Socket A has absolutely ridicilous number of different motherboards and probably dozen or so different chipsets. As the number of different motherboards is huge, some of them are inevitably duds, some are meh and some are great. Socket A lasted as a prime platform around 4 years or so. You can approach your build as targeting some period correct machine during those years, just a general early XP era build, old school voltage modding and OC stuff etc.

You don’t seem to have anything specific in mind so it might be best to go with later stuff from 2003-2004. I really don’t know where you got the informatin for nVidia chipsets, but nForce2 chipset is without a doubt the fastest Socket A platform and MCP-T offers terrific onboard sound (Soundstorm). Some of the best Socket A motherboards of all time had nForce2 Ultra chipsets. These high end boards usually also had good OC features. Dual channel is pretty much useless from the performance perspective, unless you use the onboard GF4MX graphics some boards did have. That is the main reason why nVidia added the dual channel and the benefit for for compute is perhaps few percent. But the chipset is fast and Soundstorm also is easy to CPU.

This is not to say that VIA chipset boards are outright bad options, there are many good boards that do the job just fine. Even the latest VIA chipsets aren’t just as fast as the nForce2, but for several years VIA was without a doubt the most popular chipset and many legendary boards had VIA chipsets.

My personal experience limits to AMD760, variety of ViA chipsets and nForce2. My ABIT KG7 RAID is absolutely terrific, but like I said, just by looking at your post it may be best to avoid the earlier boards and go with something newer, at least KT400.

Couple of things to note: many if not most socket A era boards suffer from bad caps, so be prepared to recap them. Most of the socket A boards use 5V for CPU VRM, so you need a PSU with good 5V capability. Pretty much the only exception to this are many nForce2 boards which have the 12V CPU VRM. These should work well with modern PSUs also.

gregorem wrote on 2025-09-11, 11:51:

Hi. I planint to make next vintage PC build, next to Socket 3, Socket 7, S370 and LGA 775. I gues asembling new rigs is my new h […]
Show full quote

Hi. I planint to make next vintage PC build, next to Socket 3, Socket 7, S370 and LGA 775. I gues asembling new rigs is my new hobby now. Probably even more than using them.

So Socket A. First fundamental question is: which chipset. I want to cover all available CPU options, so FSB from 100 (200 MT) to 200 (400 MT). We have few options:
VIA: KT600, KT880, KM400A. Are they all the same? I was read Wikipedia table, and, except extra IGP on KM400A, they all look the same. Does UniChrome causing lower AGP performance with AGP slot? Should I looked for them or avoiding some or all together?

SIS: 748 and 741 - the same question as with VIA. Mirage is probably even less useful than useless UniChrome, but does it causing any trouble or just be there and do nothing? How about quality of SIS in early XP era. SIS made quality chipsets for Tualatin, but I know nothing about SIS offer for AMD platforms.

AMD: 750 and 760. I seen post, where users praised they quality. Problem is, that they are early chipsets targeted towards server platform. So they supported max 266 MT CPU, do not offer USB 2.0 or AGP 8X and so on.
I have no interest in Athlon MP, so I probably pass it.

ALi: only MT 266 and older AGP standard (as AMD). Pass?

Nvidia: some love it, some hated it. nForce 400 Ultra support DDR dual channel! However very limited documentation make it bad choice for OC (confirmation needed) and they barely works with cpuspd (according to programmer).

ATI - know nothing.

I really want to get for some testingMobile Athlon XP, becouse they all had unlocked multipler, but for now I bought Barton 3200+ and it will be mine CPU for that setup. Do all boards and chipsets support Mobile XP? I prefer to use DDR and it will be nice to have USB 2.0.

Could you extend my knowledge about Socket A quirks, what works and what should avoiding? I really had mixed experience with nForce on LGA775, so I don't trust Nvidia chipsets. Are VIA and SIS any good?

SoA... it was a very long-lived platform which means big differences in performance (I recall a benchmark that showed an nForce2-Ultra performing 50% better than an AMD750 with the same Athlon B CPU), as well as compatibility. To make sense of it, look at it chronologically, not by brand.

Early - 100MHz (200MT/s) FSB:
AMD 750 Irongate - not aimed at servers, but more a proof-of-concept of an Athlon chipset, a holdover from Slot A. In terms of features very similar to i440BX for P3 - 3.3V AGP 1.0 (2x), 100MHz FSB and synchronous memory, although it did pair that with ATA-66. The only reason it was used for the first SoA boards was that Via had had major issues with their Slot A chipset vs the Thunderbird Athlon CPU, and all early SoA CPUs were Thunderbirds. Could be paired with both AMD and Via southbridges; Via southbridges offered more advanced features (ATA100), but had issues with SBLive cards. AMD was slower but less problematic. Only get AMD 750 if you want very early 100MHz FSB CPUs and want to pair with cards that don't like Via.
Via KT133 - the chipset SoA was supposed to go live with. AGP 4x and 133MHz asynch SDR-SDRAM support. Pretty unloved now as it offers nothing over the KT133A (which adds 133MHz FSB support), but if you wanted a mainstream Athlon or Duron in late 2000, this was what you got. KM133 was the version with integrated VGA.

Early - 133MHz FSB SDR:
Via KT133A - the chipset that let Athlon move to 133MHz FSB. Solid (apart from 686B SBLive issues), not sexy. Still probably the most common single chipset and a huge improvement on KT133 in every way. KM133A added integrated video.
SiS 730 - an integrated VGA SiS chipset, although it also allowed an external AGP card. Would have been interesting if used on better boards.

DDR arrives - 133MHz FSB:
AMD 760 - AMD again entered the chipset business to leverage DDR. Unlike with P3, this made a big difference on Athlon, as DDR SDRAM matched the double-pumped FSB and unleashed the full potential of the Athlon. Usually paired with Via southbridges.
ALi Magik 1 - how not to do DDR. Performed worse than KT133A despite theoretically having twice the memory bandwidth. Avoid unless you want oddballs.
Via KT266 - slightly disappointing DDR entry from Via. Not terrible, but not compelling either.
SiS 735 - total leftfield entry, SiS' first high-performance chipset in a few years, and in some metrics faster than AMD 760 despite being much cheaper. In particular, the ECS K7S5A paied this chipset with wafer-thin low-end PC Chips quality, but gave you an SoA board for EUR 25 at one point. That was also the *fastest* SoA board when it appeared. Sadly almost no high-quality boards using it, but the K7S5A sold like hot cakes. SiS 745 came slightly later but didn't add much.
Via KT266A - Via re-worked the KT266 into something competitive. Managed to beat AMD 760 and SiS 735 at last, if by a narrow margin. KM266 is version with integrated VGA.
nVidia nForce1 - hey, GPU vendor makes chipset. Some versions even offered dual-channel DDR. Interesting but not earth-shattering. Yet.
ATi A3 - other GPU vendor can't ignore challenge. Everyone else did ignoer them though.
SiS 746 - another little-seen SiS chipset. What makes it interesting? First chipset with AGP 8x.
AMD 760MP and MPX - dual-CPU chipsets for AthlonMP (although with a few pencil marks XPs can usually work too). MPX added 66MHz PCI. Interesting for dual-CPU boards. Be aware of issues with the USB controller in early boards, and the fact that early MP boards still drew power for both CPUs from 5V line, which means you need mammoth 5V lines up to 50A, which even modern high-end PSUs can't supply. If interested, choose a design with ATX12V connector.

FSB up to 166MHz:
Via KT333 - Via in the lead again. Same as KT266A, just a bit faster. This is the fastest chipset with (universal) AGP 4x slot, so if you want to run a fast SoA CPU with an old AGP 1.0 (2x) card like a Voodoo, this is one to watch.
SiS 746FX - pretty much the same, just with AGP 8x
Via KT400 - Via moving into deceptive name territory. You'd expect 400MT/s FSB support or at least 400MT/s RAM support. Nope. Very minor improvement over KT333, adding AGP 8x, which meant it sold better. KM400 adds VGA as usual.
Via KT400A - still no 200MHz FSB support, but at least lives up to its name in the RAM department. No prizes for guessing what KM400A brings to the party.
nVidia nForce2 - nVidia pulls it off - single fastest chipset so far, thanks to good dual-channel support.

Last gen chipsets with 200MHz FSB support:
Via KT600 - Via at last lives up to their previous name. Same as KT400A, now can officially run at 200MHz FSB.
SiS 748 - decent enough chipset, again rarely seen outside of ECS/PC Chips and Asrock. Not bad, but nothing better than KT400A either.
nForce2 400 - nVidia does a nasty: single-channel version of the nForce2 chipset. Not bad, but no better than the Via or SiS alternatives. Not the nForce2 you are looking for.
nForce2 Ultra 400 - the fastest of all SoA chipsets.
Via KT880 - rarely seen very late dual-channel DDR chipset from Via. Didn't manage to beat nForce Ultra 400 though.

So, which to choose? Totally depends on what you want. You already have faster and slower systems, so speed probably isn't even the most important consideration.

If you have a Barton with 200MHz FSB and want to test mobile Athlon XP, you want to go for one of the last-gen chipsets. In terms of performance it's simple: go for a board with nF2 Ultra 400 chipset. They also tend to be the most common. Avoid Asus A7N8X boards though - they are hopelessly picky in terms of which DIMMs they work with, yet are slower than boards from other vendors that eat anything within spec. Now, given you probably don't have a shop with a range of every nF2 board ever made on sale, there's not much point in recommending specific boards - so do it the other way round: look for what nF2 boards are available to you for acceptable price, and check whether they have BIOS support for mobile AthlonXP. Also look out for Geode NX support - the embedded version of the SoA CPU, shipped with even lower TDP than the mobile version so just as interesting to clock with.

So, you got the fastest CPU. My opinion - that goes best with "nForce2 Ultra 400". It has to be this specific variant, because anything without "400" does not support 200Mhz FSB and "400" without "ultra" is single channel only.

Benefits from dual channel are small on this platform because it can not really use all the bandwidths , but you do get a few %.

Then there are southbridges.

MCP2 is the most basic and most common variant.

MCP2-T adds SoundStorm, which is pretty much the best integrated sound you can get. Main caveat - still not as good as dedicated soundcard so if you are going to use that it becomes useless.

MCP2-S/R adds SATA and working one at that (unlike VIA), you can use modern devices with this including SSDs. Quite useful. Also 2 extra USB ports (up to 8 from 6) which often means 2 front panel headers - may be useful in modern cases).

MCP2-GB adds gigabit NIC in addition to SATA and USB. Can be useful (the fastest way to transfer data to/from this system), but also you can easily add PCI gigabit NIC if you wanted it.

VIA... IMO is a good choice if you want older system. Will work better with W98, some southbridges still support functionality needed for sound in 9x/dos, some chipsets like KT333 have universal AGP, etc. On newer system... they are just worse than nforce2. Drivers are a mess, even more so than stuff from nvidia. Performance is slightly worse. Sata, when available (VT8237/VT8237R) is poorly implemented and does not work with SATAII/SATAIII devices (only SATAI), meaning it is mostly useless.

I have very little experience with SiS and AMD chipsets on S462, so will not comment on those.

Also - caps are going to be dead, be ready to replace them.

And one very important thing is - choose a motherboard with 12V CPU VRM / extra "P4" power connector. Without that S462 is perhaps the hardest platform to power with modern PSU - way, way too much power required from 5V.

Correct me if Ivm wrong, but doesn't SoundStorm used Realtek 650 as soundcard and offered only hardware decoded Dolby and SPDiF? And doesn't VIA Vinyl offerd the same, so hardware decoded Dolby + SPDiF?

In old publication they praised SoundStorm, but Realtek 650 was avilable also with other chipsets. So whats the deal?

RTL650 was just a 'dumb' CODEC, which was paired with the Soundstorm DSP for DTS. Without the DSP, no DTS.

Archer57 wrote on 2025-09-11, 18:46:

[...]

MCP2-T adds SoundStorm, which is pretty much the best integrated sound you can get. Main caveat - still not as good as dedicated soundcard so if you are going to use that it becomes useless.

Note that driver support for SoundStorm is limited - no newer than Vista (applies to nForce2 as a whole) and no support in FOSS, Linux driver just talks to the RTL650 CODEC and ignores the APU.

[...]

I have very little experience with SiS and AMD chipsets on S462, so will not comment on those.

One more thing about SiS: the chipsets contain integrated SiS900 Ethernet. Decent enough but no fancy acceleration features - comparable to Realtek RTL8139 or Via Rhine III. One issue is specifically with the default drivers for SiS900 included in Windows XP: due to a bug, the wrong address is read for the MAC of the adapter. That means that instead of a unique MAC for every adapter, every SiS900 running with default WinXP drrivers shows the same MAC address. No problem if you only have one on a network, but if you have two or more, addressing at L2 wil collapse. Solution is simple: install SiS' own (updated, fixed) drivers instead of relying on the ones included in WinXP.

gregorem wrote on 2025-09-11, 19:49:

Correct me if Ivm wrong, but doesn't SoundStorm used Realtek 650 as soundcard and offered only hardware decoded Dolby and SPDiF? And doesn't VIA Vinyl offerd the same, so hardware decoded Dolby + SPDiF?

In old publication they praised SoundStorm, but Realtek 650 was avilable also with other chipsets. So whats the deal?

It offers sound processing on its dedicated processor built into SB, offloading it from CPU. Improves performance and offers extra effects. Kind of the same stuff creative cards do, for example.

They use realtek chip (or in some cases cmedia chip) for physical output - digital to analog conversion, amplification, etc.

dionb wrote on 2025-09-11, 20:01:

Note that driver support for SoundStorm is limited - no newer than Vista (applies to nForce2 as a whole) and no support in FOSS, Linux driver just talks to the RTL650 CODEC and ignores the APU.

Yep, that is a thing also, but... practically IMO XP is going to be the OS of choice for late S462 anyway and drivers for XP work. Modern linux, sadly, is becoming less and less viable without x86-64 support, so probably not a huge factor here.

dionb wrote on 2025-09-11, 20:01:

One more thing about SiS: the chipsets contain integrated SiS900 Ethernet. Decent enough but no fancy acceleration features - comparable to Realtek RTL8139 or Via Rhine III. One issue is specifically with the default drivers for SiS900 included in Windows XP: due to a bug, the wrong address is read for the MAC of the adapter. That means that instead of a unique MAC for every adapter, every SiS900 running with default WinXP drrivers shows the same MAC address. No problem if you only have one on a network, but if you have two or more, addressing at L2 wil collapse. Solution is simple: install SiS' own (updated, fixed) drivers instead of relying on the ones included in WinXP.

That's a nasty bug. Good to know.

Practically though... i tend to not use network with old systems at all, so most of the time it is disabled. And if i wanted to i'd very likely use PCI gigabit NIC for extra speed, in case when integrated one is 100Mb. This are so common they are essentially free and motherboards tend to have way more PCI than i'll ever need. So NIC and its specs are not hugely important.

Also probably worth mentioning for the OP - nforce2 memory compatibility can be a pain - some memory just will not work properly at all, some refuses to work in dual channel or at 200Mhz. I bought whole box of DDR1 (thankfully it is cheap) to be able to pick a few sticks which work well. VIA is much better in this regard...

Archer57 wrote on 2025-09-11, 21:09:

[...]

Also probably worth mentioning for the OP - nforce2 memory compatibility can be a pain - some memory just will not work properly at all, some refuses to work in dual channel or at 200Mhz. I bought whole box of DDR1 (thankfully it is cheap) to be able to pick a few sticks which work well. VIA is much better in this regard...

It differs a lot per board. Asus A7N8X was downright terrible in this regard, others far less picky.

Anecdote: back in 2003, I was a poor student, subletting half my pad to a richer student to make ends meet. Both nerds. He bought the good stuff, Athlon XP 3200+, A7N8X-Deluxe, some new Gf4Ti 4200 from Asus, high-end Kingston memory etc. I went to computer fairs, bought boxes of 'dead' components and tried to resurrect it. At least half the motherboards in those boxes had just been mis-flashed. I was in luck at the time - that proved true of a Gigabyte GA-7N400L, basically she same as that Asus. I added AXP2500+ and (successfully) overclocked it to 3200+ speeds - and bought the cheapest in-spec PC3200 I could find (Nanya iirc). I also lucked out with a Gf4Ti 4200 I found without heatsink. Added one and it worked perfectly - giving me essentially the same system as my housemate for less than 1/3 of the price. My Frankenstein-rig worked perfectly. His only managed to boot occasionally and wasn't stable enough to install an OS. Memtest showed memory issues, so he got some other DIMMs. Same issues. He was about to RMA his motherboard when I referred to its QVL and saw that neither of his DIMMs were on it. Mine were. I tried his DIMMs: both sets worked perfectly on my board. And mine actually worked on his. Halleiluiah! Oh, and my system benched a few FPS faster as well, so it's not as if the A7N8X gained extra performance for that selectivity.

TLDR: GA-7N400L ate all the RAM we threw at it, A7N8X failed with everything off QVL. I call this an Asus problem more than an nForce2 problem. Asus was bulletproof in the late 1990s to 2001 or so, after that they sort of lost the plot...

You bought me on MCP2-T. I like to try it.

But Asus is picki on RAM modules and slow, so we skiped that barand (their boards are also not cheap in my neighborhood - looks like people just love to overpaid for Asus sticker, no matter what crap they selling). I was read some horrific description of DFI self-corupting CMS. Are MSI ang Gigabyte reputable brands for late Socket A? Or AOpen?

One issue I ran into with nForce chipsets is lack of DMA support on PCI slots. This meant that I couldn't get DOS sound blaster compatibility under Windows 98.

Switched to a board with a Via KT600 chipset and this was no longer an issue.

I was very happy with my Gigabyte board, it was rock solid despite mistreatment. MSI kept using bad caps longer than most, but as Archer57 says, assume any caps are dead anyway. Same forAOpen. And Abit for that matter. But again: it's pretty pointless obsessing about 'the best' board, rather look at what you can acquire now and see what of that might be suitable.

dionb wrote on 2025-09-11, 21:20:

It differs a lot per board. Asus A7N8X was downright terrible in this regard, others far less picky.

Anecdote: back in 2003, I was a poor student, subletting half my pad to a richer student to make ends meet. Both nerds. He bought the good stuff, Athlon XP 3200+, A7N8X-Deluxe, some new Gf4Ti 4200 from Asus, high-end Kingston memory etc. I went to computer fairs, bought boxes of 'dead' components and tried to resurrect it. At least half the motherboards in those boxes had just been mis-flashed. I was in luck at the time - that proved true of a Gigabyte GA-7N400L, basically she same as that Asus. I added AXP2500+ and (successfully) overclocked it to 3200+ speeds - and bought the cheapest in-spec PC3200 I could find (Nanya iirc). I also lucked out with a Gf4Ti 4200 I found without heatsink. Added one and it worked perfectly - giving me essentially the same system as my housemate for less than 1/3 of the price. My Frankenstein-rig worked perfectly. His only managed to boot occasionally and wasn't stable enough to install an OS. Memtest showed memory issues, so he got some other DIMMs. Same issues. He was about to RMA his motherboard when I referred to its QVL and saw that neither of his DIMMs were on it. Mine were. I tried his DIMMs: both sets worked perfectly on my board. And mine actually worked on his. Halleiluiah! Oh, and my system benched a few FPS faster as well, so it's not as if the A7N8X gained extra performance for that selectivity.

TLDR: GA-7N400L ate all the RAM we threw at it, A7N8X failed with everything off QVL. I call this an Asus problem more than an nForce2 problem. Asus was bulletproof in the late 1990s to 2001 or so, after that they sort of lost the plot...

Nice when it works out like that 😀

Though this may be somewhat related to RAM itself - for a long time i've considered all the fancy overclocked modules... borderline scam. They use the same chips rated for lower frequencies, they just have profiles in eeprom with OC including higher voltage. That's true for XMP and alternatives in modern systems too. So simple OEM RAM, like those nanya modules would have been, is often a good choice. Apart from being cheaper tends to have better compatibility and uses the same chips - so can always be overclocked the same way if needed.

So fancy RAM causing issues when simple OEM stuff works just fine would not surprise me at all. In fact it is quite common, even nowadays. Overclocking, even "factory overclocking", always compromises stability.

In terms of motherboards - i currently have GA-7N400S-L as well as a few epox boards - EP-8RDA, EP-8RDA3I, EP-8RDA3+ PRO. They are all moderately picky in terms of memory. Have to go through a bunch of sticks and a bunch of time running memtest to find a set which works with decent settings in dual channel. Meanwhile on VIA board i can just plug in a random set of mismatched sticks, configure everything manually and it'll work. So i suspect that while boards themselves likely play a role - chipset does too. Be it memory controller itself, reference board designs they provided, whatever.

All that said nowadays the ram is cheap and easy to come by, unlike nice boards with required set of features like 12V VRM, specific SB, etc. So it makes sense to just get a bunch of memory and pick what works instead of trying to find a board with better compatibility. IMO.

I got EP-8RDA3+ PRO to replace EP-8RDA3I in my build and it does what i wanted it to - MCP2-S, which means functional SATA so i can get rid of IDE-SATA adapters and more USB, which my case needs. But... i already worked around those issues (IDE-SATA works and i got one of those "USB header splitters" which are essentially USB hubs) and decided not to change a working system for now. Overall i'd call all this epox boards good - good set of bios options, monitoring, OC etc and no obvious issues from what i've seen.

My Epox 8KRA2+ (KT600) has worked well for me, but: I of course had to replace the caps, it took a couple of power supplies before finding one that could hold the +5 volt rail, and though it has SATA ports, they don't recognize SSDs so I use one of the PATA ports with an adapter. My trusty old Epox 8RDA3+ was a better board, but I've yet to stumble across a decent deal on an Nforce 2 Socket A board.

I don't know how much (if any) experience you have with Socket A CPUs - they are very unforgiving when it comes to getting the heatsink mounted correctly, and good Socket A coolers are not easy to find these days, so keep that in mind.

Repo Man11 wrote on 2025-09-12, 00:05:

My Epox 8KRA2+ (KT600) has worked well for me, but: I of course had to replace the caps, it took a couple of power supplies before finding one that could hold the +5 volt rail, and though it has SATA ports, they don't recognize SSDs so I use one of the PATA ports with an adapter. My trusty old Epox 8RDA3+ was a better board, but I've yet to stumble across a decent deal on an Nforce 2 Socket A board.

I don't know how much (if any) experience you have with Socket A CPUs - they are very unforgiving when it comes to getting the heatsink mounted correctly, and good Socket A coolers are not easy to find these days, so keep that in mind.

SocketA was definitely the high point for EpoX and many of their models were popular among enthusiasts. They really struck gold with the 8kha(+) and released many terrific sA boards after that. Back in the day, during the around three years I had sA, I had half a dozen MBs and two of them were EpoX, both were extremely good. I would have propably continued to use the last one, 8rda3+ longer, unless I had destroyed it during my OC shenanigans. I went to Abit AN7 from there. It was terrible, retruned it to a store after few days and got Abit NF7-S V2.0 instead, which was in use till I got A64. That NF7-S was amazing, objectively probably the best board ever made for sA. Or at least one of the best.

Currently I have 8rga+ in my mid 2003 sA watercooled system (with my own original sA WC gear) and it is a great board. I have the same board as yours too, polymodded. I have only used it on a bench, but it is definitely a good KT600 MB. I used it to run some GPU benchmarks and general fiddling and it had zero issues. I also have 8K9AI, but I haven’t even tested or recapped it.

If you want a Rock Solid stable Sys then go for KT600. There is nothing wrong with it. Never had issues with them.

On the other Hand I tried minimal OC on DDR (420) on a nforce2 400 Ultra chipset and it Fried instantly. They May be faster than via but also very fizzy. In my opinion they are not worth the little Performance gain. Go for a Kt600, they are fast enough.

AncapDude wrote on 2025-09-12, 05:37:

If you want a Rock Solid stable Sys then go for KT600. There is nothing wrong with it. Never had issues with them.

There are dud boards for every chipset. Manufacturers absolutely spammed sA boards, many releasing several different board designs for same chipset. Some were buggy, some had bad bioses and never got updates as they were in the market for very short time.

I had an awful experience with Abit AN7 (nF2 Ultra) back in the day. I just couldn’t get it stable stock with more than one memory stick. It may be that it was due to early bios, because I got the board immediately when it was launched and had high hopes for it. I returned the mobo and went to NF7-S 2.0.

On the other Hand I tried minimal OC on DDR (420) on a nforce2 400 Ultra chipset and it Fried instantly. They May be faster than via but also very fizzy. In my opinion they are not worth the little Performance gain. Go for a Kt600, they are fast enough.

I’ve never heard this being a common issue and sounds like there was either ridicilous voltage for the chipset or regulators failed and voltages were out of spec. Or if the board was used, it may have been heavily OCd, memory controller was degraded and you just pushed it over the final edge. Still, many of the best sA OC boards were nF2 and I used my NF7-S v2 @DDR450 24/7/365 and after getting A64 parts, I sold it to a dude in my work place who used the system who knows how long after that.

But KT600 or many other sA VIA chipsets aren’t a bad choice. But as I said, there were many atrociously bad motherboards made for the platform, so it isn’t and can’t be a general recommendation.

As far as the performance comparisons go, this thread has huge number of benchmarks and comparison between different sA platforms for those who are curious:

Socket A: Nvidia vs Via - battle of the platforms!

Like I said in the previous post, one big advantage for modern day that many nF2 boards have is the 12V CPU VRM. Outright I can’t name a single VIA chipset board which would have that, probably because VIA reference designs and specs didn’t have this, so it is at least very rare feature. Having 12V CPU VRM pretty much removes the PSU problems off from the equation and you can easily use pretty much any decent modern PSU that is designed mainly for 12V loads, even if you have 5V heavy graphics card and hig end power hungry CPU.

I worked on fixing issues and adding new features for VIA UniChrome based graphics on Linux (OpenChrome Project) some years ago, so I can give a little insight into what UniChrome integrated graphics really is like.
There really is a chipset called KM400 prior to the release of KM400A.
The difference is that KM400 chipset apparently has a bug of the internal graphics IRQ line, but KM400A chipset fixes the issue.
The IRQ is apparently used for a page (screen) flipping, but due to the bug, KM400 chipset needs to poll certain registers for a page flip.
An interrupt based page flip apparently works fine on KM400A chipset.
Another difference between KM400 and KM400A is that 400 MT / s (200 MHz X 2 transfer edges = 400 MT / s) is supported by KM400A chipset.
KM400 chipset is up to 333 MT / s support.
They both appear to support DDR400, apparently.
The UniChrome integrated graphics is automatically disabled if you use an AGP graphics card.

Generally speaking, all AGP era VIA integrated graphics implementations disable the integrated graphics if an external graphics is used.
I believe VIA uses an AGP pin status (forgot which one) to determine the status of the integrated graphics when it comes out of hardware reset.
If this pin is shorted to ground, an AGP graphics card is present, and the chipset automatically disables the integrated graphics.
If an external AGP port is disabled, then AGP based pins can be used for driving FP (Flat Panel) signals.
It is possible that AGP based signals getting repurposed as FP signals is only available on mobile version of the chipset (i.e., KN400(A)).

KM400(A) is a second release of UniChrome graphics.
The first release was with CLE266 chipset for P6 bus based platform mainly used with VIA C3 processor.
Compared to CLE266 chipset's graphics, KM400(A) removes the IDCT (Inverse Discrete Cosine Transformation) engine present in CLE266 chipset's integrated graphics, probably to save some die area.
Internally, VIA reorganized many hardware registers, and KM400(A)'s register organization is much more like the later on UniChrome integrated graphics like P4M800 Pro, K8M800(A), and PM800 / CN400.

Contrary to some popular beliefs, UniChrome and its descendent Chrome9 integrated graphics *WERE NOT* S3 Savage4 derivatives.
This appears to be a marketing ploy by VIA marketing to associate itself with the S3 name (S3 Inc. was the original firm, and S3 Graphics was the fully owned subsidiary of VIA in Fremont, CA.) since some people still remembered S3 among the PC users.
I once spoke with an engineer from VIA, and he told me that they bought some design IPs from a firm I do not remember who it was, but certainly, they did not start from designing a VGA compatible graphics core.
That being said, S3 derived graphics and VIA UniChrome graphics have totally different hardware register design, and this pretty much proves my point.
As for what happened to S3 Savage4 based designs, S3 Graphics continued to develop it further until around 2011, and then VIA transferred the design to Zhaoxin Semiconductor (Shanghai municipal government owned fabless semiconductor company) around 2013.
Today, S3 graphics technology still lives, but mainly in China as KX-6000 and KX-7000 series integrated graphics.

gregorem wrote on 2025-09-11, 11:51:

. . .
VIA: KT600, KT880, KM400A. Are they all the same? I was read Wikipedia table, and, except extra IGP on KM400A, they all look the same. Does UniChrome causing lower AGP performance with AGP slot? Should I looked for them or avoiding some or all together?

. . .

AncapDude wrote on 2025-09-12, 05:37:

If you want a Rock Solid stable Sys then go for KT600. There is nothing wrong with it. Never had issues with them.

On the other Hand I tried minimal OC on DDR (420) on a nforce2 400 Ultra chipset and it Fried instantly. They May be faster than via but also very fizzy. In my opinion they are not worth the little Performance gain. Go for a Kt600, they are fast enough.

I disagree. Once working nforce2 is fine. The only real issue i've encountered is that it's picky in terms of memory. Once working it is stable and reliable. If anything it tends to be less buggy in terms of drivers as long as it is used with XP. For 98 older VIA chipsets may work better.

Also not sure what could have died because of OC, most obvious thought for me would be bios getting corrupted, this could happen on some boards and is not related to chipset but to the board itself.

CharlieFoxtrot wrote on 2025-09-12, 07:54:

Like I said in the previous post, one big advantage for modern day that many nF2 boards have is the 12V CPU VRM. Outright I can’t name a single VIA chipset board which would have that, probably because VIA reference designs and specs didn’t have this, so it is at least very rare feature. Having 12V CPU VRM pretty much removes the PSU problems off from the equation and you can easily use pretty much any decent modern PSU that is designed mainly for 12V loads, even if you have 5V heavy graphics card and hig end power hungry CPU.

Yep, IMO building S462 system nowadays 12V VRM is one of the most important things to look for in a motherboard. Not having to deal with vintage power supplies is a huge advantage.

Similarly working SATA on MCP2-S/R/GB simplifies things with storage - small and cheap modern SSD can be used with no adapters, while on VIA chipsets only SATA1 devices work which means vintage storage or adapters.

Archer57 wrote on Yesterday, 00:51:

Yep, IMO building S462 system nowadays 12V VRM is one of the most important things to look for in a motherboard. Not having to deal with vintage power supplies is a huge advantage.

Similarly working SATA on MCP2-S/R/GB simplifies things with storage - small and cheap modern SSD can be used with no adapters, while on VIA chipsets only SATA1 devices work which means vintage storage or adapters.

This is of course very personal preference. I have recapped some older 5V heavy PSUs for the very reason that I can use all kinds of HW combinarions with them. But everyone doesn’t want to or can’t do that so nForce2 has a big advantage and that is the reason why I generally recommend it. Old PSUs you would be using with these systems are from worst capacitor plague era and should not be trusted.

You are right about the SATA too. While it is not IMO that important as the PSU issue, VIA and some other early SATA controllers won’t work with newer stuff.

Adding a few points about the AMD Socket A platform.
Some are my personal opinions and experiences.

AMD:
Actually, AMD-75o chipset supported both Slot A and Socket A.

https://theretroweb.com/motherboards/s/gigaby … a-7ixe4-rev-1-0

Initial revision of AMD-750 chipset's USB (inside AMD-756 SB) was buggy, and many initial mainboards shipped with a PCI USB card.
This probably only affects Slot A platform since fixed version was released later, so it is highly unlikely to have a buggy version populated on a Socket A mainboard.
Some vendors matched VIA VT82C686A "Super South" with AMD-751 NB.

VIA:
It was VIA KX133 chip that had an issue supporting Socket A platform (Thunderbird), and perhaps due to this embarrassing "bug", VIA quietly discontinued the chipset.
Here is a technical explanation from EPoX about the debacle.

https://web.archive.org/web/20010618053551/ht … sp?Article=1268

Initially, they were planning to call the updated (bug fixed) chipset that supports Socket A to be called KZ133, but someone knowledgeable about Nazi past must have intervened and got the name changed to KT133 before the release.

https://www.cnet.com/tech/tech-industry/via-r … olocaust-gaffe/

KZ is an abbreviation that means "concentration camp" in German (konzentrationslager in German language).

NVIDIA:

nForce's SB based PCI performance is said to be pretty low.
While it might not be a chipset issue, I have encountered sudden death issue of ASUS A7N8X-E DELUXE.
I still have the 2 boards that died.
I remember that at least one of them worked, but a few years later, it stopped working suddenly.
I still have the mainboard, and I tried to repair it, but it did not work.
I will like to repair it someday.
I always felt like NVIDIA is not that great of a chipset vendor.
The spec might be high-end oriented, but appears to ship with silicon bugs all the time like with nForce4's Ethernet (this one is for AMD HyperTransport based platform).
I personally never had good experiences with NVIDIA chipset in general.

ATI:

The rumor is that USB is pretty buggy that they had to spun many revisions of their SBs.
Some boards utilize ALi / ULi SB like M1535(D)+ instead of their IXP SB.