Well, that 495SX board requires 8kx8-bit TAG RAM for a single 128kB (4x 32kx8-bit) bank of cache and dual 8kx8-bit TAG chips for the full 256 kB (dual 128kB banks) configuration that's installed.

My board still has the 20 ns cache and tag chips installed. (I tried substituting some faster 32kx8-bit SRAMs for TAG, but this board didn't like that at all in spite of the mostly identical pinout: I also tried isolating the extra address pins which are NC and CS2 on normal 8kB chips, but this board must make use of CS2 as it then hung at the RAM test rather than during DOS booting)

I'm pretty sure some boards aren't fussy about 32k vs 8k SRAMs installed as cache, even if they don't need 8k ones. (and even back then, I'm pretty sure the dropping cost and large supply of 32kB density SRAMs would've made that an appealing option to simplify logistics at the user/shop end)

Anyway, I'm not sure what exactly is going on in my case of if I can enable write-through cache on this board, but things that really like fast access to DRAM seem to run better with the L2 disabled entirely. The 640x480 PC Player benchmark in particular likes that and is faster with no L2 than with 2-2-2-2 cache timing for the L2 and even slightly faster than 2-1-1-1 timing if I don't have auto-configuration set in the advanced-chipset options. (which seems to be the only way to enable page-mode burst reads, or I'm guessing that's where the speed comes from)

2-1-1-1 timing + auto-config manages to be a little faster with L2 than without for PC Player VESA mode.

For most DOS benchmarks, the cache has a lot more benefit or at least not a disadvantage and both Quake and Doom prefer L2 to just faster DRAM, though I haven't done runs of 640x480 Quake yet.

I also worked out the VLB wait state settings on the board and set it for the 33 MHz (or slower) 0 wait option, which sped things up a little and also seems to work at 40 MHz with the Trident 9400 and Diamond Stealth 24 and I also switched the Trident card's own waits to zero but noticed the DRAM timing jumper was set for 60/70 ns and it has 80 ns 256kx16-bit chips (1MB 32-bit) soldered to the board, so I set that to 80 and cured the video artifacting problem it had had in SVGA/VESA resolutions. (looks like someone had previously set it up for 70 ns DRAM timing + waits for 40/50 MHz bus speed)

https://stason.org/TULARC/pc/graphics-cards/U … -VERSION-1.html

That Trident card is too bad like that now and both it and the S3 card seem to have cleaner video outputs than the Avance Logic one (there's a jailbar moire pattern on the Avance card that I thought was just scaling artifacts on my LCD VGA monitor, but those artifacts are almost absent on the Trident and totally on the Diamond card, plus seem absent in the Avance Logic card when using unchained 320x240 modes ... or maybe it's related to signal quality when running in 60 vs 70 Hz: since Doom runs in unchained mode and has the same artifacts as 13H/MCGA mode... and I haven't tried setting up a custom batch file in any of my games or benchmarks to adjust the VGA synch rates, though that's supposed to be possible for both vsynch and hsynch: ie allowing 15.7 kHz 60 Hz output too)

The Avance Logic card also seems to make the system unstable when 2-1-1-1 cache timing is enabled, but the other cards (and ISA cards) don't seem to do that.

I don't have a Tseng 4000 VLB card to directly compare, but I think the Avance Logic card might be closer to that speed range, though I think the Stealth 24 is supposed to be reasonably fast too (and better than the Mach 32 for example).

It's possible the Opti chipset is also bottlenecking these cards and the vogons wiki also mentions the 495SX added buffering to the VESA bus to improve stability (or compatibility) at the expense of performance, but I'm not sure that's true on the whole or simply referring to the board/chipset level wait state functionality.

I've also seen some reference to 'slow' ISA implementations on PCI 486 (and presumably Pentium/later) boards vs 'fast' ISA, but I'm not sure that's related to the behavior I noticed when comparing ISA video cards in Socket 7 boards vs ISA/VESA in that OPTI board.

I wish I could remember where I came across the slow ISA description but forgot to make a note of it. I think it came up when I was looking for documentation on some SiS 496/497 chipset based boards.
(I think it was the PCI400 and Zida Tomato 4DPS, which also seem to be virtually identical boards ... or actually identical depending on the specific revision of each: some revisions seem to include discrete keyboard controllers for example and others seem to lack them and have a bunch of resistors and capacitors in the spot where the keyboard chip was ... or would be: it actually seems like it's later revisions of boards that add the keyboard controller so maybe they had issues with the chipset integrated MCU)

Anyway, I'll need to double check my notes and screenshots, but I think 4.9 was the best score I got at 120 MHz with the Avance Logic card and L2 disabled. And I believe it dropped to 4.5 or 4.6 FPS with the cache enabled (in 2-2-2-2 timing)

And at 33 MHz fsb (100 MHz) with both the S3 and Trident cards I got 3.9 with the L2 cache off, 4.1 with cache (both with auto-config enabled), and 3.8 with auto-config disabled and 2-1-1-1 cache timing selected.

For some reason, the S3 card managed faster 3Dbench scores than the trident one, but not in PC Player, even at 320x200. (though as mentioned earlier, at 320x200 the L2 cache was already preferred enabled even with worst case)

Hmm, I also don't think there's a way to increase DRAM wait states and have page-mode burst cycles enabled. I could try messing with it more, but I think the BIOS ignores all the user input parameters for RAM/cache timing and maybe the ISA divider setting too when auto is enabled. (otherwise it would be easier to confirm that page mode is being used: turning up wait states and trying to enable/disable page-mode would show a much more dramatic bandwidth difference than minimal wait states, and burst read cycles would probably be 6-2-2-2 with 'zero' waits and 9-2-2-2 with 3 wait states)

I'm not sure why they bothed including that auto-config function and not user configurable DRAM timing and page mode enable/disable.
(also allowing 2 or 3 clock page mode cycle times would be useful for a wider array of DRAM speeds, but I'd assume a cheaper approach would just be to lock that at 2-tick timing and disable page-mode entirely for fast FSB + slow DRAM combinations, though I suppose locking it at 3 ticks would be a more conservative, cheap approach: also other considerations like if the DRAM controller runs at the bus clock or a fraction/multiple of it and/or if it's single-phase vs poly-phase clocked logic)

That Mustang SiS Socket 7 board does have fairly detailed EDO and FPM DRAM timing options in the BIOS, though that can be confusing in itself.

Oh right, and Memspeed.com got some interesting figures:

At 33 MHz x3 with the Cx486 DX4 (ST)

l2 off, auto timing on:

46780 kB/s (up to 8kB)
34000 kB/s (16kB onward)

l2 on auto on:

47700 (4k)
46700 (8k)
41800 (16-256kB)
10800 (512-640 kB)

Though come to think of it I wonder why the L1 speeds are so low. That's something I noticed in cachecheck, too. For a clock-tripled chip that seems a bit odd unless it's just a reflection of it being in write-through mode.

Or is there just relatively high latency in the L1 cache?

Oh, apparently that Trident 9400CXi I've been trying out is actually one of the faster Trident VLB cards out there, along with the 9440, certain models of the 8900 (Tiido mentions the 8900D is fast and 8900C is not) though some also seem dependent on the RAM installed or RAM banks populated (and I've got 2 banks of 1MB 32-bit DRAM in mine now) impacts speed as well. (bank interleave could be supported)

I think the Trident 9000 series is supposed to be on the slow end of things, too.

Apparently, the faster Trident chips are fairly close to Tseng ET4000, so that would line up with that Avance Logic card being about the same as the ET-4000. (it'd probably do better in a faster VLB chipset board, though)

Oh and that Avance Logic card is an:
ALI2201 Vesa Local Bus VGA Rev. 2.0

using this graphics chip:
Avance Logic Inc.
ALG2201.0F

there are two jumpers on the avance card I haven't touched, both are open and I don't know what they do, but I suppose trial and error might be in order.

Apparently, the faster Trident chips are fairly close to Tseng ET4000

I have not yet posted Tseng ET4000/W32p PСI results in my PCI bench thread, which trade blows with Trident 9440-3 PCI.

https://www.ebay.com/itm/353073570841

I think it's a joke we can't share rare items anymore but here's potentially a once in a lifetime opportunity.

Boxed CYRIX evergreen 5x86-133 from a deceased estate, new unused, would love to see all the boxwork and manuals if it's different from the AMD version.

There might be some stuff in Evergreen's documentation specific to Cyrix's voltage and heat related warnings and the 3.6 or 3.7 volt requirement of those chips. (vs the more common 3.45 volts that was typical by then, or the previous range of 3.0 to 4.0 volts some boards had)

I'm not sure if Feipoa already figured this out, but I remember reading somwhere in the Ultimate 486 (or was it 586) benchmark comparison that Cyrix DX4 486s were getting consistently better per-clock performance than their 8k cache AMD counterparts and he suspected it was due to the write-back cache. (and then falling behind when pitted against the 16k DX4s from Intel and AMD)

However, I'm seeing significantly better memory bandwidth and read/write/move performance across the board (L1, L2, and DRAM) with Cyrix 486s (DX2 and DX4) than AMD or Intel DX2s or an AMD DX4 (or a Pentium Overdrive, but that's at 1x multiplier).

And write-back vs write-thru settings don't change this memory speed advantage, including in the L1 itself. (I've been trying out a Soyo SiS 471 based board that has L1 mode switchable in the BIOS, I haven't used a DOS utility to try changing that on the older OPTi 495SX board)

I've also only done the DX4 comparisons on that Opti board so far since this Soyo board seems to only want to run 1x and 2x multiplier CPUs so far and I haven't found documentation for jumper settings beyond what's printed on the board. (the ZIF socket also isn't really ZIF and seems on the cheap side ... or at least is more finicky about pin condition/straightness than is typical: ie even if the CPU seats fully in the socket, good contact might not be made)

Enabling Write-back mode also didn't help in cachecheck or speedsys scores when the fastest memory settings were enabled but might have a slight impact with slower DRAM timings. However, system stability is definitely better with WT mode enabled and the IBM BL-DX2 seems quite happy at 2x50 MHz here though my VLB cards are finicky. Oddly enough my ATi Graphics Ultra does quite well with the ISA bus set to 25 MHz. The CL-5420 with VRAM does OK with WS enabled at 25 MHz but is only slightly faster than the ATi GU in Doom and Quake but slower in 3DBench and identical in PCPlayer. (and the CL-5422 is a bit more finicky and also more sensitive to which CF card or HDD I use, but I've noticed that ever since I started using CF cards: the CF to IDE adapter I use also seems to matter, so maybe it's a bus noise problem and/or RF noise coming from the card itself ... and the super loud old Quantum 170 MB IDE drive I was using was more dependable for all tests, overclock or otherwise)

I installed a 12 ns tag RAM chip, but the cache still isn't super happy at 50 MHz even at slowest timings in the BIOS but also isn't much or any faster than the DRAM according to speedsys and cachecheck. (at fastest cache timings it's slightly faster than the DRAM, but not by much) With the Pentium Overdrive I didn't have the cache stability issues, but performance was also much lower so it's probably thrashing memory a lot less. (Cyrix at 2x50 was giving 14 us times for read blocks in Cachecheck vs 16 ns for DRAM and 13ns for L1, but the POD at 1x50 did 25 ns for the L2, 30 ns for DRAM, and 20 ns for the L1)

Speedsys score was 40.51

Memory Bandwith = 167.21

8k L1 cache = 87.33 MB.s
memory throughput = 65.66

ANd reading the graph:

Write = 93.75 MB/s
Read = 63.75 MB/s
Move =34 MB/s
(all uncached)

Not so surprisingly, the 50 MHz POD was still faster in Quake at 12.9 vs 8.2 FPS using the same ISA video cards at 25 MHz. (and with VLB at 2x40 the Cyrix does about 7.8 depending how I tweak it)
OTOH that 8.2 was achieved with no L2 cache and Quake seems to like that with the POD at least where it got 11.1 FPS with the L2 disabled.

I can't frame count it, but Tomb Raider seems faster/smoother at both 2x40 and 2x50 than the POD at 1x50.

My Cyrix DX-2/80 with integrated heatsink will boot DOS at 2x50 MHz but isn't nearly as stable as the BL-DX2/66. (even with a fan blowing on it)

The BL-66 just has a 40x40 mm generic black anodized chinese heatsink with a 40 mm fan screwed onto it and stuck to the CPU with sticky white thermal grease. (old vintage Techspray silicone free stuff, I believe zinc oxide based) Arctic silver ceramique (alumina based stuff) would probably work too, though I think the Aluminum oxide might scrub off CPU markings more easily.

So you want a Cyrix 5x86-133?

I noticed Feipoa mentioned his Cyrix 5x86 tested some years ago wouldn't POST at all when set to 5V and he used 4.0V for some overclocking tests.

He made the assumption that the chips don't want to work at 5V at all, but I think this may be a board (or chipset) specific issue.

They may or may not last very long at 5V, but I did basic POST tests and entered the BIOS with several QFP mounted Cyrix 5x86-100 chips (passive socket adapter PCB mounts, no voltage regulator circuits) and didn't have problems there.

I've avoided actually testing any of those 5x86s further while overvoltated for fear of damaging them, particularly as they might tolerate the overvoltage better than overheating and I noticed a worrying lack of heat coming off the heatspreader surface compared to all ceramic CPUs I tried. (I did short tests without heatsinks, feeling them by finger-touch to the top-middle of the package)

I suspect the poorer thermal conductivity of the plastic encapulation of the PQFP parts ends up insulating the die more and slows the heating of the package. If that's the case, the die itself could be getting far hotter than the surface of the package and would fail to dissipate heat effectively even with a heatsink+fan mounted to it.

It might be OK at 100 or maybe 120 MHz like that, but I'd think going for a higher overclock (150 in that particular board) would be risky. Those same 5x86s did POST at 50 MHz FSB too, which made further tests temping.

I'd think AMD 5x86s (and 350 nm DX4s) would be even worse in this respect in plastic vs ceramic packages. I've done similarly brief (well under 1 minute) POST tests of a couple of ceramic AMD x5 CPUs like that and they get very intensely hot in a very small area (presumably due to the small die size) extremely quickly compared to all the DX, SX, DX2, and DX4 CPUs I've tried like that. So the heat dissipation is extremely concentrated over the die on those 350 nm AMD parts.
They also tended to post at 50 MHz FSB (presumably 200 MHz) but I didn't risk further testing like that.

It's enough to make me wonder if some of the reported failures with overclocking at 5V were more heat related than overvoltage related. 5V is a pretty significant jump beyond the absolute maximum specs of those AMD parts and just generally high for 350 nm silicon, but the thermal issues seem severe enough to be the bigger issue if not really aggressively tempered. (especially given the typical means of heatsink attachment on 486s and especially if tower mounted rather than horizontal with gravity assisted heatsink contact)

If any of the 5V overclocked AM5x86 failures were on QFP parts, that's even less surprising.

Also:
Correction on the Cyrix L1 cache performance, it appears to be somewhat slower than AMD 486s, comparing DX2 chips at the same clock speeds in the OPTi 495SX 3486 board (and older model 5V AM486DX, too).

However the external I/O performance is definitely faster on the Cyrix parts and Speedsys gets easily confused between L1 and L2 cache because of that (at 33 MHz FSB with fastest timings, the Cyrix L1 is only slightly faster than the L2 so it thinks there's only one cache) That I/O (or memory bandwidth) performance still falls away with the L1 disabled, though.

The 'auto configuration' mode also ends up being slower at 2x33 MHz than manually configuring the fastest timing options, but this isn't true at 2x80 or 3x33 MHz so there's some mystery register settings in play.

Quake also ended up faster on the AM486 DX2 than the Cyrix DX2 (I was using my Cx486 DX2/80 with integrated heatsink, but its performance seems identical to the Blue Lightning) and Doom was faster as well.

It may be the L1 cache bandwidth playing a role there since Doom is supposed to like fast on-chip cache a lot and Quake may as well.

The Cyrix DX2 at 80 MHz was getting the same 13 us Cachecheck scores in the L1 range as the AMD DX2 at 66 MHz. OTOH the Cyrix DX2 at 66 MHz was getting L2 and DRAM speeds about the same as the AMD DX2 at 80 MHz. (and the AMD DX2 seems pretty happy at 80 MHz with a heatsink+fan, and is the older 24361 model, definitely 5V and I believe using the same .7 micron process as the AM386 ... )

Hmm, except apparently the Blue Lightning DX2 CPUs I have as well as the ST DX2 66 and DX2 80 are actually 5V parts as well, or I think all but the last one are definitely and the last doesn't have any voltage indication on it. (no 3V or 3.3/3.4/3.6/4.0V markings like some Cyrix DX2s)

If there are accurate:
http://silirium.ru/ibm-486.html
http://www.cpushack.com/cyrix-486-cpus/

The Cyrix+IBM DX2 are .8 micron parts and probably 5V ... except they tend to be numbered by IBM as 486-V266 or -V666 or such and I'm not sure if the V in there indicates 3.3 or 4.0V as with Cyrix branded parts. (or for that matter, if some of the 5V parts use the same wafers as lower voltage parts)

I don't have any 1x multiplier Cyrix 486DX (or SX or Fascache) CPUs to compare. There was a Cyrix DX-50 with mobo combo that sold a few months back at a fairly reasonable price (or probably low given how uncommon Cyrix DX-50s seem to be), but I only saw it right after it sold (I can't find it in ebay's completed listings now, so it was probably more than 3 months ago).

Someone also scored a Cyrix 486DRX 66/33 a couple weeks ago for $25, but that thing was a BIN and only on there for a couple hours before it sold. (assuming it still works, someone got a deal there)

Though I'm starting to suspect the 8kB L1 cache of the also TI 486SXL PGA-132 (386, not 486 socket) might make more of a difference than the clock doubled Cyrix DRX with 1kB of L1. And the 8k Cache PGA-132 SXL2-50 might be better off than the SRX 66/33 for the same reason (overclocking aside for either one).

Unlike those, there's recently been what looks like several batches of new old stock TI 486SXLC-50 (PQFP-100 386SX pinout) CPUs on the market from China, Romania, and maybe elsewhere. I I've got a few now that I plan to eventually graft onto appropriate 386SX boards, though I wish there was some sort of option for a socket instead of soldering them all on. (I could potentially rig up a PGA-100 socket on boards with the solder pads/holes for that around the QFP-100 mounts and mount the 486SXLC-2 chips to QFP-100 to PGA prototype boards, but that might not be worth the cost or trouble compared to just soldering to the motherboard)

I need to get a decent hot air pencil soldering kit before doing any of that anyway. (I have some more, non-working Cyrix QFP 5x86s that need to be resoldered due to lifted and bent pins/leads ... some might need to be totally desoldered and straightened before being resoldered: it looks they were all stacked/nested in piles at some point and had too much pressure put on them that crushed/shifted/skewed them)

I did some tests with a Trident 9440 now too, and it's definitely faster than the 9400CXi and the Diamond Stealth 24 and seems to be faster than the Avance Logic card, too. Albeit I'd need to set the 9440 down to 80 ns DRAM jumper settings to be totally fair with the 9400.

Oddly, the 9440 does well at 50 MHz on the Soyo SiS471 based board (with WS enabled) and is definitely my fastest card in that board, but it's markedly less stable in the OPTi 495SX board at 40 and 50 MHz with or without wait states. (I didn't try turning waits back on on the motherboard end, though, so that might have helped)

It also does weird things in 640x480 VESA mode on the Opti board where the first few scanlines end up displayed interlaced or something (every other line is dropped on the monitor) but the rest of the screen is normal. It might just be a synch stability issue.

I figured out where I'd seen the mention of "Slow ISA" before: It's in the results list of the 3DSpace benchmark. All the ISA video card results from PCI motherboards put (slow ISA) after the results, including the 4DPS SiS 486 boards with AM5x86 CPUs at 160 MHz. (with ISA CL5420 that gets only 51% but with a Trio3D/2X it gets 99%)

My 5V ST 486DX2-80 in a Symphony chipset 386/486 board manages 54% with 70 ns DRAM set to 1 WS in the BIOS. It's probably slightly faster with the CL5422, but I happened to have the 5420 in there at the time, plus the video output is nicer on that card, at least on the old GEM LCD monitor I've been using. (several cards have jailbar moire-like artifacting, especially in 70 Hz modes while others don't, though 320x240 tends to avoid this on all the problem cards for some reason and even lacks the more obvious vertical scaling artifacts ... but maybe it's just a matter of optimized scaling and filtering hardware on both the digital and analog side of things for common 4:3 square pixel resolutions)

Performance of that 5V rated ST 486 seems identical to the Cyrix and IBM branded DX2 chips, which isn't surprising, but I just realized I haven't been consistently keeping track of the CPU identification results in CHKCPU.

Well, I can start with this 5V Cyrix/ST 486DX2-80

Model: 1Bh
Stepping: 32h

Well, it turns out that IBM Blue Lightning DX2-66 I've been using (at up to 2x50 MHz) has the same CPU model and stepping info in CHKCPU as the ST DX2-80.

1Bh
32h

I'm not sure who manufactured these, though. Are these SGS parts rebranded by IBM or IBM parts furnished to SGS like the ST branded 5x86 and 6x86?

In any case, it seems like this Blue Lightning has overclockability in common with IBM's 5x86 and 6x86 parts. The ST DX2-80 runs fine at 80 MHz, even when running quite hot (I tried passively cooling it, which seemed stable, but worryingly hot to the touch, so went back to using a fan) but it really doesn't want to do more than that. (that Cyrix branded DX2 80 will do 2x50 with mediocre stability, better if kept really cool, so it seems in-between the ST and IBM ones; I'll get the CPU ID for that next)

I did some multimeter testing on the Soyo SiS 471 board and it looks like it's set to 5V on the CPU socket Vcc lines. (reads 5.04-5.05V)

I did the same for the Acorp SiS 496/497 board and it appears to be at 3.45V (reads 3.458).

I previously did some testing with the same IBM BL DX2-66 in the Acorp board and also had it running at 2x50 MHz without much trouble, so it seems that particular CPU is fairly stable at that speed at both voltages.

What's weirder is that I I had my Intel 486DX-50 running in the latter board as well, so apparently that's fairly tolerant of undervoltage. It also POSTed with the DX-50 at 60 MHz but I didn't attempt booting at that speed. It'd be surprising of the DX50 proved stable at that low a voltage, let alone overclocked.

I don't know where the voltage select jumpers are on these boards, or the CPU multiplier select jumpers. I've only worked out the FSB speeds so far and the Acorp board has the multiplier set to 2x (or what corresponds to 2x on Cyrix DX4 processors).

The SiS 471 Soyo board may not have a multiplier select or may do something odd. All the CPUs I've gotten to POST in it so far are 1x or 2x multiplier ones and the DX4s won't work (and AMD DX2-80s too, which I've since realized default to 3x when left floating).

This includes CPUs that have worked in other boards at 5V, albeit Socket 1 PGA-168 type boards.

Well, before posting this I just pulled the Acorp board out and tried the DX-50 again and, provided with a 90 mm fan blowing on it (no heatsink at the moment) it seems to go fine at 60 MHz in DOS. (using an ISA IDE card and ISA video card at 1/3 PCI bus speed, so presumably 20 MHz)

Except doing a bit more testing with a Cyrix 5x86 100 installed revealed more weirdness where CHKCPU was reporting 60 MHz (2x30 MHz) where it should have been 2x50 and another try at 33 MHz FSB gave 33 MHz with 16.7 MHz bus speed. Speedsys scores roughly matched those, too.

So I closed the Turbo jumper and that solved things for the 5x86, which apparently runs at 120 MHz OK at 3.458V. Though I'm not sure what to say about the DX-50 as it was reporting 33, 50, and 60 MHz in CHKCPU. (as internal speed and external bus speed)

The turbo jumper might also be responsible for some unusual performance of other SiS 496/497 boards (including slow memory throughput).

So it's probably always good to try toggling that on a board. (fast mode on my 386SX-40 PCChips M396F also works with turbo closed)

Turns out the Cyrix 486DX2-80 with integrated green anodized heatsink is a later stepping than the IBM DX2-66 and ST 486DX2-80.

Model 1Bh
Stepping 34h

This one and both the IBM DX2 and ST DX2-80 didn't want to POST in the Acorp board at 3.45V even down to 25 MHz FSB, so they may actually be 5V parts. That or something else weird is going on.

That Intel DX50 OTOH does run at that voltage and I tested it again with the turbo switch jumper closed. It seems totally fine at 50 MHz and marginal at 60 (CHKCPU reads 61 MHz). It doesn't get very warm, but to do much testing I still had to have a fan blowing directly on it (no heatsink, just fan). It ran several benchmarks OK, but wouldn't do PC Player without crashing part way through and Quake timedemo could get all the way through the demo, but crash and exit back to DOS (or to Phil's DOS Bench 1.4 menu) just before the score could be displayed (immediately after that last grenade gets fired and explodes). I tried a few times and it did the same thing there.

It'd probably do 60 MHz fine at 5V and maybe 66, though I don't think this board has a 66 MHz setting.

The Cyrix 5x86 100 seems to do fine at 2x60 MHz as well and got some really high scores in cache speed and memory bandwidth (plus 14.8 FPS in Quake timedemo when using a Voodoo3 PCI card, also S3 virge but with corrupted colors due to the 60 MHz bus: my Rage 2+ was fine but a bit slower at 14.1 FPS). I have a Virge DX around somewhere that might be more suitable. (presumably most GPU cores that got used on AGP cards should be fine at 66 MHz PCI ... though I now wonder if ATi's older Mach64 based cards are better on high PCI clocks than some competitors: might make up for the relatively slow performance in some VGA tests)

Oh, in VESA/SVGA modes the Virge 325 (Diamond Stealth 2000) didn't seem to have issues, but in VGA modes it did, both unchained and 13h. Might be the way the palette/CRAM is handled. (PCPlayer 640x480 and Tomb Raider at 640x480 were artifact-free)

PC Player 640x480 got 9.2 FPS with the Voodoo 3.

3DBench got 90.9 in all 3 PCI cards tested.

The internal cache of the 5x86 also appears to be nearly 4x as fast as the Cyrix/ST/IBM DX2 and DX4 at the same clock speed. That may be one of the major areas it gets a performance boost even without other enhanced features enabled. The external memory performance is a little faster, but Cyrix 486s already seem to be on the top end of things there.

I'll need faster RAM to get through all the benchmarks at maxed out speed at 60 MHz fsb, but at 50 several 60 ns sticks I have work fine and the L2 cache seems to work at fastest timing settings. (though I'm not seeing a difference in speed between 1 and 2 cycle burst settings)

As it is I can get most benchmarks to complete, including Speedsys at fastest 60 MHz settings, but the extended memory test fails. Still, 201 MB/s bandwidth in Speedsys is pretty fast, or 167 MB/s at 2x50 MHz, where it's stable.
Having the L2 cache enabled in write-through mode maintains the same DRAM speed as with it disabled entirely, but Write-Back mode kills DRAM speed by about 3x. (at 2x50 MHz, Cachecheck was getting 15 us for DRAM block reads with write-thru or cache off and 44us with write-back enabled)

Looking over my photos again, I missed one configuration where the DRAM appears to be in the fastest settings at 60 MHz FSB along with L2 cache enabled, but does pass Speedsys's XMS test.

The DRAM is still set too fast for Speedsys or Cachecheck to specify L2 cache performance, but you can compare the values in the charts and graph at least.

I also don't see an option in the BIOS to set L1 cache policy (unlike the Soyo SiS 471 board) just the L2, but I guess that's something Cyrix register utility tweaks could also do. Except the latter isn't so useful if a given CPU needs to be set to WT mode to be stable at given settings and be able to boot DOS. (which that Cyrix DX2-80 needs to run at 100 MHz and the IBM DX2 is also happier with)
I guess that's one good thing about the boards that set WT mode on the L1 by default in the BIOS. (or rather, don't know to set WB mode, since I assume the CPU initializes in WT)

The SiS 471's performance also seems pretty close to identical to the 496 chipset both for DRAM and cache performance. (comparing the 496 with L2 in WT mode)

I also previously ran that same Cx5x86 for a while at 5V at 3x33 and 3x40 MHz in a Symphony chipset 386/486 board and it didn't seem to have problems, but obviously long-term use might be a different matter. I had a 40mm heatsink stuck on with thermal paste and a 90mm fan blowing on it then and ran it in 5-10 minute bursts on and off over a little more than an hour. It obviously doesn't need 5V to run at those speeds.

The test results obviously aren't going to be competitive with the SiS 496 board (aside from L1 cache) though the L2 cache at least appears to be in write-through mode since disabling it doesn't change DRAM performance and the cache itself is faster than the OPTi 495SX board's. It does seem more finicky about overclocking the same 5V CPUs I'd tried in the OPTi board (like the AMD DX2-66 I had at 80 MHz without trouble). I think the OPTi board actually had faster DRAM speed with the L2 cache disabled, but I'll need to compare notes there.

So what usually is the first sign of a dud 5x86? Iv been skimming this thread trying to get into the Socket 3 fastlane myself and while i can get my ADW-133 to work fine at its rated speed, i can't for the life of me manage to get it to fully boot DOS at 160mhz. Not even from a bootdisk. It just stops at "Loading MSDOS..."

The only way i managed to get it fully into DOS is by disabling L1 cache, wich tells me something isn't happy at the OC speed. 150mhz wouldn't even post on this board (LS486, its the bigger motherboard variation. I think the original?)

Sometimes it's also motherboard dependent. (at the same bus speed and voltage settings)
Also you can try different DRAM and cache timing: just because a configuration works for one CPU doesn't mean it will work with another.

Then there's voltage settings: you might need a little more than 3.45V to get the AM5x86 working right at 160. (like 3.6V if the board supports it) Some boards support 4.0V without a wide range of other steps up from 3.3 or 3.45V and that may work as well with sufficient cooling. 5V seems to be a bad idea with the AM5x86, so avoid that. (it might work for a while, but tends to be in permanent damage territory if used for longer periods) Same voltage concerns would go for the similar .35 micron AMD 486DX4 CPUs. (I believe that's all the 16kB write-back cache versions)

Rather than disabling L1, switching from Write-back to Write-through cache policy can help overclock, but not all BIOSes give a user selectable option there.

As for trying 150 MHz, try setting the slowest RAM settings and disabling the L2 cache while at 133 MHz and then switching to 150. Also make sure you've got the CPU set up to 3x multiplier (try it at 3x33 or 3x40 first) and aren't actually using 4x50 MHz. Very few AM5x86s will run reliably at 200 MHz, and most probably won't even post at that speed at stock voltage.

Once you have things working with L2 cache disabled, you can start working up through faster RAM timing settings and enabling cache. Sometimes you can get away with faster DRAM timing when L2 cache is disabled and then need to slow it down slightly to get things working right with cache enabled. In my recent testing, most cases where Speedsys checks out OK with the extended memory test also end up working fine with L2 cache enabled. (note a configuration on the edge of stability might test good at first but fail if things are running for a while and get fully warmed up: or end up warmer inside a case then on an open test bench or end up in a warmer room)

If you have a bunch of different RAM sticks to play with, even at the same rated speeds, swapping those around can get you past the flaky stability range too. (like playing around with 60 ns modules to get into the range normally requiring 50 ns ones)

It seems like AMD and Intel 486s have pretty close performance, though the Intel DX4 with 16kB WB cache might have more of a distinction the 8kB models are pretty close (the 5V DX2 CPUs I have ended up with nearly identical scores across the board, including in Doom and Quake) and the enhanced 8kB models seem pretty similar as well.

They all seem to have faster L1 cache performance than Cyrix 486DX2 and DX4 CPUs but slower external memory (cache and DRAM) performance as well, so actual benchmark scores between the Cyrix and Intel/AMD 486s ends up being a really mixed bag.

At the same clock speeds, the AMD and Intel 486s end up significantly faster in Doom and Quake in my tests and I'm going to bet that's related to the L1 cache speed and maybe some 486 and Pentium assembly language optimizations that don't cross over to the Cyrix 486.

For example, in my Acorp SiS 496 PCI based board with a Rage II+ (PCI clock at 1x) and both CPUs at 2x50 MHz, L1 in WB mode L2 in WT mode, cache and DRAM all at fastest settings and lowest waitstates the BIOS allows.

Cyrix (ST) 486DX4 gets
Doom
2097 realticks
Quake
9.7 fps

AMD enhanced 486DX4 gets
Doom
1863 realticks
Quake
11.5 fps

3Dbench ended up at 66.6 FPS for both CPUs, but PC Player and Landmark both ended up slightly faster for the Cyrix 486 as did the Speedsys scores. (except for L1 cache speed and read speed in the first 8kB range)

PC Player ended up being 17.3 vs 18.0 (320x200) and 7.3 vs 7.5.

CPU score for Speedsys was 37.28 vs 40.08

Without any register enhancements enabled, the main speed advantage of the Cyrix 5x86 over the 486 seems to be the larger and faster L1 cache. I suppose I could disable the L1 and see if I ended up with identical figures there, though that might not show much difference in the numbers even if there is one. (given how much performance tanks with L1 disabled)

The 5x86 (without register enhancements enabled) manages to beat out the AM486 and AM5x86 clock for clock in Quake performance but ends up still loosing out when compared relative to advertised PR numbers, even with an FSB advantage. Ie at 2x50 MHz the Cx5x86 ends up slightly behind the 4x33 MHz AM5x86 in Quake and only slightly pulls ahead at 2x60 MHz and then falls well behind the AM5x86 at 4x60 (at 14.1 vs 16.7 FPS using the Rage II+ ).

That particular AM5x86 is only moderately stable at 4x60 but I'm only using 3.458V. It eventually hung/froze during testing, but did well enough to get a fair chunk of results.

Incidentally, that AM5x86 is the one with engraved markings and the AMD 486 DX5 printed on it. (it seems like most of the laser engraved variants also have the 486 moniker on them and none of the white print marked ones have 486 included on the labeling)
Like these examples:
http://www.chipdb.org/cat-133-22.htm

So I guess mine is probably not a remarked chip as I previously suspected, though the lack of black ink in the laser grooves still seems odd.

That particular AMD DX4 also seems to run fine at 2x60 MHz and 3.458V (like the Cx5x86) though the above tested ST 486DX4 would only POST at 2x60 and only when cold, freezing by boot time (even with slower memory, no L2 cache and L1 in WT mode) so it probably needs more voltage to achieve that. It runs fine at 5V and 3x40 MHz in the OPTi board, though obviously is pretty bottlenecked there by the slow chipset. (it probably doesn't need nearly 5V to actually run at 120, probably 3.6-4.0V, but it has tolerated over 100 hours at that setting so far without obvious ill effects ... I'd say a few hundred hours at this point, but I haven't been keeping close enough track to be sure)

Also, one disadvantage of the AM5x86 (and DX4-100 models using the same core) is that there's no 2x multiplier, so you can't try 2x60 or 2x66 like you can with older models (and with Intel and Cyrix DX4s). Though there's 3x and some will do 180 MHz with some overvoltage and 150 MHz much more reliably. They can do 3x66 in rare cases.

While 5V will probably kill them, I think it might be possible to get away with something above 4.0V but below or just at the absolute maximum rating in the datasheet. Chips are normally supposed to resist permanent damage if run below the absolute ratings and the .35 micron AMD parts are listed at 4.6V absolute maximum, so if a particular board has a 4.1 to 4.5V setting (and you verify this with a multimeter at the socket) it might be worth trying.

OTOH higher voltage doesn't always help stability (and can sometimes cause erratic behavior), so backing off to the minimum voltage a chip will POST (or BOOT an OS) at could be a more useful bet.

The die area is also very small on the .35 micron AMD parts, so applying thermal compound right around the center of the package is more important for good contact (and avoiding lumps near the outside that might cause poor contact at the center).

Cyrix's datasheet on the 5x86 lists 4.0V as the absolute max for supply voltage (and 6.0V absolute max on any single pin: I think that means on supply or I/O pins) though does cite that the CPU is 5V tolerant I/O as well. (which would be relevant on socket 3 boards with split-rail supply and I/O voltage arrangements: for 5V RAM and 5V TTL compatible signaling in the chipset and peripherals)

The 5V I/O tolerance might be part of why my Cx5x86 at least seemed stable at 5V up to 120 MHz, though might eventually die at that setting. (it also POSTS and tries to boot at 150 and 5V in the OPTi 495sx board, but didn't get to DOS and I stopped trying after a few minutes of fiddling with it) OTOH Feipoa didn't have good luck getting his 5x86 to POST in a 5V board.

The document I'm looking at also has the Cx5x86 rated for 110C while under load, and that's really hot for a typical CPU and more in the range of

I'm also not using big Socket A style heatsinks with my 5x86, but am using large and/or really aggressive fans and open case or no case testing set-ups. (the generic chinese 40 mm fans on ebay seem much more aggressive and high flow than the 37 mm fans on the new old stock 486 clip-on type coolers I have (and the one in my Dad's old 486 cooler, though that one has worn out bearings and doesn't spin up as fast as it should: otherwise I think it'd be a bit more powerful than the old stock ones I got, though the heatsink is also shallower on that one so it might need more flow).

Since those clip-on type heatsinks are useless (as far as actually clipping on) to QFP mount CPUs that's also less relevant anyway. I've been using 40mm black anodized aluminum heatsinks (more, generic modern ones, probably mostly aimed at router/RasPi/set-top-box type embedded stuff, but they're about the right size for 486s, including on boards with little room around the socket, plus good for some 386 socket arrangements)

I haven't seen any heatsink/fans that actually make use of the Socket 3 external mouting clip points and those certainly aren't Socket 5/7 compatible. (Socket 4 also seems to have clip-on points, but all the Socket 4 coolers I've seen are CPU-clip-on types as well)

Given the lack of secure mounting pressure (or sometimes no pressure at all) the clip-on types have, you need fairly sticky thermal paste to hold it together. (it actually seems more like the clips are there to prevent the heatsinks from sliding off the CPU in a tower mount orientation while assuming you've got thermal compound making the actual contact: and while hot, the thermal paste tends to be fluid enough to allow stuck-on heatsinks to gradually slide off without clips or such, though in a horizontal case or open testbench, that doesn't seem to be a problem)

In my experience it's pretty hard to get Cyrix 5x86 chips to POST at 5V. I have 5 or 6 of these chips, and I don't think any of them would do it. 150MHz on a Cyrix 5x86 is not realistic at any voltage with L1 cache enabled. 133MHz is pretty much the best you can ever hope for with this chip. On the other hand, almost all of my -100 chips seem to run fine at 120MHz.

I also had pretty much zero luck getting am5x86 to run at 180MHz with L1 cache enabled. The best one I tried was a laser engraved 16BGC model made in 2001. That one would boot into NT4, but the system would crash after a few minutes. I also tried about 5-6 of these CPUs. I have a trim pot to vary the voltage, and I use a 1.5cm thick heatsink with active cooling and thermal grease. Almost all of my am5x86s can run at 160. The only exception is my ADZ, which crashes just after POST (even with modified voltage). In my opinion, if you can't at least get to the DOS prompt at 160MHz with stock voltage and active cooling your CPU is probably subpar.

I should note that both the boards I had my 5x86 running at 5V are 5V only Socket 1 boards and both are 386/486 combo boards, at least one of which (the OPTI 495SX based one) is also specifically intended for Cyrix 486DLC CPUs.

Turned brass/bronze type connectors on a non-ZIF type socket.

The pins on the PCB adapter the QFP chip is mounted on also seem slightly thicker than normal ceramic 486 pins. 3 of the 7 5x86s I got as dead/scrap damaged CPUs would POST on the OPTi board, though 2 I didn't test as the leads were too crushed together to safely separate and might need to be desoldered to straighten and/or have lifted pins that are required to work or didn't work for other reasons. (I could try them again at 3.46V, but had set them aside intending to work on them when I get a hot air rework station probably along with some better precision tools: I was mostly working with tiny flathead screwdrivers and needles to try and nudge crumpled QFP leads apart)

I think the lot came from CPUs that had been stacked on top of eachother in some container and had something heavy put on them at some point, maybe new old stock or maybe just sockpiled socket-pulls without noticeable wear on the pins.

Both of those 386/486 boards also measure at 5.0V at the socket and on the ISA slots. Between 5.01 and 5.04 and well under 5.1V. (I didn't invert the board and measure the socket pins with a CPU installed, but did test the 5V rail on the ISA slots while a CPU was installed) So not being in the 5.1-5.25V range might also be part of the stability there. (which is still within the spec for 5V devices and unfiltered 5V output of AT/ATX power supplies is often in the >5.2V range)

I just figured out the 5V jumper on my Acorp SiS 496 board and tested the socket at 5.11V (or 5.109V) and I don't have a large sample size here, but if other boards are more like that it might be pushing things further over the edge into erratic operation territory.

I also haven't found a jumper block that allows other 3.0 to 4.0 voltage settings if there is one present. I recall seeing separate blocks for the 3V CPU range on a Taken or Tomatoboard manual and quick reference card I was looking at recently, and I think the 5V setting was a separate jumper while the other voltage settings were done via a block of 2 or 4 jumpers.

The individual 5x86 I've been doing testing with might also just be a particularly stable, overclockable example that avoids erratic behavior when overvolted since it seems to overclock to 120 MHz (2x60) at just 3.46V and just happens to still run fine at that speed when overvolted.

I did actually have the Cx5x86 up to 131 MHz (using an 87.5 MHz oscillator) in the Symphony 386/486 board. But that board itself is a pain to get working at that speed with any CPU I've tried and it's ISA only so actual testing was limited and additional cache wait-states had to be enabled via one of the 486DX-50 specific jumper settings printed on the board. (there's 3 DX-50 jumpers, but I can't seem to get anything to POST when all 3 are set and the one related to the cache is the only one that seems to help stability

Oddly enough, while I couldn't get a 386DX40 working at that speed, but I did get a TI 486DLC-40 to run at that speed without much issue beyond some quirks I got with 486DLCs at 25/33/40 MHz: ie Speedsys froze on all of them during initialization. (the Cyrix/IBM DX2s that will run at 2x50 in other 5V boards also didn't want to POST or at least complete POST without freezing at ~44 MHz FSB) I havn't tried the 486DX50 again in that board but I imagine it would be the best for troubleshooting/testing that setting, but it also might just be one of the many 486 boards poorly suited for >40 MHz. (the OPTi one seems fine at 50 MHz other than the 20 ns cache being too slow to cope)

In any case, all the ST 486DX4s I have (all from the same chinese seller) will run at 5V though that one I've been testing the longest seems to be one of the most stable as well though won't POST at 150 MHz (while the one that will post routinely at 150 MHz isn't all that stable at 120 or even 100 MHz ... there might be something else wrong with it or it just doesn't like 5V or >33 MHz FSB since 100 and 120 at 3.46V were done at 2x multi).

That one ST DX4-100 I've been running for hours at 120 MHz in the OPTi board still runs fine at 2x50 MHz and 3.46V after the overvoltage abuse but will only boot DOS when cold and then crash/hang shortly after when at 2x60 in that Acorp board and 3.46V.

However, I tried it again at the 5V setting in that board (or nominal 5.11V) and it's quite happy maxed out at 2x60 with the same high memory bandwidth figures as the Cx5x86 at the same speed. So it seems like the external I/O section of the Cyrix 5x86 and 486DX2 didn't change at all. Performance also looks identical to the Cyrix/IBM/ST DX2-66 and DX2-80 from what I can tell. (and all of them seem to have the same L1 cache speed disadvantage to the Cyrix 5x86 at approximately 37.5% of the read speed in cachecheck)

I assume the SGS manufactured 486DX4 (as well as the TI ones, unless they're rebranded STs) are on a different manufacturing process than the earlier DX2s but it wouldn't be that much of a stretch if they stuck with the same process after it had matured and yields were good. The change to add a 3x PLL (and simple logic for the 2x and 3x select) would be a minimal change to the masks.

If it is the same process used with the 5V marked ST 486DX2 that makes it less surprising that the DX4s run fine at 5V.

CHKCPU does give the same model code for all the Cyrix/ST/IBM 486DX2 and DX4 CPUs I have, but with 3 different steppings.
All are model 1Bh and the DX2s are 32h and 34h while all the DX4s are the same 36h.

And I believe the 'h' means those are hexadecimal numbers (1B and 32/34/36 in hex) which would translate to:
Model 27, steppings 50, 52, and 54.

However I don't think that necessarily says anything about which manufacturing process was used without further information.

Also the Intel DX-50 seems fine at 5.11V and 60 MHz. Interestingly, I can't get the Pentium Overdrive to POST at 60 MHz FSB but it's still fine at 50. I'd thought running it at 3.46V might have been the problem with it before (ending up underpowering the regulator) but maybe it's sensitive/limited on the I/O bus end of things. It's still stuck at the 1x multiplier unless I can rig up a compatible tach-fan installation for it. (I was honestly hoping for 1x60 or 1x66 to compare directly with that DX-50, Socket 4 Pentiums, 5x86s, and Socket 5 CPUs with 1x multipliers: Cyrix M1/6x86 and some uncommon models of AMD K5, at least some 5k86 marked ones)

None of my 6x86 CPUs seem happy at 1x100 MHz even though they might overclock well (PR-200 6x86Ls seem to do 3x66 reliably if run at 3.3-3.5V) though I don't think I've seen one that won't run at 1x83.3 MHz. (including the single PR-90 I have)

Oh right, and I got an Intel DX2-50 to run at 2x40 MHz without apparent issues. (and a Quake score of 9.6)

Anyway, have some pics of that ST DX4-100 @ 2x60 MHz and one of the the Cx5x86's cachecheck tests for comparison. (it gets detected as a 486, but it's the one with the 4 us L1 cache scores)

And for what it's worth, here's some of my results from by brief testing of the Cyrix 5x86 at 3x 43.75 MHz (87.5 MHz oscillator installed) at 5.0V in the Symphony SL82C461/82C362 cipset board. (with 15 ns cache chips and 16 MB of 60 ns 30 pin SIMMs installed)

Though I also discovered 4 or 8 MB of 70 ns SIMMs (4x or 8x 1MB) I had on hand also seemed to go OK at the same settings (2 wait states, which is the max).

OTOH it was pretty easy setting up the 5x86 at 3x40 MHz in that board and with the faster cache and RAM timings, partially due to the known-good system stability at that speed ruling out other variables, and was really faster than at 131 MHz anyway. But still obviously slower, probably more bottlenecked by the ISA video card than the RAM throughput. (in fact, RAM speeds were pretty close to the same as what the SiS 496 gets in WB L2 mode)

Update summary:

Acorp SiS 496:
ST 486DX4-100
@ 133 MHz (2x66) 5V

Cyrix 5x86-100GP
@ 133 MHz (2x66) 5V

Other 496 board:
AMD 5x86-133 P75
@ 180 MHz (3x60) 3.98V or 3.68V

AMD 5x86-133 P75
@150 MHz (3x50) 3.30V

A couple nights ago I figured out the 66 MHz setting on that Acorp board. (turns out it's just all 3 jumpers enabled; had to use the DX-50 to test it, and it does 1x66 at 5V with slowest cache settings)

I have another unknown (and apparently older) SiS 496 based board that I figured out voltage settings on with a jumper block. It seems to do 3.30 (measured 3.298), 3.40 (3.39-ish), 3.51-ish, and 3.98 (which I assume is the 4.0V setting) when using a single jumper on the block. (4 pairs of pins) Using 2 jumpers on the block gives up to 3.68V and capping all 3 lower jumpers probably outputs 3.7-3.8V but I forgot to try that. Having the 4.0V jumper set outputs the same 3.98V regardless of other combinations.

There's probably a 5V bypass jumper, too, but I didn't find it.

I did, however work out the FSB jumper settings, which appear to have 3 active jumper locations like the Acorp board and a 4th that doesn't seem to do anything. (may just be to hold an extra jumper) I worked out 25, 33, 30, 50, 60, and 66. 66 is all 3 jumpers closed like the Acorp board. I think the remaining 2 combinations were redundant or wouldn't POST. (there may be a 30 MHz setting on the Acorp board I missed, but I only found 25, 33, 50, 60, and 66 there along with no-POST combos and a second 33 MHz setting)

It also has a 3-pin turbo button header, but the third pin doesn't seem to do anything, at least when jumpered to the middle pin (behaves the same as when open, which is slow, and fast mode is pins 1+2 closed, like the Acorp board).

The board is set to the 3x multiplier and doesn't seem to want to work with my ST 486DX4 or 5x86 properly. They POST and enter BIOS but the 5x86 showed no L1 cache when booted (and I tried L1 enabled and disabled in the BIOS, and tried write-back mode but the system refused to boot at all then) and the DX4 wouldn't boot DOS at all, or wouldn't even get to the POST table.

It may be an issue with pins contacting the socket, but it seems more like a BIOS issue. It's a 1994 dated BIOS, which I think predates both the Cx5x86 core and ST's DX4 derivative of the Cx486DX2 so there may be some issues there. That or something else weird is going on. (it definitely recognized the Cyrix CPU as Cyrix and had options for WB and WT L1 cache as well as burst-write enable, but there may be an issue with using the 3x multiplier ... and the only options for CPU selection were Auto and Cyrix with it installed)

The AMD 5x86s I tried worked fine, though, aside from only allowing the cache in WT mode. Actually, more than fine: the same one I'd been running at 160 MHz (4x40) at 3.46V without issue runs at 180 MHz (3x60) at 3.98V and seems to still be fine when backed off to 3.68V, maybe lower but it might be ambient-temp + cooling capacity dependent and I don't think 3.51V was quite enough. (I did testing late at night or early in the morning when things had cooled off here given the heat wave we're having and lack of air conditioning)

The memory/cache performance wasn't all that impressive and there's only a single bank of 128kB cache in this board (and I'm not sure what the jumper settings are for other configurations) but overall it was still quite fast. I believe the best Quake score I got was 17.5.

When totally cold it will post at 3x66 MHz, gets to the POST table, then hangs when trying to boot. As soon as it's warm it won't post anymore at 3.98V and I'm not sure it's worth risking it at 5V even if/when I figure that setting out.

I backed it off to 50 MHz FSB and that 5x86 seems totally happy at the 3.3V setting and 150 MHz (might need a little more voltage if running in a hot room and/or poor case airflow) and I could max out most or all the cache and DRAM settings at that speed. In spite of the single bank of 15ns cache it seems to do OK with 2-cycle write, 2-cycle read, and 1-cycle burst-read settings.

The often finicky sound engine of X-Wing (floppy) also went fine at 3x50 MHz with the fast settings, but see below on that.

The Acorp board, in addition to running the Intel DX50 at 66 MHz also will run the ST DX4 at 133 MHz (2x66) at the 5V setting and I used it as such for quite a while. Then I decided to risk the 5V setting with the same 5x86 I've been testing with and it too runs (apparently) fine at 2x66 MHz.

I think having the 90 mm fan set next to or on top of the CPU heatsink and also blowing on the 496 helped stability too as the chipset gets quite warm at 66 MHz as well. (I should probably rig up some legs to mount through the screw holes on that fan to use more easily for set-ups like this)

I'm not sure how long it would last long-term at 5V (or 5.1V in this board) even with adequate cooling, but I ran it long enough to get a good batch of tests and some gameplay in. 150 MHz still wasn't stable enough to boot from the little I tried. (and the added heat output seems riskier there combined with the voltage)

I'm not sure this board even has selection for other 3.xV values but I suspect 4V might not be enough to do it and might be somewhere closer to 4.4-4.5V but I assume 5V is both overkill and probably degrading stability at least to some extent. (4.4-4.5V would also be nice to try with that AM5x86 for 200 MHz)

At that point I went back to 120 MHz at 3.46V with the 5x86.

I then discovered that, as with many configurations in my Socket 7 system (and faster ones, like PIIIs), X-Wing's sound doesn't work or is totally garbled. I thought it might be the 15 MHz ISA bus (60/4), though that's not a problem with slower processors ... and the asynch 7.16 MHz ISA setting made things even worse. (and no 1/5 or 1/6 ISA dividers ... unless the PCI divider is also set which I both haven't found and don't want to use since the Rage II+ is fine at 66 MHz)

I, this problem remained all the way down to 2x40 MHz with a 10 MHz ISA bus regardless of wait states and memory/cache timing (short of disabling the L1). It took 33x2 to cure that issue.

So I got the idea to try the exact same test with the ST 486DX4 at 133 MHz ... and it works flawlessly at the fastest settings I could already get DOS working in (which are close to the same as the 5x86).

I'm now convinced that X-Wing (and maybe some other 16-bit real-mode variations of the 1991~1993 vintage IMUSE MIDI engine, maybe Monkey Island 2) has issues with very fast L1 cache speeds. That would explain why disabling L1 usually solves the problem regardless of ISA and FSB or board level cache speeds and why the problem doesn't usually manifest in 1x multiplier CPUs in my Socket 7 testing up to 83 MHz or with 2x for most CPUs I tried up to 75 MHz FSB and beyond that things get CPU-model dependent. Albeit I had more than just sound issues there, but random game crashes (those turned out to disappear when running a Virge 325, Rage XL or Pro PCI or AGP card, but didn't like hot-running Voodoo 3 or Savage 4). The main sound issue was just no sound at all.

Disabling L1 nearly always fixed all of X-Wing's problems, but also runs the game a low slower than optimal ... especially with higher settings. Granted, the Collector's CD-ROM version avoids most of these problems (and you can import/export save files easily) but it's fun to try the original 16-bit version ... and in a different sense than getting it working on a 286 build. (ie mostly playable and often single-digit framerates at minimum settings and massive slowdown on the scaled bitmap explosion effects vs: solid 70 FPS and no screen tearing at maxed out settings)

In any case, X-Wing (or its sound engine) does work fine with the 16.67 MHz ISA bus and 66.67 MHz FSB settings with an ST 486DX4 and with an AMD 5x86 up to 150 or 160 MHz at least, but doesn't seem to like a 5x86 at 80 MHz or more. And it may also have to do with the speed difference between external and internal RAM access throwing off some sort of timing related code. (given fast 1x SS7 multiplier settings avoided this and I'm pretty sure the 6x86 cache is at least as fast as the Cx5x86)

And given the Cyrix 486 core seems to perform very similarly to the 5x86 at default register settings with the sole exception of the much faster 5x86 cache, while the AMD and Intel L1 caches are somewhat faster than the Cx486 but much slower than the 5x86, it seems likely that it's cache-speed related.

None of the other ISA-bus sensitive games have problems with the 5x86 at 120 MHz (Doom has sound issues sometimes over 16 MHz, and Jazz Jackrabbit usually won't run at all) seemed to have issues with the Cx5x86 at 60 MHz like X-Wing does. Tie Fighter didn't either. However, none of them seems to have problems at all with the ST DX4 at 133 MHz.

So I guess that ST/Cyrix DX4 at 133 MHz is my fastest 'pure' 486 so far ... though honestly, the behavior of the Cyrix 486 family is pretty different from the AMD and Intel 486s, closer to the distinction you see among Socket 5 and 7 CPUs.

I'll have to try the (I believe .5 micron) 8kB cache AM486DX4 at 3.98V and 3x multiplier later on: maybe 3x50 or maybe 2x66 at 5V, but it didn't seem happy for the little while I tried it in the Socket-1 5V boards, same for the DX2-80 I tried, which runs as a DX4 at 3x by default: ie no multiplier jumper set. (AMD 486 models 25398 and 25253)