Write allocate is a setting on the CPU that you can toggle regardless of chipset. Whether it's faster on or off depends on the application. Games that do software rendering are usually faster with it off, so that framebuffer writes don't cause unnecessary reads/thrashing.
Linear Burst was only supported on some chipsets. Maybe someone here knows exactly what it was, but I don't think it made a huge difference. I seem to recall that it was something Cyrix did differently than Intel just to avoid patent infringement.

Some makers managed to work some optimisations into the BIOS on intel boards but they were at risk of incurring Intel's petty wrath and losing priority on Intel chipsets and tech info, so maybe didn't put too much effort into it. However, another thing to remember to wring the best out of a Via board is to use DIMM compatible with the 4x bank interleave feature.

Many non-intel chipsets handled 75mhz+ FSB much better, some of the sis motherboards had 33mhz async PCI which only works with Cyrix.

From what I remember only one board was 75mhz FSB certified at the pr200 launch

bakemono wrote on 2026-10-28, 02:08:

Write allocate is a setting on the CPU that you can toggle regardless of chipset. Whether it's faster on or off depends on the application. Games that do software rendering are usually faster with it off, so that framebuffer writes don't cause unnecessary reads/thrashing.

I'm not even sure Write Allocate was a Cyrix thing or more of a K5 or K6 option in the BIOSes of the era. Good to know regardless.

BitWrangler wrote on 2026-10-28, 02:20:

Some makers managed to work some optimisations into the BIOS on intel boards but they were at risk of incurring Intel's petty wrath and losing priority on Intel chipsets and tech info, so maybe didn't put too much effort into it. However, another thing to remember to wring the best out of a Via board is to use DIMM compatible with the 4x bank interleave feature.

Good catch. Was 4-way Bank Interleaving a thing in VIA chipsets all the way back in 1996/1997? I thought it was more of a Apollo Pro133/Pro133A thing and later.

rmay635703 wrote on 2026-10-29, 00:38:

Many non-intel chipsets handled 75mhz+ FSB much better, some of the sis motherboards had 33mhz async PCI which only works with Cyrix.

From what I remember only one board was 75mhz FSB certified at the pr200 launch

I'm leaning towards this as well, that the recommendations came because of better support for the 75MHz and latter 83MHz FSB, rather than any specific BIOS/chipset feature that unlocked some "super secret ultimate performance mode" on the processors. That being said, I've had exceptionally good experiences running all my 430VX and 430TX boards at 75MHz, even if the PCI bus likely gets a bit overclocked. Some people even swear by 83MHz on these boards, but I haven't had good experiences with that. And Super 7 boards essentially make this a moot point today, if you want to run Cyrix processors with weird FSBs you can just grab one of them.

I think the reason I made this thread and kept thinking about the subject is that I always see people comparing Cyrix CPUs with Intel ones from the era and they all do it on the same platform, which makes sense, it's convenient and saves a great deal of time, but it's also kind of paradoxical in the sense that I've read so many conflicting reports that Cyrix CPUs were really undermined when not run on a VIA or SiS chipset or something. Which in itself doesn't make a ton of sense, given that chipsets other than Intel's at the time were not as fast and definitely not as stable as Intel's (at least in certified settings, such as 66MHz FSB).

What I'm getting at is that maybe those features did in fact give the Cyrix processors a performance boost, but overall performance of 430VX and HX chipsets likely made up most of the difference, to the point where it didn't make much sense not to go with them instead of VIA/SiS/ALi/etc. Just strange that I can't seem to find a single thread here on VOGONS comparing performance across 2 or 3 different chipsets.

Thank you for the answers so far.

Garrett W wrote on 2026-10-29, 08:39:

Good catch. Was 4-way Bank Interleaving a thing in VIA chipsets all the way back in 1996/1997? I thought it was more of a Apollo Pro133/Pro133A thing and later.

It appeared in the VP series. See these notes for an enabler util if the BIOS forgot to add it...
https://soggi.org/files/misc/tools/Memory-Int … bler-readme.htm (link at top of page takes you to the utility download page to find it)
Some enabled/disabled benchmark results given there.

Then also in the FIC motherboards page at DOSdays, it is mentioned that the PA-2010 needs it disabled at 83Mhz, and that is a VPx. So not sure if it was new to the VPx and whether it's FIC or Via's fault it doesn't work at 83mhz on that board.

I always heard it needed 64Mbit chip on the SDRAM to enable 4 bank interleave but that page above says the driver will still enable on 16Mbit, but not sure whether it is cautioning not to because the driver doesn't have a check, or pointing out that with only 2 banks per chip you'll need 2 modules for it to work instead of 1+ with 64Mbit.

Edit: actually would be interesting to see if the PA-2010 just has a boot/POST glitch and that driver still works at 83 if you leave it unset in CMOS setup.

Kowalski! Analysis?
https://www.vogonswiki.com/index.php/Super_So … et_7_Benchmarks

I think those tables might give us a few clues but will have to look at it later unless anyone beats me.

I've run the IBM/cyrix 6x86 CPUs on a few different boards including the 430VX and 430TX, and I think they mainly benefit from bus speed. Later socket 7 boards with higher bus speeds typically used chips from VIA, SiS and Ali. This may be what you're referring to? The newer chips set to low speed seem to be about the same as older ones. If there's some test in particular that you're interested in, I can try to check it. For an example, try a 6x86 at 200MHz (2x100) vs 210MHz (3.5x60).

The 6x86L has less cache and is particularly helped with higher bus speed. The 6x86M has a larger cache and can keep up with the K6-2. The full speed, on-die cache of the K6-3 and "plus" CPUs seems to be the best help for slower bus speeds.

BitWrangler wrote on 2026-10-29, 12:26:

It appeared in the VP series. See these notes for an enabler util if the BIOS forgot to add it... https://soggi.org/files/misc/t […]
Show full quote

It appeared in the VP series. See these notes for an enabler util if the BIOS forgot to add it...
https://soggi.org/files/misc/tools/Memory-Int … bler-readme.htm (link at top of page takes you to the utility download page to find it)
Some enabled/disabled benchmark results given there.

Then also in the FIC motherboards page at DOSdays, it is mentioned that the PA-2010 needs it disabled at 83Mhz, and that is a VPx. So not sure if it was new to the VPx and whether it's FIC or Via's fault it doesn't work at 83mhz on that board.

I always heard it needed 64Mbit chip on the SDRAM to enable 4 bank interleave but that page above says the driver will still enable on 16Mbit, but not sure whether it is cautioning not to because the driver doesn't have a check, or pointing out that with only 2 banks per chip you'll need 2 modules for it to work instead of 1+ with 64Mbit.

Interesting, one more thing to try I guess if I eventually start doing some tests. I've just had a look and both FIC PA-2010 (VP1 based) and PA-2007 (VP2 based) have a jumper next to the cache chips that can be jumpered for Intel Burst or Linear Burst, so I'm thinking @bakemono might be onto something about the way Intel did things was patented.

As for the benchmarks for the VOGONS wiki page, I don't think they are especially helpful for what I am trying to discern.

nickles rust wrote on 2026-10-29, 22:23:

I've run the IBM/cyrix 6x86 CPUs on a few different boards including the 430VX and 430TX, and I think they mainly benefit from bus speed. Later socket 7 boards with higher bus speeds typically used chips from VIA, SiS and Ali. This may be what you're referring to? The newer chips set to low speed seem to be about the same as older ones. If there's some test in particular that you're interested in, I can try to check it. For an example, try a 6x86 at 200MHz (2x100) vs 210MHz (3.5x60).

The 6x86L has less cache and is particularly helped with higher bus speed. The 6x86M has a larger cache and can keep up with the K6-2. The full speed, on-die cache of the K6-3 and "plus" CPUs seems to be the best help for slower bus speeds.

No, I'm not talking about the bus speed, I think it's fairly easy to understand how this would affect things. I've seen posts and magazine articles recommending VIA boards instead of Intel ones even for CPUs using the 66MHz FSB such as the PR166.

If you have a 430VX or TX board and a VP1 or VP2 board and are willing to do a couple of benchmarks I'd be grateful! I would maybe compare a classic PR166 (L variant or not shouldn't matter) on both boards and see if the BIOS on the VIA side has any options relating to the Cyrix CPU.

Garrett W wrote on 2026-10-30, 09:13:

As for the benchmarks for the VOGONS wiki page, I don't think they are especially helpful for what I am trying to discern.

Damn yeah, giving them a thorough look there's almost no results for Machine 4 which would be our Intel TX baseline, then all the others barely have one CPU speed that matches another and the Via machines barely have any numbers there either.

This article would be fairly informative, apart from I can't be assed subscribing to Toms now they put that fence up...
https://www.tomshardware.com/reviews/review-s … rboards,46.html

However, it seems like there's less than a whole percent between best 430TX board and the best VP2 board in this graphic, the x axis cut off at 65 so exaggerated spread there's less than 4% between top and bottom.

I wouldnt believe anyone recommending non Intel socket7 chipset. Especially when combined with PB cache Intel could be up to 20-30% faster than some popular because cheap at the time VIA/SIS abominations https://dependency-injection.com/early-pentium-chipsets/
Im not familiar with any Intel chipset having problems at 75MHz, and afaik 83MHz gave only problems with some disks (writes) when mobo clock gen didnt support Async PCI clock 33MHz output.

nickles rust wrote on 2026-10-29, 22:23:

The newer chips set to low speed seem to be about the same as older ones.

I tried a 6x86 in a 430v board at 167MHZ (2.5 x 67) and also in a MVP. The 430v has EDO and 256kB cache; the MVP has SDRAM and 512kB:

1BBBBBB   YYY   Y  TTTTTTT  EEEEEEE
2BBB   B  YYY   Y    TTT    EEE
3BBB   B  YYY   Y    TTT    EEE
4BBBBBB    YYY Y     TTT    EEEEEEE
5BBB   B    YYY      TTT    EEE
6BBB   B    YYY      TTT    EEE
7BBBBBB     YYY      TTT    EEEEEEE
8
9430v
10BYTEmark (tm) Native Mode Benchmark ver. 2 (3/95)
11NUMERIC SORT:  Iterations/sec.: 71.649463  Index: 1.851647
12STRING SORT:  Iterations/sec.: 7.861871  Index: 3.455768
13BITFIELD:  Iterations/sec.: 14694742.721916  Index: 2.520613
14FP EMULATION:  Iterations/sec.: 4.540573  Index: 2.182968
15FOURIER:  Iterations/sec.: 1076.246993  Index: 1.218632
16ASSIGNMENT:  Iterations/sec.: 0.503029  Index: 1.916521
17IDEA:  Iterations/sec.: 150.150235  Index: 2.297280
18HUFFMAN:  Iterations/sec.: 71.943926  Index: 1.999275
19NEURAL NET:  Iterations/sec.: 0.558460  Index: 0.944941
20LU DECOMPOSITION:  Iterations/sec.: 28.820696  Index: 1.701541
21...done...
22===========OVERALL============
23INTEGER INDEX: 2.269308
24FLOATING-POINT INDEX: 1.251331
25 (90 MHz Dell Pentium = 1.00)
26==============================
27
28MVP
29BYTEmark (tm) Native Mode Benchmark ver. 2 (3/95)
30NUMERIC SORT:  Iterations/sec.: 71.296321  Index: 1.842520
31STRING SORT:  Iterations/sec.: 7.880460  Index: 3.463938
32BITFIELD:  Iterations/sec.: 14717674.190382  Index: 2.524547
33FP EMULATION:  Iterations/sec.: 4.531848  Index: 2.178773
34FOURIER:  Iterations/sec.: 1078.738305  Index: 1.221453
35ASSIGNMENT:  Iterations/sec.: 0.505562  Index: 1.926169
36IDEA:  Iterations/sec.: 150.777348  Index: 2.306875
37HUFFMAN:  Iterations/sec.: 72.255625  Index: 2.007937
38NEURAL NET:  Iterations/sec.: 0.559511  Index: 0.946719
39LU DECOMPOSITION:  Iterations/sec.: 28.967062  Index: 1.710182
40...done...
41===========OVERALL============
42INTEGER INDEX: 2.272737
43FLOATING-POINT INDEX: 1.255198
44 (90 MHz Dell Pentium = 1.00)
45==============================

nickles rust wrote on 2026-10-30, 17:37:

I tried a 6x86 in a 430v board at 167MHZ (2.5 x 67) and also in a MVP. The 430v has EDO and 256kB cache; the MVP has SDRAM and 512kB:

All within margin of error despite mpv3 being newer with in theory very fast SDRAM controller 6-1-1-1-3-1-1-1 so pretty much at cache speed.
Triton cache PB or async?
Bios settings maxed out on both?
Bios quality. Some bioses dont even allow setting fastest possible timings.
Actually measuring FSB on both boards 😀 yes its that silly 😀 Having read over 50 PLL datasheets in last week there are some stupid things out there like PLLs rounding up setting to hundreds of kilohertz up or down willy-nilly, from memory record holder was generating 3.5 MHz when set to 4MHz (12% off). 66 vs 66.6 is pretty standard discrepancy.

BitWrangler wrote on 2026-10-30, 13:46:

This article would be fairly informative, apart from I can't be assed subscribing to Toms now they put that fence up...
https://www.tomshardware.com/reviews/review-s … rboards,46.html

However, it seems like there's less than a whole percent between best 430TX board and the best VP2 board in this graphic, the x axis cut off at 65 so exaggerated spread there's less than 4% between top and bottom.

Hey, now we're getting somewhere! Looks like even with Linear Burst, the TX is either equal or ever so slightly ahead. That being said, the VP2 boards are equipped with 1MB of L2 cache so it ain't exactly fair (and wouldn't have been cheap!). I wish they had done a few more benchmarks though! And maybe added in some 430VX boards which were very cheap and aren't really noticeably far off the 430TX.

rasz_pl wrote on 2026-10-30, 14:38:

I wouldnt believe anyone recommending non Intel socket7 chipset. Especially when combined with PB cache Intel could be up to 20-30% faster than some popular because cheap at the time VIA/SIS abominations https://dependency-injection.com/early-pentium-chipsets/

MPE's work on this is fantastic, I wish they updated this with some more chipsets though. Still doesn't truly answer my question since the only VIA chipset in there is the Apollo Master.

nickles rust wrote on 2026-10-30, 17:37:

I tried a 6x86 in a 430v board at 167MHZ (2.5 x 67) and also in a MVP. The 430v has EDO and 256kB cache; the MVP has SDRAM and 512kB:

Hey that's pretty nice, thanks! Your VIA board seems to be VP3 based and not MVP3 by the way. Which one is it out of curiosity?
Would you mind giving a classic 6x86 166 a go as well, if you have one? Are there any options on the VIA board pertaining to Linear Burst or anything of that ilk? Thanks in advance.

Yeah, mpe's results are fine far as they go, but a bit like trying to infer triton III performance from mercury, given they were everyone else's first gen.

rasz_pl wrote on 2026-10-30, 20:54:

Actually measuring FSB on both boards 😀 yes its that silly 😀 Having read over 50 PLL datasheets in last week there are some stupid things out there like PLLs rounding up setting to hundreds of kilohertz up or down willy-nilly, from memory record holder was generating 3.5 MHz when set to 4MHz (12% off). 66 vs 66.6 is pretty standard discrepancy.

Good point. I remember some minor controversy about Abit always having their 66Mhz at 68 for a year or two. Then when caught they put it as optional Abit turbo speed in CMOS setup, which was in their PII boards. All this happened pre 1998 in the web memory hole (i.e. mostly uncaptured by archive.org and maybe most argued about in lost compuserve conferences)

Edit: Ah, old THG article remains on the .ru domain. https://thg.ru/mainboard/19960702/ where it is said to be documented. Not sure then what the "sneakiness" was about it that I remember vaguely. Possibly that they were "accidentally" leaving it set on review samples, or on demo tradeshow or press event machines. Possibly 3rd party builders using Abit boards were setting it to make their machines look better.

BitWrangler wrote on 2026-10-31, 13:07:

Good point. I remember some minor controversy about Abit always having their 66Mhz at 68 for a year or two.

and here is one of the clever PLLs that could facilitate it PhaseLink PLL52C59-14 https://theretroweb.com/chips/4791
It was available in three sub versions.
-14L - no idea about this one, datasheet unavailable, boards using it only list official 60 / 66 FSB choices.
-14A where A is for AMD fsb: 50, 55, 60, 66.6, 75, 75 (Asyn PCI), 83.3, 83.3 (Asyn PCI) MHz - really nice for running 83MHz fsb with safe PCI clock
-14T where T is for Turbo 😉 fsb: 50, 51.3, 55, 60, 61.6, 66.6, 68.4, 75 MHz - I submitted datasheet to theretroweb

sneaky "One turbo input pin accelerates 3 CPU frequencies (50,60 and 66Mhz) by 2.5%" and bam you get magical ranking wins for free 😀

Edit: wait, https://thg.ru/mainboard/19960702 you linked is specifically about this PLL 😁 and I stumbled upon it last week independently 😀

The board I tested above is an EPoX EP-58MVP3C-M and the main chip is labeled VIA VT82C598MVP. The only different bios setting I see is the ADS# delay. It doesn't seem to do much that I can tell.

I don't have an earlier version of the 6x86. I think it would just be slower (at the same clock) due to less cache. Some of the last versions also got slower (at the same clock) for some weird reason. These threads have some good info:

www.vogons.org/viewtopic.php?t=31111

www.vogons.org/viewtopic.php?t=86803

Ah, so it must be Speedsys being unable to recognize your board for some reason. I think that settles it, even with the fastest timings and all relevant options on an MVP3, it's really not ahead of a 430VX with half the cache.

I'm trying to think of any benchmark that might actually show a difference, but I can't really think of anything that might matter. Maybe try Quake at 320x200 using Phil's VGA Bench suite? Anybody have a better idea? To me this really just seems like an open and shut case now.

I am familiar with the last MII CPUs getting slower per clock by the end there. I wonder what was going on, probably looser timings on the cache?

I have an MS-5158 430TX board which is stable at 83MHz and will even run at 100MHz with a few caveats (L2 cache disabled, IDE for the HDD replaced with a PCI controller card.) I had not planned to push the machine that far, the PC I bought with this motherboard in just happened to be capable of it.

I have some nostalgia towards the cyrix MII. The 2nd CPU I ever bought was a PR233 that I overclocked in my PCChips M571 (SIS 5598 based) but it was a very slow board even when the FSB was cranked up. I think it was a poorly optimised implementation of that chipset. Necroware did a video of an ASUS SP97-V and got some great results, though he did not try a cyrix in there.

I wouldn't mind getting an end of line MII - whatever the fastest one is and seeing how it compares to my K6-3+