Hmm that is actually an interesting question, I don't have an exact answer for that.

The two things that I can add, in regards of the Pentium(There are other 586 chips, but I will ignore those that fit into 486 motherboards as they will have a 32-bit Bus interface by design):
I think the feature should be more widespread then one would assume at first because the Pentium processor was most often times paired with a PCI Bus and PCI Bus has a 32-bit data path, so the chipsets should already have circuitry to send 64-bit data in two 32-bit chunks.
Also the performance decrease that will happen will depend on the processes running on the processor, the complete system performance depends on a lot of factors, so depending on the process the Pentium runs on, for example working with a small dataset, the halving of the memory databus throughput wouldn't matter that much, also for processes that mostly interact with I/O or Storage HW. For applications that primarily work with a lot of data, yes the performance penalty would be very real.

VIA Apollo VP (VT82C580VP) works well with a single SIMM installed.
This reduces memory speed, but it works.

Here are my benchmark results on the FIC VA-502:

1Slot 1   | Slot 2   | Slot 3   | Slot 4  | MemSpeed
2---------|----------|----------|---------|----------
34 MB FPM |   -      | -        | -       | 11 MB/s
48 MB EDO |   -      | -        | -       | 13.2 MB/s
5         |          |          |         |
68 MB EDO |   -      | 4 MB FPM | -       | 11.4 MB/s
78 MB EDO |   -      | 8 MB EDO | -       | 13.4 MB/s
8         |          |          |         |
98 MB EDO | 8 MB EDO | -        | -       | 18.2 MB/s
108 MB EDO | 4 MB FPM | -        | -       | 16 MB/s
11         |          |          |         |
128 MB EDO | 8 MB EDO | 4 MB FPM | -       | 14.9 MB/s
138 MB EDO | 4 MB FPM | 8 MB EDO | -       | 14.9 MB/s

The best results will be achieved if Slot 1/2 or Slot 3/4 is completely populated. If only one module is used, the controller operates in 32-bit mode.

There were several threads in the distant past on these motherboards both here and on vcfed

Pushing a 486 beyond 200MHz - achievable?

VLB and Pentium... How?

486 Speed on Pentium Base - Meet OPTI - Galaxy V Pentium Mainboard

rmay635703 wrote on 2026-06-01, 14:56:

There were several threads in the distant past on these motherboards both here and on vcfed Pushing a 486 beyond 200MHz - achiev […]
Show full quote

There were several threads in the distant past on these motherboards both here and on vcfed
Pushing a 486 beyond 200MHz - achievable?
VLB and Pentium... How?
486 Speed on Pentium Base - Meet OPTI - Galaxy V Pentium Mainboard

thanks, i checked up the threads but most are on the rare opti 586 chipsets which i am not expecting to find.
now i can only confirm that some intel and via chipsets work, is there any info on sis ones?
and how did sis perform against via in the 586 territory?

I have an AOpen AP57 SiS 5571 Pentium board that works with a single SIMM installed. Obviously, performance suffers with a single SIMM, but as this only affects L2 cache misses, the amount of slowdown highly depends on the cache miss rate, and thus on the application you are using. I might get around to get some Quake scores on a single-SIMM Pentium system with that chipset, to compare them to a 5x86-160 system. Compared to the Intel 430 series, the SiS 5571 is worse anyway, even with 64 bit memory installed.

noshutdown wrote on 2026-06-01, 07:50:

for a long time, its been well known that 586 and later cpus have 64bit fsb and should work with 32bit simms in pairs.
however, i have only recently read that certain 586 class mainboards and chipsets reportedly work with a single 32bit simm, probably by double banking or reverse interleaving(two dram reads/writes for a fsb transfer). needless to say that performance must be heavily degraded, but could probably make the slowest 586 system if needed.
i have little info on this, has anyone confirmed which 586 chipsets actually work in this way, likely via and sis?

SPRING CIRCLE COMPUTER, INC. SF586 REV. VIP3A Socket5 motherboard based on Forex FRX58C613A/602B/601B Chipset.

https://theretroweb.com/motherboards/s/spring … f586-rev.-vip3a

Checked id single SIMM mode with 32MB FPM module

mkarcher wrote on 2026-06-02, 02:07:

Compared to the Intel 430 series, the SiS 5571 is worse anyway, even with 64 bit memory installed.

The 430FX is pretty dire if you don't have a pipeline burst COAST in it. .. I don't imagine the LX or NX are much cop, not that they are easy to find to check.

Sis5511/2/3 chipset supports it too.

I have a Gemlight GMB-P54SPV (Clone of DTK PAM-0036S) that supports it too. If you want even worse performance than single simm, enable the onboard 6205 VGA controller too by not plugging in a PCI VGA.

BitWrangler wrote on 2026-06-02, 04:16:

The 430FX is pretty dire if you don't have a pipeline burst COAST in it. .. I don't imagine the LX or NX are much cop, not that they are easy to find to check.

guess that async cache is simply not much faster than edo dram, thats why plb cache is needed and not the chipset's fault, just don't compare it with other chipsets using plb cache.

i looked up some datasheets, and it claimed that via supported single 32bit simm up to vp2(595), while vp3(597) and later had that removed for whatever reason.
sis supported that up to the 5598, 5591 with agp mentioned only asymmetrical mode and not 32bit mode, 530 is sdram only.
intel docs are in same case as sis5591.

BitWrangler wrote on 2026-06-02, 04:16:

mkarcher wrote on 2026-06-02, 02:07:

Compared to the Intel 430 series, the SiS 5571 is worse anyway, even with 64 bit memory installed.

The 430FX is pretty dire if you don't have a pipeline burst COAST in it. .. I don't imagine the LX or NX are much cop, not that they are easy to find to check.

Yeah, that's what happens if I don't write precisely what I meant. Not your fault. I didn't even think about the LX or NX, because I didn't have much contact with them. When I wrote "430 series", I was thinking about FX and later chipsets, obviously supported by the "correct" cache type, i.e. PB cache.

Actually, I know someone who has an Intel Plato with the 430NX chipset in storage, possibly in unknown condition. If demand is high, I might be able to get some benchmarks from that board, if the board works at all.

mkarcher wrote on 2026-06-02, 17:09:

BitWrangler wrote on 2026-06-02, 04:16:

mkarcher wrote on 2026-06-02, 02:07:

Compared to the Intel 430 series, the SiS 5571 is worse anyway, even with 64 bit memory installed.

The 430FX is pretty dire if you don't have a pipeline burst COAST in it. .. I don't imagine the LX or NX are much cop, not that they are easy to find to check.

Yeah, that's what happens if I don't write precisely what I meant. Not your fault. I didn't even think about the LX or NX, because I didn't have much contact with them. When I wrote "430 series", I was thinking about FX and later chipsets, obviously supported by the "correct" cache type, i.e. PB cache.

Actually, I know someone who has an Intel Plato with the 430NX chipset in storage, possibly in unknown condition. If demand is high, I might be able to get some benchmarks from that board, if the board works at all.

mpe has a good comparison here: https://dependency-injection.com/early-pentium-chipsets/ and discussion Early Pentium Chipset Comparison

BitWrangler wrote on 2026-06-02, 04:16:

mkarcher wrote on 2026-06-02, 02:07:

Compared to the Intel 430 series, the SiS 5571 is worse anyway, even with 64 bit memory installed.

The 430FX is pretty dire if you don't have a pipeline burst COAST in it. .. I don't imagine the LX or NX are much cop, not that they are easy to find to check.

430FX certainly performs much better with PLB (regardless of whether it's onboard or on COAST) than with async cache, but it was a huge leap forwards compared to NX, which only supported async cache and also only could handle FP DRAM, no EDO.

It's easy to be negative about i430FX, but it's hard to understate how amazing its performance was back in the day, it pretty much single-handedly moved Pentium from being an overpriced, hot, underperforming white elephant to an affordable, efficient and downright fast option that was finally clearly faster than 486 systems in applications people actually ran (i.e. not just floating-point benchmarks). But it did need paired SIMMs. Its moment in the sun was brief though, as a year later the i430HX and VX eclipsed it and another year later the i430TX was in a completely different league.

SiS chipsets all could run on single SIMMs, including the 5511, which introduced UMA (memory shared between CPU and onboard or later integrated VGA). If you want the single worst performing So5/7 solution clock for clock, get a board with 5511 and 6202 VGA, and add a single SIMM to it. Its already excessively narrow bandwith will then get shared between CPU and VGA.

jakethompson1 wrote on 2026-06-02, 17:18:

mpe has a good comparison here: https://dependency-injection.com/early-pentium-chipsets/ and discussion Early Pentium Chipset Comparison

thats a very good comparision but it also indicates that plb cache seems to have largest impact on performance.
is there another comparision showing how the intel 430hx and tx perform against newer 66fsb chipsets including vpx/vp2/vp3, 5571/5582/5597, ali4(m1531) and others(excluding 100fsb super7 chipsets)?

noshutdown wrote on Today, 02:04:

jakethompson1 wrote on 2026-06-02, 17:18:

mpe has a good comparison here: https://dependency-injection.com/early-pentium-chipsets/ and discussion Early Pentium Chipset Comparison

thats a very good comparision but it also indicates that plb cache seems to have largest impact on performance.
is there another comparision showing how the intel 430hx and tx perform against newer 66fsb chipsets including vpx/vp2/vp3, 5571/5582/5597, ali4(m1531) and others(excluding 100fsb super7 chipsets)?

There was a historical chipset to chipset benchmarking thread here once upon a time. That also showed memory bandwidth.

Search function is your friend

It is sometimes my friend... pulling up too much, or missing stuff I know damn well is here, but anyway, on digging thus far, this thread is interesting...
Socket 5 & 7 Motherboard VGA Benchmark comparison
Though personally, I think some of those boards would show different rankings with a faster common CPU, maybe a 233MMX rather than the p100. Baked in latencies are bad until they let you run faster CPU at tighter cache and RAM timings.