@pshipkov
I haven't tried the Diamond v2.09 BIOS but re-checked and soft reset works with the #9 BIOS.

Boca Research's PCBs are nice too, even for their lower-end cards like this GD5426.
Almost like STB's, but not quite to that level of design.
Number Nine seems to be the pinnacle, for the most part. Miro/ELSA are excellent too.

Ah, I see. ALi Panda is an interesting one.
Curious about the M1429 chipset for a 386.
I have one of these boards in a Leading Edge 486D, but have not used it much.
Came with a 486DLC stock. Seemed pretty speedy, even with the old HDD and Win95.

PCChips really made a monstrosity there!
Those are indeed VLB from what I have seen.
I guess only 2-3 were meant for use at a time?
Curious about HiNT CS8005 chipset. HiNT P-EISA is not a bad performer, from my testing.
2.5MHz might be generous if all slots are filled 🤣
Not sure where this photo is from. Usually TRW guys put a name in the bottom right. Probably an old eBay listing?

I think EISA is definitely transitional, but the boards equipped with EISA were often overbuilt.
Generally, it seems they were server/workstation grade. Built for speed, stability, and power.
EISA itself is quite nice. Plug and play never works right on ISA, but on EISA it's all easy. Plug in a card and add it to the config.
VL/EISA is the ultimate combo. Boards like AnonymousCoward's Acer J3 are where things get interesting.
EISA is quite fun. I see you have an ELSA Winner 1000 XHR. It's interesting to use one in an EISA board.
On an EISA 486? No difference/minimal difference. On an EISA Pentium? I saw massive gains.
EISA is good for "wide" bus transfers; 10/100 LAN, SCSI, etc. With proper busmastering, it is quite nice.

Don't know if you've seen this, but found this interesting doc re: WB L1 support on chipsets.
Geared mostly toward Cyrix/TI/IBM/ST 486/5x86C, but applies pretty universally to Intel/AMD.
Here's another with more chipsets w/r/t the IBM 486DX2.

I remember in the late 90s or early 2000s seeing pictures of that PCChips DX 9000 board with the 8 VLB slots, but for many years I was convinced it was photoshopped until I found it in TH99.

About the Boca VLB video card you linked - really like the sparsely populated late design like that. They look unusual (in a good way, to me at least) with these big empty areas.

M1429 for 386 CPUs is ok. It won't blow your mind, but is above the average for sure.
It will be up to Feipoa to provide some concrete metrics once he is done with trenches around his house.

Old industrial hardware components like the 8xVLB PCChips motherboard get scrapped on a common basis when their time is up.
Little chance we get to try one of those. Hope future proves me wrong.

Thanks for the chipset links.

---

Adding one more 386 motherboard to the list.
Biostar MB-1325VT/1333VT version 2, based on VLSI VL82C33x-FC chipset.

This chipset is usually present on SX boards but this was a DX implementation. That piqued my interest.
It was obvious that it will not be a great performer given the lack of L2 cache but wanted to see what's there anyway.

BIOS is full of options. They seems to be set to their optimum values.
Any attempt to further reduce timings results in no POST completion.

Things work fine at 40MHz.
Board can overclock to 45MHz. Becomes unstable after that.
Performance is really bad - running tests took forever.
Didn't bother capturing exact numbers - they won't reveal anything interesting.
Took couple of screenshots from quick synthetic tests for reference.

A slightly unusual piece of hardware which is interesting for what it is, but that's everything about it really.

@pshipkov
Agreed. "Minimalist" PCBs are interesting for this era.
I appreciate seeing fewer tantalums needing replacement 😀

Interested to hear his results. I may have to look at my ALi Panda.
Seems like a nice assembly.
There are 1-2 other VLB-only 386 boards I've seen, all using the M1429.
I guess the Panda is a good representative for that class of hardware.

Agreed. Hopefully one of those monstrosities turns up.
Like @AnonymousCoward, I thought that was a photoshopped image the first time I saw it.
PCChips never fails to amaze.

------
Intriguing 386 board.
Curious as to why it was implementing an SX-oriented chipset.
I have seen the reverse w/ a 386SX and MX chipset. That makes sense.
Nice looking assembly, shame on the metrics.
Synthetics look like this board is at the limit of what it can do.

Seems to be some variant of the VLSI TOPCAT minus the data buffer?
Not sure. Curious if any of the MR-BIOSes help with performance.
According to catalog, should be 301 or 302.

-------
@feipoa
Update re: Diamond BIOS v2.09 and S3 Trio64 -P variant.
Mine indeed does NOT allow for soft resets with this BIOS.
The -P variant of the chip doesn't like BIOS v2.09 and soft resets, as you suspected.

------
Updated the SuperEISA post with more VLB cards.
Going to add some visuals as well.
Still need to unbox Mach64 VRAM for addition.

I kind of started watching the EISA space now.
At least 15-20 years too late, but still ... : )

Not sure what was the motivation for producing such motherboard.
In fact, i wonder what market would require such slow assemblies and in what volume to motivate companies to make them.

Middle size 386 boards are my favorite, but Biostar offered "the basics only" in this case.
Not sure either about the exact chipset model and stuff. Took it for what it is. Shared the data because i had it.

(btw, another 386 mobo MX chipset will show-up here in the coming days)

I like how your SuperEISA post swell with numbers for VLB video cards.
Small note about Trident TGUI9440# cards. Very fast and solid. Some of the best VLB all-rounders. The AGI especially. Good to see yours kicking a.s over there.
The best part yet - they are still frequent on ebay and at reasonable prices.

Isn't the the 9440AGI supposed to be slower than a 9440 sans postfix?

@pshipkov
EISA is worth watching for sure! You won't be led astray 😀
Lots of fun, and neat parts/cards to check out too.

Sometimes I feel that these slower setups came as generic IBM clone things.
At least, that's how I rationalize it.
Otherwise, why buy such a board? I constantly wonder that about my MX 386SX board.
Why put that DX chipset on a 386SX board? It's a Trangg Bow board, so that says a lot 🤣

Agreed on middle-size 386s. Biostar does have some nice boards, but this is a strange one.
I feel like VLSI chipsets are all about the implementation. Seems like some assemblies are quite fast, even for 386s.

Eagerly awaiting your MX setup! I should break out mine too. They are quite fun! 😀

I will keep adding and make charts once I finish. I plan to do the same for the Chips4041 as well.
The SuperEISA is wonderful to work with. Truly plug and play.
AnonymousCoward is 100% right that it is a beast in all regards. His post is the reason I wanted this board.

Currently awaiting a Miro 20SV S3 964 to add to the test.
May start ISA testing in the meantime.
Picked up a 3COM 3C509B-TPO for comparison with the BocaLAN ISA and VLB cards. That will come very soon as well.

@pshipkov and AnonymousCoward
Trident TGUI9440s are great for sure. Fast, great image quality, and easy to use.
Seems like they vary in build quality. Mine is certainly cheap, but has good image quality.
The AGI is one I know little about. I would like to compare that with a 9440-1. (Don't know if the later -2/3 were offered on VLB).
I notice the 9200 was a CXR, 9400 was a CXi, and 9440 was AGI. I wonder what these all mean?

The AGI drivers do not work with mine. I can't figure out why.
Maybe the 9440-1 and 9440AGI are internally different.
I now want to test an AGI. Guess I found another card to find! 😀
There is also a 9680 VLB. It was Japanese market centric. Wonder how that is?

Different levels of Trident cards are like different flavours of dog poop.
But seriously, I never understood why Trident made so many different versions of their ultra budget ICs.
For a very brief period in the mid 90s...maybe 1994, Trident almost became good! After dealing with Trident 8800, 8900, 9000 cards I had very low expectations for the 9440, but I was pleasantly surprised. I think their 96xx cards for PCI were also decent. I remember reading that the 9680 had the VLB interface, but I could never find an actual card. So that's pretty cool I guess. I hope we can cook up some more if they're good. I honestly didn't even know people in Japan used VLB. I figured they all had PC98s with proprietary interfaces.

From what I understand, the 9600 series uses the same VGA core as the 9400 series, so if you're expecting increased DOS performance over the 9440 you're not going to find it here. I haven't consulted the datasheets, but I'm guessing it's the 64-bit version of the 9440, with a beefier RAMDAC and support for more RAM.

Hi. Mr. pshipkov,
Can you post original BIOS DUMP of Asus ISA-386SIQ ?
Thank you!

@crackintosh
I don't have this motherboard anymore.
Attached you a generic SiS461 BIOS that works just fine on it.
It is identical to the original one really.
Board was tested with it.

@WJG6260
Keep your hopes low for the MX motherboard. It doesn't wow. : )

@Anonymous Coward and @WJG6260
There is a lot of mediocre stuff from Trident for sure, but they had couple of good models too.
It is not obvious at first sight which ones are the good ones in the pile of models / numbers / abbreviations.
From my experience TVGA9000C ISA and TGUI9440AGI VLB are top-notch video cards.
9000 PCI are good too.

I have numbers in previous posts for TGUI9440 VLB on Asus VLI-486SV2GX4 and ProVidia 9685 PCI on Asus PVI-486SP3.
These are the last "classic" 2D video cards from Trident for the corresponding VLB/PCI buses.
Both setups with 0-wait states across the board.
Ran a quick test with ProVidia 9685 PCI on Biostar MB-8433UUD-A rev.3.1 since that board is one of the fastest at 160MHz and to get one more sampling point. Again - 0-wait states entirely.
Couple of numbers for comparison:

1           VLI (9440)  PVI (9685)  UUD (9685)
2Wolf3D:      114.8       104.7       105.8     (fps)
3PC Player:    20.1        24.6         9.4     (fps)
4Doom:         52.2        54.4        52.4     (fps)
5Quake 1:      14.1        16.9        16.1     (fps)
6WinTune 2:  5820        6474        6741       (KPixels/sec)

Numbers are very similar, so you are probably right that they may be very similar inside.
But then most other video cards of that period show similar results. : D

Anyhow, point is that the 3 mentioned Trident cards are pretty good.

---

Micro Express Inc. Forex 386 Cache v3.61 based on Forex FRX46C411 / FRX46C402 chipset.

There is information online touching on Forex based assemblies but without providing concrete metrics.
In addition to that, some people hold them in high regard and their comments make me curious.
Decided to check some of that myself - performance and CPU upgrades support.

Few notes about 386 Forex chipsets:
The 411/402 model is a middle generation.
There are 2 later versions:
FRX46C421A - more integrated than 411.
FRX46C521A - single-chip implementation.
The 521 one is seen on very late 386 boards with integrated clock generator limited to 40MHz which makes them uninteresting for overclocking exercises.
The 421 based PCBs are definitely worth exploring, but for that i will need to find a working replacement for the already dead board here.
With that said - these chipsets seem to have relatively shorter lifespan than the average hardware from that time.
In addition to the 2 more Forex dead boards here (one 411 and one 421, partially scavenged for parts) i often see Forexes on sale for parts online.
Luckily the one in question is still alive and kickin'.

Partially upgraded system with 10/12ns rated L2 cache chips and FasMath gray top FPU, but original CPU and clock oscillator still in place.

Assembly seems brand new. I suppose it was put in a computer and never touched until again until now.
It is kind of nice when i am the first person to remove CPU and L2 cache chips after they got inserted in the factory.
But also, it feels somewhat inappropriate to intrude on the "pristine" assembly.
... did it anyway.

Took some effort to max-out and stabilize the system at 45MHz.
It is on the edge with 32Mb RAM and 256Kb L2 cache.
50MHz is impossible under any circumstances.
Some of the BIOS settings had to be lowered but still performance is still better than 40MHz with tighter timings.
I/O RECOVER CONTROL must be ENABLED or no boot into DOS.

Chipset/BIOS don't recognize BL3 CPUs. Only SXL2 and DLC.
Same applies to Cirrus Logic GD-5434 video cards - instabilities.
A bit disappointing.
Tested with AMD 386DX and TI 486 SXL2 processors, as well with video cards ET4000AX (DOS) and ET4000/W32i (Windows).

Here is what SpeedSys shows for 386DX (incorrectly shows frequency as 50MHz, it is 45MHz):

And some more stats.

SpeedSys for TI 486SXL2 (incorrectly shows CPU model and frequency, they are TI SXL2 at 45MHz):

And some more stats.

One thing is immediately clear - IDE performance is clearly lacking.

benchmark results

Not bad at all.
That motherboard is very capable 386 hardware despite the lack of support for BL3, unstable CL GD-5434 (which is the case with many 386 boards) and an upper limit of 45MHz.
Hope to see one day what FRX46C421A / FRX46C521A can do.

Finding a board with a removable CPU and 400 series chipset seems pretty challenging. I tried to find a 421 board for a while, but the CPU was always soldered.

Agreed.
Earlier boards based on Forex 46C402 chipset version are wide PCBs with CPU sockets.
Later boards tend to be more on the budget side with surface mounted 386DX CPUs.
Have seen only one or two FRX84C421 based ones that have CPU socket.
The single FRX46C521 one i know of has surface mounted CPU.

Wasn't there also a 300 series chipset from Forex, or am I confusing that with OPTi?

This is what's in VOGONS wiki:

36C300/200 [36C300/46C402] WriteThru
36C311 Single Chip 386SX with Cache
46C411/402 WriteThru
46C421A/422 WriteBack
46C521 WriteBack (Single Chip)

I think the first one is the one I'm referring to.

This thread: FOREX 386 motherboard thread (you participated there as well)
It shows Forex 300 based motherboard.

The more interesting part is towards the end where user 386_junkie runs his Forex 421A motherboard at 50MHz with 386DX CPU and most relaxed BIOS timings.
Shared results show lower performance than what i got above with tightest timings and 45MHz frequency.
Hard to extrapolate anything from it, but it feel like 421A is not a major upgrade over 411A.
Searched online around this topic but nothing useful came from it.

@AnonymousCoward and pshipkov
Trident really made way too many ICs to take account of.
There are about 55 different sub-versions for each too! 9440, 9440AGi, 9440-1, etc.
Thing is, I have found that the 9440 is honestly good in all forms as a upper-mid tier card.
Not the best, not the worst. Stable, predictable, and pretty fast.

The 8900 has some good cards, and there are some implementations that are pretty quick.
Feipoa's chart here really highlights that the 8900D-R in at least one implementation is quite good for DOS graphics.

The 96xx PCI cards were definitely pretty good, agreed there. I believe the 98xx cards were/are decent too.
The 9880 AGP is at least on par with a Voodoo 2, afaik. That's not too awful.
I hope we can cook up more as well! I wonder why there's a DOS-V intended VLB card as well.
I always figured the same re: PC-98 and proprietary setups.

Good to know with regard to the 9600 series. I guess the Windows acceleration is decent on them then?
The beefier RAMDAC and 64-bit setup sound nice. Will have to pick one up for messing around with.

------
@pshipkov
Liking the numbers on the 9440 and 9685.
Great to see them both together, thank you for that!
Curious about that 9685 UUD PC Player result. Otherwise makes sense.
96xx faster in Windows, great in DOS. 9440 great in DOS and solid in Windows.

Curious Forex assemblies. I've always wondered about them.
They made a VLB Pentium chipset, but I always thought they were like Efar and got out of the game early and just relabeled.

This 411/402 assembly is nice. CPU-intensive metrics are great.
A shame about the BL3 and GD5434 issues.
Being faster than UMC480 is interesting.
I wonder if any early 486 boards used this setup?
A FRX84C421 w/ CPU socket might be worth its weight w/ WB L2.
Really interested in this setup and in Forex more generally.
Seems like a good avenue to develop and explore.

------
Ran some quick tests on the Intel Batman's Revenge Socket 4 board and compared it in DOS interactive graphics to the SuperEISA running at 160MHz. To no surprise the SuperEISA creams the competition and gets close in Quake. Both were tested with near-equivalent (as near-equivalent as possible) VGA cards; the ARK1000VL and ARK2000PV.

Wanted to add some diversity to the mix. Figured Socket 4 would be an interesting angle, being a logical competitor to the 486.
The P60/66 are interesting historically for a few reasons. There's the FDIV bug, the transition to a 60/66MHz FSB, and just early P5 weirdness.

All things considered, this was a fun little test.

BIOS settings were as follows:
PCI Byte Merging Enabled
ISA Speed Enhanced

The Batman's Revenge itself deserves a few words itself.
It's a good board; it lacks any real "options," like any Intel board from any era, but it's stable and predictable.
Timings are not the fastest nor the slowest. BIOS updates are simplistic and the BIOS interface is cut-down.
IDE drives > 2.1GB lock the board. It will not POST without XT-IDE or some other drive overlay.
It is available in two variants: one supports the 5.15v "specified/required" for 66MHz operation; the other lacks the circuitry, but still can do 66MHz FSB.
The tested board does not have the 5.15v option.

The Pentium 60/66 are often called good alternatives to a DX4-100. This is true to the extent that the ALU is nothing to write home about if all you are doing is DOS gaming. For DOS interactive graphics--the hallmark and keystone for most benchmarks on Vogons--the Pentium 60/66 are best avoided. They're expensive these days, hard to find, and more-or-less lack the popular "retro" factor that surrounds the 486 these days. There's no doubt that grabbing an average 486 board and slapping a DX4 or Am5x86 in it is probably a better bet--these days, at least.

They are hot, face immature/bugged "standard" Intel platforms (430LX), and yet they shine, despite it all.

The 430LX bugs are worth discussing first. It appears that the Intel 430LX could not properly support WB L2 cache. Considering that Intel had such functionality implemented in the 420-series Saturn/Saturn II/Aires chipsets, this is incredibly surprising and confusing. The 430LX is non-integrated. It's four chips, and that's not counting a fifth if you want EISA. Not awful, but again, there's single-chip 486 solutions offering comparable/better performance. Chips manufactured before 1994 seem to have some issues with PCI/EISA bridging. This is, frankly, unacceptable. It seems that different implementations contained various bug-fixes and revisions; I have never personally seen a singular chipset redesigned over a two-year period of implementation to this extent.

Clock-for-clock, the P60/66 are impressive. We see some metrics here where even in the 486's wheelhouse of DOS interactive graphics, the P60/66 does okay. It's not far off in PC Player, Superscape, or even Wolf3D, despite the near-100MHz clock speed deficit. Impressive, especially for fast tech on an old, hot node.

The P60/66 also isn't quite that hot. A good fan and heatsink are really all you need. The fan isn't even necessary if there's airflow around a large enough heatsink.

Despite its reputation as a clunker; as a hot, overpriced chip not worth its weight in gold (and it's a heavy, large one 😀 ); as an under-achieving, dismal piece of history that led to the legendary P54 chips and beyond, the P5 is a solid foundation for what it is. It's a spartan chip, sure, and the platform is woefully immature, but chipset options out the wazoo that make exploration fun. The performance is good, not great, and with a better chipset like the SiS50x (to be tested later 😀) that allows for the of EDO, things should get interesting. VLB Pentium boards are great to mess with too, and they are pretty common on Socket 4 assemblies.

Impressions: Predictable, but intriguing. Clock-for-clock impressive performance. Nothing out of the ordinary motherboard-wise. Intel Batman's Revenge is typical "Intel." P60/66 interesting avenues. Nothing Am5x86-level, but nothing to scoff at.

I think of P5s like first gen jeeps and landrovers... peak power is laughable... but stick it in 4 low and you can pull a house down... slowly. Mine I was running at the turn of the millenium as a family utility box, and it could play MP3s on winamp 2.2 and websurf at the same time... top end 486 machines could only manage to play lower rate MP3s and nothing else. Strangely it seemed the smoothest most robust machine on Win95 I've ever seen.. smooth doesn't mean real fast, it just kept doing things at steady pace no stutter. First version of Carmageddon played tolerably on it also. Think that was low detail software render though.

P60 and P66 were 1993 products, and I think they were late to market. Nothing in 1993 (in the x86 world) could touch them, especially if you were a professional and needed the FPU. I think the issue was never so much that they were slow, more that they cost too much. Yes, an am5x86 at 133 offered similar performance, but those appeared in late '95. In the 90s, basically every year new products would come out making whatever you bought the previous year almost obsolete.

From what I remember reading the writeback L1 cache issue was finally fixed in one of the later revisions of the mercury chipset.

By the time the 5x86 was an option, power users had moved up to P133. But yeah they caused quite a stir in science community, like your own supermini on your desktop. Which was why the FDIV bug was such a big deal for them, lost confidence in that market.

Excellent write-up.

Batman's Revenge
Bold and mouthful.
Did DC even permit Intel to use the word "Batman" in the name of the motherboard ?

Never liked much Intel motherboards - they are too rigid in my opinion.
This guy seems to exhibit some of these characteristics.
Drive size locks. Lack of options.
But still very interesting early-gen assembly.

I have near zero experience with Socket 4/5/7, apart from some one-off encounters now-a-days.
Back in the day they were completely out of reach. Didn't even pay attention to them since they were exoplanet material.
But looking at your benchmarks - it took the competition 2-3 years to catch up to P5 @66MHz using 160MHz clocked 486 CPUs, as AC mentioned as well.
Nice.

DOS interactive graphics are a bit below top-notch 486@160MHz system, but Quake 1 is ahead, even at mere 66MHz.
If possible, can you run a 3D Studio rendering test ?
Curious what it will show. I bet it will be at least 10-20% better than what we see with the best 160MHz 486 systems.

The FDIV bugs were a big stink back then and maybe it was for some industries/applications, but for home users doing casual computing - a no story really.

Except for the early days of Athlon, Intel has never really had competition in the x86 world...and the only reason that happened was because Intel got greedy and tried to screw everyone over with serial architecture while AMD was busy buying up DEC's engineering department. Prior to that, Cyrix was the only one that managed to come close, with the 6x86-P200+.