Some more numbers:

Performance at 180mhz. Note, the FPU speed here is actually slightly lower then what my Cyrix at 2x60mhz scores in the same motherboard with enhancements enabled, however I have yet to find a game in which the AMD at 180 wouldn't feel faster. Obviously this is a bit of an apples to oranges comparison though as the cyrix needs much relaxed cache timings compared to the AMD.

Well, your 200mhz configuration is impressive. Best quake 1 score with 486 cpu so far.
Tell us more about it - bios version, ram chips (looks like samsung), post few more scores like Doom test.
Did you run any other more challenging benchmarks to verify stability?

Well done, I wonder if my SIV utility correctly reports the CPU speed, please will you try it?

Ideally use SIV V5.82 Beta-05, do Menu->File->Save Local and post the save file.

Below you can see it's OK on my system, but I have never tested it with a 60/66 MHz FSB speed.

I wonder why CPUZ reports the L1 cache size + technology for your CPU, but not for my CPU when that are basically the same CPU.

At x4 it repots the technology, but still no L1 cache size.

I wonder if your system would be more stable a 4 x 50 MHz, have you tried it?

pshipkov wrote on 2025-07-04, 05:55:

Well, your 200mhz configuration is impressive. Best quake 1 score with 486 cpu so far.
Tell us more about it - bios version, ram chips (looks like samsung), post few more scores like Doom test.
Did you run any other more challenging benchmarks to verify stability?

Thanks. I've been wanting in the club for some time now and working hard at it 😁

Didn't someone score around 25 with a 1:1 fsb/pci ratio? I could probably milk out an extra fps if I used one of my cards that can do that speed, however not that motivated since the G200 outperforms those under windows regardless of the benchmark score.

Bios version is stock for this board 05/06/1996. Extensive testing confirmed findings of others that all the other bioses available have a gap in the cacheable range between 56-64 mb, leading to lower performance, while not giving any other observable benefit. I tried a few award bioses of similar boards as well, but these all had other issues. Note, this bios also works flawlessly on a v1.5 board (that has a version with the memory gap from factory).

The edo ram has 5v tsop Samsung ic's - km44c16104bs-6. I tracked this down after some recomendations from Feipoa regarding Samsung tsop ram. I also have a 64mb 40ns (!!!) module with soj Samsung chips that performs flawlessly, however since there seemed no difference stability wise I obviously stuck with the 128. I also had great results with a 64mb module of Samsung tsop FPM - benchmark results were 1-2 us/mbps better with this and no instability was noted (I think one of my benchmark result photos is with this module). However, the FPM "felt " a tad slower in windows - not sure if EDO is playing a role here, or, if the FPM is at the edge of it's capabilities and I am getting cache misses or something that affects the performance.

Yeah I can post some more results.

So far, close to a week in, all seems stable, even leaving demanding windows games paused with winamp in the background etc all day has not triggered anything. I've messed with a few of these less successfully and usually causing lockups or errors isn't a problem

Wonder if you could get by that shutting down issue by clocking down on the fly, just before closing Windows. My 486SOM's clock generator can switch frequency live, clocking up or down as needed. By red-ray's inspiration, I added switches rather than jumpers, to be able to switch as I please ..

The multiplier, though is read at reset, so that's not a live switch ...

pshipkov wrote on 2025-07-04, 05:55:

Well, your 200mhz configuration is impressive. Best quake 1 score with 486 cpu so far.
Tell us more about it - bios version, ram chips (looks like samsung), post few more scores like Doom test.
Did you run any other more challenging benchmarks to verify stability?

Here is a few more benches

Doom

Pcplayer 320 and 640

Did you do your test with the G200 or the Virge?

S3 Virge can't work stable at 60 MHz. At least I haven't encounter one that could.

Very impressive results

Disruptor wrote on 2025-07-05, 14:59:

Did you do your test with the G200 or the Virge?

Sharp eye. That round in particular was done with an S3 virge (with a 75mhz clock) and I believe FPM memory. I was wanting the virge in there as the g450 I usually use for dos tests (as it scores identically to my g200 which usually lives in a system/doesn'tserve as a test card) does not have windows 98 drivers that work on a 486.

The virge consistently scores about o.2 quake fps LESS then g200, while using FPM consistently scores about 0.2 more then using edo. SO, virge + FPM or g200 + edo yields an identical quake score. The virge obviously lacks heavily under windows compared to the g200

The Serpent Rider wrote on 2025-07-05, 15:30:

S3 Virge can't work stable at 60 MHz. At least I haven't encounter one that could.

I agree. I tested several and always had problems eventually. The ones with a lower clock rate did a little better but never reliable.

HOWEVER, I'm NOT running this one at 1:1 fsb/pci, so the pci is comfortably running at 30mhz. Honestly, I can't be bothered to hot switch it every boot to achieve a 1:1 pci at this speed (which is the only way to force it on a 919). I've had mixed results with running 60mhz pci on SiS and Ali boards - sometimes there is an extra fps to be had, sometimes not. I remember the Ali board in particular scoring over 100 in 3d bench with a sis6326 and a 2x60mhz cyrix which was very impressive, however the improvement to any other benchmark (or to the feel of games under windows) was very minimal

Yes, thats evident by your Doom framerate. It is around 60, which if far below the 80 if you run at 1:1.
Basically, runtimes that can saturate the pci/vlb busses will suffer significantly.

Thanks for the additional details.
Linked you from a big thread related to this stuff.

pshipkov wrote on 2025-07-06, 05:36:

Yes, thats evident by your Doom framerate. It is around 60, which if far below the 80 if you run at 1:1. Basically, runtimes tha […]
Show full quote

Yes, thats evident by your Doom framerate. It is around 60, which if far below the 80 if you run at 1:1.
Basically, runtimes that can saturate the pci/vlb busses will suffer significantly.

Thanks for the additional details.
Linked you from a big thread related to this stuff.

Oh really? I didn't realize it'd affect doom so much. What VGA?

I think I found my sis6326 to handle that pci speed best (as it's an agp chip, what a shame that it's so mediocre otherwise), followed by one of my slower virge's (was ok in dos but would artifact in windows). I remember noting that the slower virges would suddenly end up faster then the fast ones memory bandwidth wise, which seemingly must have automatically added wait states to cope with the bus speed (but were still unstable).

If the 919 had an easy way of forcing 1:1 permanently so I wouldn't have to fiddle with it every boot (could anyone edit the bios? The boards with this chipset that use Award instead of Amibios have it selectable) I would perhaps explore this more one day

Yes, running the PCI bus with some divider will have a significant impact on such runtimes.
Assuming that a decent to good graphics card is used, the tight spot will be the bus.
In general, for DOS it is better to use VLB cards - the good ones are slightly faster than the good PCI cards.
For Windows, PCI cards (the good ones) are better than the VLB ones.

PCI - Matroxes, Voodoo3, etc
VLB - Ark1000, S3 Trio64, or the 96* variants.

How are you adhereing that oversized socket 370 heatsink to the socket 3?

180 MHz at 4.0 V was my setting max on the M919 with 2-1-2, 1/0 ws, 1024K, 128 MB EDO. As I recall it, 3.45 V was mostly stable except for some demanding Windows 95 software titles.

Windows 9x shutdown issues weren't uncommon even with non-overclocked systems. If at 200 MHz, you shut down by the option to "restart in MS-DOS mode", does the system still hang? Once in DOS mode, hit the power button. I try to shutdown on most of my Windows 95 systems like this.

bertrammatrix wrote on 2025-07-04, 03:30:

Performance at 180mhz. Note, the FPU speed here is actually slightly lower then what my Cyrix at 2x60mhz scores in the same motherboard with enhancements enabled, however I have yet to find a game in which the AMD at 180 wouldn't feel faster. Obviously this is a bit of an apples to oranges comparison though as the cyrix needs much relaxed cache timings compared to the AMD.

That's very common and I think it is why the Cyrix doesn't shine so much with overclocking compared to the AMD. However, there have recently been some very interesting tests at 150 MHz using a QFP Cyrix 5x86-120. Seems someone got it stable on an M918: Re: Cyrix 5x86 Register Enhancements Revealed I have 4 of these latest S1R3 QFP chips I [eventually] plan to try my luck with at 150 MHz. 3 units I still need to solder to PGA boards. They have the latest datecodes for S1R3 chips, meaning branch prediction should work in Windows, not just in DOS.

Some comparative results from the other thread, on an M918 w/cx5x86-150, 3-1-1-1, 1024K, 32M:

L1 = 309 MB/s ------> compared to 207 MB/s (Am5x86-200)
L2 = 105 MB/s ------> compared to 112 MB/s (Am5x86-200)
RAM (read) = 69 MB/s ------> compared to 62 MB/s (Am5x86-200)
Quake = 21.5 fps ------> compared to 23.4 fps (Am5x86-200)

However, if he is able to find a stable EDO/SRAM configuration to allow for 2-1-1-1, Quake jumpers to 22.6 fps. Still slower than the AMD at 200 MHz, but the Cyrix FPU should clobber the AMD's.

bertrammatrix wrote on 2025-07-06, 19:45:

If the 919 had an easy way of forcing 1:1 permanently so I wouldn't have to fiddle with it every boot (could anyone edit the bios? The boards with this chipset that use Award instead of Amibios have it selectable) I would perhaps explore this more one day

I'm sure one day someone on the forum will be motivated enough to use a relay or switching transistor to do the 1:1/2 to 1:1 PCI switch automatically, perhaps using an Arduino. It's an idea I pitched in around 2012, but no biters yet. From my experience with the M919 PCI switch, it is best to do just after boot at an idle DOS prompt.

pshipkov wrote on 2025-07-07, 16:09:

Yes, running the PCI bus with some divider will have a significant impact on such runtimes. Assuming that a decent to good graph […]
Show full quote

Yes, running the PCI bus with some divider will have a significant impact on such runtimes.
Assuming that a decent to good graphics card is used, the tight spot will be the bus.
In general, for DOS it is better to use VLB cards - the good ones are slightly faster than the good PCI cards.
For Windows, PCI cards (the good ones) are better than the VLB ones.

PCI - Matroxes, Voodoo3, etc
VLB - Ark1000, S3 Trio64, or the 96* variants.

Switching to PCI at 60 MHz will also allow for a whole new round of tweaking, depending on the entirety of expansion cards you are wanting to use in the system. Seems that I ran into all sorts of issues with this on socket 3 boards.