What are your criteria?

If you're looking for best performance, forget the i430NX and (slightly less drastically) the SiS501 and go for the i430FX.
If you're looking for good build quality, I'd go with Gigabyte or Asus over Elitegroup.
If you're looking for least fuss, avoid the PCI/I-P54TP4 as it has a Dallas RTC that almost certainly needs sorting.

Oh, and both SiS501 boards have the ubiquitous and buggy/crap CMD IDE controller, and the i430NX has no I/O controller at all.

Only decent reason to go for i430NX: 256MB RAM support, all of which cacheable.

Critera: Best over all performance with the largest cacheable range. Must work with K5-PR200. The presence of a Dallas doesn't matter, it will get converted. Onboard CMD IDE controllers don't matter, I use PCI-based host controllers for HDDs. PS/2 header nice, but there is one board that doesn't have it, but uses AWARD and a standard KBC-40, so it will get converted. How much can these SiS 501 bards cache with 512K memory?

Best overall performance: no question, the PCI/I-P54TP4 with i430FX. It's a year newer than the other two chipsets and it shows.
Largest cacheable range: the exact opposite, GA-586IP as i430NX can cache 256MB, where the i430FX can "only" do 64MB (note that in 1995 16MB was seen as extravagant...)

Not sure about cacheable range of 501, documentation on early SiS designs is notoriously sparse. My guess would be 64MB in WB, 128MB in WT, but can't confirm. I do have a 501 board (So4, not So5 - but that shouldn't matter for these purposes) at home so could theoretically test.

As for K5-PR200 - no idea. Given the P54TP4 is almost a year younger than the others I'd give you highest chance there, but you probably just need to test it yourself.

430NX can take 512MB. I assume all of it can be cached.

I'm hoping to get a good 430NX board for my dual gold top P100s. This would have been ridiculously expensive back when it was bleeding edge. The 430FX with a P120 or P133 was also a performance leader for a while, but it was soon displaced by faster things and became budget very quickly. Personally, I'd do two Socket 5 boards. One with a 430NX and a proper Pentium, and a 430FX board with a K5-200.

The PCI/I-P54SP4 and the SI54P-AIO both have the buggy-as-hell CMD640 IDE controller, so avoid using these.

I'd use the PCI/I-P54TP4, as the NX chipset is kind of slow.

Made a post about the PCI/I-P54TP4 some time ago: ASUS PCI/I-P54TP4 Socket 5 motherboard thread/review

If you manage to get the P54SP4 working with a K5, please let me know. I've tried mine with a PR100 and wouldn't POST.

P54TP4 is one of the most stable boards I have used, ever. The one I had came with Pipeline Burst Cache. From 1995 to 1998 I ran it with a P75, later overclocked to 100/120/133 with undervolted 5v fan. Still use it to this day for Dos. Very stable overclocking board.

The P54TP4 sounds like a real winner, however the pipeline burst option makes it feel more like a mid-path between sockets 5 and 7. Apparently, this same chipset was also used in socket 7 boards. Also, the missing option also has the feel that this was a cheaper variant of its fraternal brother with the PB option. Has anyone compared PB and async cache on this same board to see what the difference was?

I like now the NX board was a socket5-only board, but will have to see how much slower it is than the FX.

Nobody voted for the SiS 501/502/503. Is it that lousy?

I think the 501/502/503 was a better choice if you're going with Socket4.

The 430FX chipset was intel's first chipset designed to use pipelined burst cache. I can't remember if the pipelined cache option was available right at introduction, but it was definitely available shortly afterward. I think pipelined burst cache is okay on Socket5, but you should use a COAST module even if they're less dependable.

FX supports EDO as well. But I'm guessing not the NX and SiS 500 series. The SiS white paper suggests that it has a cacheable limit of 128 MB when using 512K, 1M, or 2M cache. Interesting that the cacheable range doesn't increase with increasing RAM. More cache just for greater 'hits' I guess.

Seems as if the 430FX boards which sport the COAST slot identify themselves as socket 7, although they do not necessarily have split voltage capability. Which begs the questions, what defines a socket 5 motherboard? I figured it was the split voltage alone, but perhaps not?

There are mountains of socket 7 boards which don't have split rail capability. It was an optional part of the specification, and most boards only started adopting it after the MMX chips came out. 430FX was a crossover chipset. Usually if the board only supports 1.5x and 2x multipliers, it's considered socket5. If it supports 2.5x and 3.0x then it's socket 7...and of course socket 7 boards will have bigger VRMs to handle the extra current required by the faster Pentiums.

The difference between Socket5 and Socket7 is minimal, kind of like with Socket2 and Socket3 on the 486 platform.

Didn't that NEC Proserva system have dual socket 5's, yet it uses a 3x multiplier to run the Pentium 200's?

I didn't have a close look at it, but didn't it have an HX chipset? HX is a proper socket7 platform, and it should have 3.0x out of the box. If the socket5 was used in that machine, they were probably just clearing out old parts.

feipoa wrote:

FX supports EDO as well. But I'm guessing not the NX and SiS 500 series.

SiS 501 supports EDO too (iirc it was the first Pentium chipset to do so), but its memory performance is clearly slower than the i430FX. Even with FP, both are a lot faster than the i430NX.

Yes, the Proserva is HX. Odd to use a socket 5 in there. It would be interested to see if some of the systems have "socket 5" and some have "socket 7" written on the ZIF.

I searched for "EDO" and "extended data" in the SiS 500 series manual and the search returned nothing. Manual was dated: January 9, 1995, Rev. 2.0.

feipoa wrote:

I like now the NX board was a socket5-only board, but will have to see how much slower it is than the FX.

NX was the first chipset for the P54 and while it has some nice features (fully cachable 512 MB of RAM with tag installed - FX can only do 64 MB; SMP support), it was just as slow as the older LX with limited PCI 2.0 support. It's also more or less Intel Pentium 75-133 only. It doesn't like non-Intel CPUs much.

If you want a historical build from the early Pentium days, it's a good choice. But otherwise, go for the FX board.

feipoa wrote:

Which begs the questions, what defines a socket 5 motherboard?

Well, the former is 320pin the latter is 321pin and that's what defines the socket 5 vs socket 7.

What makes things complicated is that some vendors mounted 321-pin Sockets 7 on boards with no split-voltage support (or with support through an external VRM that was not present and even not populated). And that includes not just crappy vendors, but also Intel who was considered as reference design maker, so this was widespread.

Given this I would personally not use "Socket 5 motherboard" definition, but rather talk about a motherboard with or without dual-voltage support.

My Socket 5 build would have the Gigabyte board 😀

There is some more rare 430HX socket 5 boards that support POD200 😐 These would of been high end, dead end boards. 😒 While not historically period correct it would of competed with early socket 7 233MX, and might of been faster.