I won't be surprised if am the Young Micro Systems most avid end user to date with many of their 386/486 motherboard models collecting dust nearby.
The interest is entirely based on my curiosity about Symphony Labs silicon which YMS used in their products. Much less about the company YMS itself.
Unfortunately none of their assemblies turned-out exciting so far. They don't respond well to tight BIOS timings and overclocking.
Until now. Well, mostly.

Young Micro Systems VS486G-3VL (VEGA 486 Green PC)
Based on Symphony Labs Wagner (SL82C491, SL82C492).

It is the most feature complete 486 VLB motherboard from the YMS series.
There is even a voltage regulator on board. Crazy ! : )

In crisp condition.

--- Am5x86 at 160MHz (4x40)

All BIOS settings on max, except CACHE BURST READ CYCLES 2:4 = X-2-2-2 (best is X-1-1-1)
Motherboard is not picky about RAM or level 2 cache chips.
Ark1000VL video cards works well with tight wait states (noting it since this is not always the case).
There is not much more to say really. Things work as expected, except SpeedSys refusing to cooperate and Quake 1 test showing some funny result.

Looks like skewed game clock at play. Otherwise the test completes just fine and the game is playable entirely.

Coretest reports 12207 Kb/sec with Promise EIDE 2300Plus controller (turbo mode + speed 8 ) + CF card.

benchmark results

One sentence conclusion - above average performer.

--- Am5x86 at 200MHz (4x50)

There are 6 jumpers for setting up base frequency, but no configuration results in 60/66MHz. Going with 4x50.

Had to slow things down for complete stability.
CACHE SHORT WRITE CYCLE = NORMAL (best is SHORT)
BURST WRITE CYCLES 2:4 = X-2-2-2 (best is X-1-1-1)
BURST READ CYCLES 2:4 = X-2-2-2 (best is X-1-1-1)
BURST WRITE CYCLE 1 = 3-X-X-X (best is 2-X-X-X)
BURST READ CYCLE 1 = 3-X-X-X (best is 2-X-X-X)
AT BUST CLOCK SELECT = CLKIN/4 (best is /3)

benchmark results

Coretest reports 13523 Kb/sec with Promise EIDE 2300Plus controller (turbo mode + speed 8 ) + CF card.

Not bad at all really.

--- Intel Overdrive P24T at 100MHz (2.5x40), POD100

System is inherently unstable with the modified P24T processor.
Tried hard to improve on that but without success.

benchmark results

---

Despite the POD100 issues, this is the best Symphony Wagner based VLB motherboard i have seen so far.
No hustle, overclocks pretty well, supports 3.45V CPUs.

I've seen these boards around. I didn't know they were also made by Young Micro Systems. Do you think these boards were mainly sold to OEMs? I've only ever encountered them loose.

I wonder if Symphony had a P5 chipset in their pipeline. Too bad they didn't get on board early and pump out a nice VL/EISA P5 chipset. I bet that would have kicked some serious butt.

Agreed. I had no idea about Symphony and YMS until few years ago.
Back in the day - not even a notion about them.

---

Somebody pointed me to this Voodoo 1 6Mb mod.
Pretty cool.
https://youtu.be/pwGdw0eZVCQ

Will add to the 386 story 2-3 more motherboards based on less popular chipsets, for completeness.

First in line is Matra 486SLC2 VESA based on EFAR Microsystems 82EC392, 82EC495 (relabeled OPTi 82C392, 82C495B) chipset, hidden under cool stickers with the word "PATH" written on them.

Still in great condition despite some scratches on the VLB and ISA slots which indicate some rough handling here and there.
Warranty labels still in place. Traces of small battery leak that was taken care of before causing real damage.

One day will spend the time to find what "PATH" stands for. Anyway.
The motherboard is interesting for several reasons.
The obvious ones - a mixture of rare IBM BL2 (SLC2) processor soldered on-board, 256Kb level 2 cache (unusual for 386SX class system), and a VLB interface.
EFAR chipsets are seen predominantly in cost reduced assemblies, or late hybrid 386/486/VLB ones.
The first type are perf and scalability crippled in various ways, the second type come with 25/33/40MHz clock generators instead of crystal oscillators which limits overclocking options.
This motherboard comes with up to 80MHz clock generator, but also has wiring for crystal oscillator which opens the path to overclocking and discovering where the performance peak really is.
Removed the clockgen from its socket. Soldered 4 pins. Ready to roll.

It will be great if one day i manage to obtain one of these late EFAR based mobos to test with BL3, SXL2 and standard 386DX processors. Ok, back to tpoic.

It was established in the past that the fastest 386SX class system is Alaris Leopard (OPTi 82C295) with the same IBM BL2 CPU, so i was really curious how the EFAR chipset will stack against it and handle the unusual combination of hardware this Matra motherboard is.

Only 25 and 33 MHz base frequencies are supported via jumpers, but i was able to find a 40MHz configuration. Unfortunately between warm resets the frequency drops down to 20MHz from time to time.
This was another incentive to move to crystal oscillator.
The system takes 80MHz crystal with easy, but struggled with 90MHz. No lights at 100MHz.
While my hopes for higher overclock faded quickly, at least 80MHz became reliable between resets.
All jumper configurations on max.
The AWARD 4.51 BIOS resembles 486 layout and type of options. There are quite a few timing settings.
All parameters on max except:
CACHE DATA BUFFER OUTPUT = DISABLED (best is ENABLED)
DRAM QUICK READ MODE = DISABLED (best is ENABLED)
DRAM QUICK WRITE MODE = DISABLED (best is ENABLED)

Similar to Alaris Leopard this motherboard was not happy with most VLB EIDE controllers - their DOS drivers hang during initialization.
DTC 2278E worked fine but had to be slowed down a lot which kind of defeated the point of using an EIDE controller. If nothing else it proved that 2278E is one of the most compatible guys around.

Ark1000VL works fine.
One unexpected outcome - the S3 Trio64 VLB driver hangs Windows. This is an omnipresent occurrence regardless of jumper/BIOS settings.
Happens even with the slowest possible base frequency and most conservative wait states.
Used Orchid Kelvin 64-VLB 2Mb (Cirrus Logic GD-5434) instead - slower than Trio64 in Windows, but still pretty good.
(Will double check that stuff at some point later)

benchmark results

Matra / EFAR 82EC392, 82EC495 is in the top performance tier for 386 class hardware and on par with Alaris Leopard / OPTi 82C295.
It is same or faster in some of the DOS interactive graphics tests, slightly slower in the FPU intensive ones.
Really liked this motherboard. Not an entirely trouble-free experience, but somehow it grew on me a lot.
Maybe because it is a great/better alternative than Alaris Leopard, which until now was in a class of its own in the area of all things 386SX.
Also, this was my first time with EFAR silicon that i really wanted be good, and it did.

That's an interesting discovery with the EFAR SLC2 system. The Matra is pretty rare compared to the Alaris offerings, so I've never had a chance to test one.
I remember somebody speculating that all "PATH" labelled chipsets were Acer, but in my opinion PATH is probably just the name of the vendor, and the sticker can appear on just about anything. I'd also heard that EFAR is rebadge of something else. The model numbers are similar to OPTi, but that's just pure speculation.

I didn't know that these 16-bit boards had issue with EIDE controllers, because I only use SCSI. The funny thing is that I'm pretty sure I tested mine with the AMI FASTDISK VLB caching controller, and I remember it working. Normally I just use a 1542 or 1522 on a system like this.

From what i know Matra was a French company with a foothold in several industries, including defense and aerospace, also some personal microcomputers stuff.
Maybe this motherboard was produced for some industrial purpose of theirs. At least the presence of warranty stickers suggests something like that.

As for EFAR chipsets being relabeled - no idea.
I don't think the 495 number is related to OPTi.
That OPTi version is a single chip design versus the 2 chips for EFAR.
In the other hand performance is basically identical with Alaris Leopard / OPTi 82C295, which can be a coincidence of course.

VLB EIDE controllers are more reliable and much faster than SCSI counterparts from the same time period, especially when the systems gets pushed around.
ISA SCSI or ISA/VLB caching controllers are just older generation hardware, so they are even further behind.
So, i kind of ruled them out as viable options, but now when you brought it up - i should have tested the Alaris and Matra boards with SCSI VLB or ISA adapters given the fact that EIDE is a problem.
Adding it to my to-do list. Will try it at some point soon.

pshipkov wrote on 2023-02-28, 07:26:

VLB EIDE controllers are more reliable and much faster than SCSI counterparts from the same time period...

Are they? SCSI was always touted as the faster, more reliable alternative. Anyone done or seen a reliability study between, say, the AHA-2840VL SCSI card vs. DTC2278 IDE VLB card?

This is what i thought long ago.
Turned-out to be an urban legend.
Empirical evidence suggests otherwise.

Remember sharing some thoughts about it here.
Also, take a look at page 5 in this thread.
I keep updating the data there from time to time - few more controllers of interest left on the radar.

I remember seeing those, but I think a much more in-depth study is needed.

Segregate ISA, VLB, and PCI tests, as well as bus overclocked, and non-overclocked. Broken down further by slow system vs. fast system, e.g. ISA SCSI on a 286 may show more benefit than, say, ISA SCSI on a BL3 system when compared against ISA IDE. Same may apply for VLB systems. Ideally you'd use the same class of fixed disk media for the tests on SCSI and IDE, e.g. if using a CF card on IDE, use the same card on SCSI, perhaps via an ACARD device. Using a old 1 GB, 5400 RPM SCSI drive vs. a modern 80 GB Ultra 133 HDD wouldn't make for a fair comparison. Another item to look for would be CPU load when multi-tasking. For example, doing timed HDD transfer tests (with a stop watch) while continuing to load google search results. Check the time to transfer and also the CPU load.

Running an in-depth study would be incredibly time consuming, but would also be very interesting. I'd wager that such a study exists somewhere already, but you won't know if the results have been compromised due to special interests.

pshipkov wrote on 2023-02-28, 07:26:

As for EFAR chipsets being relabeled - no idea. I don't think the 495 number is related to OPTi. That OPTi version is a single c […]
Show full quote

As for EFAR chipsets being relabeled - no idea.
I don't think the 495 number is related to OPTi.
That OPTi version is a single chip design versus the 2 chips for EFAR.
In the other hand performance is basically identical with Alaris Leopard / OPTi 82C295, which can be a coincidence of course.

OPTI 495 chipsets are not single chip. Aren't they always paired with a 82C392?
In any case, the 495 is a chipset for 32-bit CPUs, so it would be a bit of a waste.
But who knows, maybe the board maker had a big pile of 495 chipsets they needed to dump and rigged it for a 16-bit CPU.
There are software utilities to adjust opti495 chipset registers. You should test it on your EFAR chipset to see what happens.

*edit*
The 495XLC and SLC do not have 82C392s...I suppose they must be the highly integrated versions.
The original 495 and the 495SX do however. The 495SX comes with a truncated version called the "82C392SX" which I think is just a regular 82C392 in a smaller package. So it's possible that your EFAR chipset is an OPTi 495 sans postfix.

Well now I know why I was really daydreaming about BL3s, apparently it was my subconscious telling me I had one buried in the basement. Bottom paragraph Re: What retro activity did you get up to today?
Think it's an opti chipset opti i/o etc but main chip is under an IBM sticker, similar deal to the cougar I guess, IBM made it. Couldn't see any numbers on mine, might have to completely disassemble to see anything, for figuring if this motherboard went anywhere else, or if IBM supports in in any way, jumper settings would be nice.

@feipoa

This area is not well studied indeed.

Like everyone else, I have some observations about the different hardware platforms/configurations you mentioned. Can share some thoughts as well, but the subject was VLB local storage controllers.
These i test on Asus VLI-486SV2GX4 2 or 2.1 with Am5x86 processor running at 160MHz (4x40). All settings on max. Same CF card type (with Acard adapter for SCSI) and HDD model that the controllers cannot saturate.
Initially i was running many more tests, including in Windows 3.11.
Noticed that results from the different tests are more or less in line with each other for the given controller.
So over time resorted to the faster tests which are still indicative of the overall performance.

Subjective, personal stance here:
I see 486 VLB class hardware as DOS/Win3.x first and foremost, so local storage access from multiple processes/threads is a minor (no) factor. Single thread peak perf is what matters. For this i can safely say that EIDE VLB >> SCSI VLB.
Some of these controllers have drivers for Win NT and other operating syatems from that time, few have for Win 95, but never spent the time to test properly multithreaded local storage access, which is what SCSI is supposed to be better at.

---

@Anonymous Coward

Never saw 2 chip implementation of OPTi 495. Do you have a visual reference ?
To me 2 chips is pre 495 silicon - the 393 and so on.
Can you also point me to the OPTi chipset tweakers please ?

---

@BitWrangler
Very cool find. But picture or didn't happen. : )
From the notes you share about your place i picture some vast dark dungeon with many dead ends filled with arbitrary artefacts.

pshipkov wrote on 2023-03-01, 18:53:

@BitWrangler
Very cool find. But picture or didn't happen. : )
From the notes you share about your place i picture some vast dark dungeon with many dead ends filled with arbitrary artefacts.

It's a small dungeon, kinda dimly lit, 4 dead ends, but yah, very arbitrary artifacts.

I got a few pics yesterday but wasn't very happy with them due to taking them unsteadily one handed, and had to have the flash on because light wasn't great. Then what the flash seems to do is make light dust shine 3x brighter and make smudges and darker features 3x darker, so everything looks 3x more filthy than it really is....

But here's the front of the neglected beast and it's heart...

Has two IBM 4MB SIMMs which I guess it must have came with, spot the odd cache chip, I cut the speeds off, but that one is 15ns whereas the others are 20, the tag is soldered down. Not sure if it just randomly happened in production or was an in service replacement. Can see the brown slot there, that's inline with the ISA riser, I'm assuming it's probably VLB but could be something else. CPU fan is on with only 2 screws just like twitter dudes. I didn't want to pull that off to expose CPU or it would be annoying to get it somehow reattached. There is a CDROM audio cable loose in the case so guess it has had sound and CDROM, don't know which.

edit:pic fix I hope
editII: I just realised there's 9 socketed cache and one on the board, so 15ns is more tag? but it seems illogical placing, like it should be other side in the row of 5.

The plain Jane 495 is a relatively obscure chipset. I've only ever seen it used on 486 boards, but it should work with the 386 also.
The 495 chipsets are just 391s that support 486 chips, aren't they?

Here's the 495sx one I have:

@BitWrangler

Fair enough about your dunge (tm).

From your pics it looks like some IBM made mobo - the slanted RAM modules hint at that.
Good chance it is BL3 chip under the heatsink. If it is - and if it is functional - it will be good to give it a spin.

---
@Anonymous Coward

First time seeing a pre 495S# version.
Thanks for the reference.
Ok, so maybe you have a point - this EFAR chipset can really be relabeled OPTi.
Especially in given the striking similarity in performance.

I went through few S# models.
Got rid of them. Only 2 left - really good assemblies actually.

I forgot to mention. CTCHIP can modify OPTI registers, but if you have the databook you should also be able to do it manually in debug.

This popped up randomly in other searching
https://www.datasheetarchive.com/pdf/download … 2520motherboard
App note for a regulator to use with BLs, not sure if it has any new info, but if that reg is on your board maybe can sneak the voltage a tad higher with a pencil trick if you find "R2"

pshipkov wrote on 2023-03-02, 02:53:

@BitWrangler […]
Show full quote

@BitWrangler

Fair enough about your dunge (tm).

From your pics it looks like some IBM made mobo - the slanted RAM modules hint at that.
Good chance it is BL3 chip under the heatsink. If it is - and if it is functional - it will be good to give it a spin.

Wouldn't it be easiest for BitWrangler to turn it on to determine the installed hardware? If this is a BL3 system w/VLB on an OEM Patriot system, I'd be interested as well. Keep it "local". haha.

pshipkov wrote on 2023-03-01, 18:53:

This area is not well studied indeed. […]
Show full quote

This area is not well studied indeed.

Like everyone else, I have some observations about the different hardware platforms/configurations you mentioned. Can share some thoughts as well, but the subject was VLB local storage controllers.
These i test on Asus VLI-486SV2GX4 2 or 2.1 with Am5x86 processor running at 160MHz (4x40). All settings on max. Same CF card type (with Acard adapter for SCSI) and HDD model that the controllers cannot saturate.
Initially i was running many more tests, including in Windows 3.11.
Noticed that results from the different tests are more or less in line with each other for the given controller.
So over time resorted to the faster tests which are still indicative of the overall performance.

Subjective, personal stance here:
I see 486 VLB class hardware as DOS/Win3.x first and foremost, so local storage access from multiple processes/threads is a minor (no) factor. Single thread peak perf is what matters. For this i can safely say that EIDE VLB >> SCSI VLB.
Some of these controllers have drivers for Win NT and other operating syatems from that time, few have for Win 95, but never spent the time to test properly multithreaded local storage access, which is what SCSI is supposed to be better at.

If it is just VLB SCSI vs. VLB IDE, limited to Win31, then by the shere number of options for VLB IDE, there's already a greater probability that VLB IDE has a faster option. There weren't all that many quality VLB SCSI controllers to choose from. I would imagine that, at the time, people wanting SCSI would be using EISA or MCA. If limiting to Win31, then this changes things as well because fast Win95 drivers for VLB IDE might be hard to find or non-existent. It is even hard finding quality W95 drivers for integrated PCI IDE.

Yah think all the 2 and 4 CPU server boxes for netware would have been EISA and the performance demand there would have been high.

I will attempt to meditate on what card a college buddy had for VLB EIDE, had drive overlay software, DMA drivers for DOS OS/2 Windows, that was 27 years ago though, so memory not that fresh. Anyway, I had a couple of winbond and UMC cheapies neither of which matched up to his. I wanna say he was getting 3+MB a sec random read, which is more like mid pentium class typical i/o

Efar has a somewhat interesting history as a re-labeler. From what I gather, based on the work of the guys at The Retro Web, it seems that, for example, the Efar 802GL is a clone (?) of the SiS471. Note the board linked here that has two variants on the same PCB. One is equipped with the 802GL and associated buffer/RTC chip, and the other has the SiS471 and SiS407 accompaniment.

This is interesting, as it seems that there was at least one board with such a chipset and a PCI/VL bridge: The Alaris Tornado 2.

This is my board, and it's dead as a doornail (at the moment). I'd like to get it going sometime sooner than later and see what's what; I believe it's equipped with a MR-BIOS, like most of the Alaris parts.

The PATH boards are another story. I've done a little research on them, and there's one other than the Matra board. It's an unlabeled board that is kind of to the Matra as the Leopard LX is to the Leopard. It lacks VLB, and it has a completely different chipset. It's equipped with the ALi M1217. Here's a photo from TRW (I believe this board is mt777's, based on TRW's credits):

Oh, and the whole notion about 32-bit chipsets with 16-bit CPUs is an interesting one. It's not a proposition that I can say is impossible; in fact, there's one other such board out there and, yes, it's an Alaris part with an IBM 486SLC2. This board, the "Alaris Cougar 486SLC2," as described on The Retro Web, makes little sense from a design point of view. It's effectively an Alaris Cougar BL3 board, but with a 16-bit external bus 486SLC2 in place of the BL3, whilst retaining the Socket 3. In my eyes, this is the first time I've ever seen a 387SX on the same PCB as a PGA169 socket. It unnerves me.

There's a lot to be said regarding these chipsets that can probably be garnered from their databooks, if extant. I haven't read through the OPTi 499's, for example, but I admittedly am curious as to how performance of a 16-bit chip is on a 32-bit external bus chipset. It should work, in theory, as even VLB has 16-bit operating modes and cards like the Cirrus GD542x are 16-bit host bus native. That being said, I can't imagine that it's a pretty implementation 🤣