatsoever.
My primary goal is to test agp cards.

I have MS-6309 for that. It has the typical Nvidia drivers falling back AGP to 2x speeds, but using Rivatuner, one can set it back to 4x no trouble at all.

mmx_91 wrote on 2026-06-22, 18:32:

I have MS-6309 for that. It has the typical Nvidia drivers falling back AGP to 2x speeds, but using Rivatuner, one can set it back to 4x no trouble at all.

Nvidia?
It is a Via Apollo 133A board.
I do not like Apollo 133a anyway.

I have an Amptron M930 LMR - it is one of the boards listed in the thread, and it has been a reliable board for me (after I recapped it of course). But I can't vouch for it working with 3.3 volt AGP cards because I don't have one to test it with. I had a look through my stash the other day and even the ATI 7000 AGP I have is a 1.5 volt card.

Mamba wrote on 2026-06-22, 04:33:

[...]

I loved the k7s5a, but is it true that I would need a behemoth psu when it comes to legacy voltages?

What CPU do you intend to install- just do the math: the hottest CPU you could install on it is the Athlon 1400C. It draws 72W from the 5A line. That means 14.4A. A PSU with a 20A 5V line - like any modern mainstream PSU - will be fine. And that's with the hottest CPU that would run on it. Anything else - including newer, faster CPUs - will draw less. For testing purposes I'd probably pair it with a 133MHz FSB Thoroughbred, say the AXP1700+, which draws 49.5W i.e. just under 10A on the 5V line, or failing that a 1000C (54W, 11A)

dionb wrote on 2026-06-21, 21:20:

Are you implying that Socket A is picky with RAM and voltages or that it isn't? […]
Show full quote

Are you implying that Socket A is picky with RAM and voltages or that it isn't?

I'd say SoA in general is very accepting, with the exception of later Asus boards which are utterly inexplicably picky even when boards from other vendors with the same chipsets have no trouble.

For universal AGP you're looking at late P3, early to mid SoA and early P4.

But If you want to test RAM too, it would make sense to choose a board with SDR and DDR SDRAM, and if you want to do anything PCI-related it might be better to avoid Via chipsets. To me that seems to point towards SiS chipsets.

By far the easiest to find with those specs would be the ECS K7S5A. Its PCB is wafer-thin and early BIOS releases were appallingly buggy, but ECS fixed the latter and so long as the caps are good (expect to have to re-cap if it hasn't already been done) it's stable. It was also the fastest single SoA board in most benchmarks at time of release (and the fastest overall in anything not hopelessly skewed to P4).

I mean avoiding Nforce based boards for a socket A test bench is a good idea, they have a lot of issues in general not just with memory either.

If I was going to pick a Socket A for testing it would be a rock solid KT266 board if I didnt need ISA or a KT133 with ISA, both would work well as a test bench without dragging in the compatibility issues the later Socket A chipsets had due to VRM changes, removal of ISA and Floppy controller and just general instability with AGP 8x early revisions and chipsets, same applies to DDR support better to just stick with SDR for testing purposes.

This isn't to say you couldn't use a later board for a test bench but you would be adding in extra compatibility issues you don't really need. (Not to mention VIA drivers for later chipsets were ...mid at best and at worst downright painful to bug stomp)

Trashbytes wrote on 2026-06-23, 10:13:

[...] […]
Show full quote

[...]

I mean avoiding Nforce based boards for a socket A test bench is a good idea, they have a lot of issues in general not just with memory either.

If I was going to pick a Socket A for testing it would be a rock solid KT266 board if I didnt need ISA or a KT133 with ISA, both would work well as a test bench without dragging in the compatibility issues the later Socket A chipsets had due to VRM changes, removal of ISA and Floppy controller and just general instability with AGP 8x early revisions and chipsets, same applies to DDR support better to just stick with SDR for testing purposes.

This isn't to say you couldn't use a later board for a test bench but you would be adding in extra compatibility issues you don't really need. (Not to mention VIA drivers for later chipsets were ...mid at best and at worst downright painful to bug stomp)

nForce boards have AGP 3.0 slots anyway, so out of consideration for that reason.

I disagree on the memory issues, that really was an Asus quirk, nor a broader nForce2 issue. A7N8X boards were great if you had exactly the right RAM, but boards from other vendors with the same chipset were absolute sluts that swallowed whatever DIMM you gave them. My GA-7N400-L was a prime example. I ended up swapping DIMMs with my flatmate whose A7N8X Ultra only wanted to run stable with my DIMMs. My board was happy with both sets and several others during testing stuff that didn't work on his board.

But again, wrong AGP revision. Via chipsets are great for RAM and AGP compatibility but PCI is an issue, with earlier southbridges (686B...) with issues with SBLive but even later ones having poor performance. Still, if PCI isn't critical they would be a fine option, although I'd then sooner go for P3 with ApolloPro133A (694X) due to lower power requirements and less chances of a core crunch when grappling with a big AGP card.

dionb wrote on 2026-06-23, 12:12:

nForce boards have AGP 3.0 slots anyway, so out of consideration for that reason. […]
Show full quote

Trashbytes wrote on 2026-06-23, 10:13:

[...] […]
Show full quote

[...]

I mean avoiding Nforce based boards for a socket A test bench is a good idea, they have a lot of issues in general not just with memory either.

If I was going to pick a Socket A for testing it would be a rock solid KT266 board if I didnt need ISA or a KT133 with ISA, both would work well as a test bench without dragging in the compatibility issues the later Socket A chipsets had due to VRM changes, removal of ISA and Floppy controller and just general instability with AGP 8x early revisions and chipsets, same applies to DDR support better to just stick with SDR for testing purposes.

This isn't to say you couldn't use a later board for a test bench but you would be adding in extra compatibility issues you don't really need. (Not to mention VIA drivers for later chipsets were ...mid at best and at worst downright painful to bug stomp)

nForce boards have AGP 3.0 slots anyway, so out of consideration for that reason.

I disagree on the memory issues, that really was an Asus quirk, nor a broader nForce2 issue. A7N8X boards were great if you had exactly the right RAM, but boards from other vendors with the same chipset were absolute sluts that swallowed whatever DIMM you gave them. My GA-7N400-L was a prime example. I ended up swapping DIMMs with my flatmate whose A7N8X Ultra only wanted to run stable with my DIMMs. My board was happy with both sets and several others during testing stuff that didn't work on his board.

But again, wrong AGP revision. Via chipsets are great for RAM and AGP compatibility but PCI is an issue, with earlier southbridges (686B...) with issues with SBLive but even later ones having poor performance. Still, if PCI isn't critical they would be a fine option, although I'd then sooner go for P3 with ApolloPro133A (694X) due to lower power requirements and less chances of a core crunch when grappling with a big AGP card.

Yup I agree with going for P3, OP was interested in Socket A .. which would make KT133 a good starting choice .. same for P3 one of the 133A or 133T chipsets with SDR and Universal AGP or AGP 4X, a great spot for a test bench. Universal AGP would be a sweet spot for a test bench actually, wouldn't matter what non 8X AGP card you put in it, it would likely work. (Some of the really early 2x cards might not .. like the Fury MAXX which only posts on certain BX boards, I know because I have a working one and it refuses anything but a BX chipset and even then it only posts in two of the 6 boards I own. Haven't tried it in Super 7 yet, perhaps I should)

Could even go for a 815E board too.

The best, most stable, compatible, and also very fast board that I ever tested (with universal AGP slot) is the Shuttle Spacewalker AK32A. Bonus: it comes with both SDR and DDR slots and works great with both memory types.

Also, I can confirm that it does not have any issues with PCI sound cards such as the Creative SB Live 5.1, Audigy 2 ZS, Aureal Vortex 2, Yamaha YMF724F-V (the latter also works exceptionally well in DOS with this board, and most games run without issues with SB Pro 2 compatibility using DSDMA).

Thank you for all the advices.
Will try to find one, hope already recapped.