I've glanced through the 499 databook a few times. I was always confused as to what made this chipset part of the "Premium" line. Feature-wise it seems pretty much the same as the 495, except that the 82C206 is integrated into the main IC.

What does that opti 82C601 chip do?

I think the level of integration designates the Premium moniker. Otherwise, your guess is as good as mine.

Seems like the 601 is just a buffer chip, but, unlike the 602, lacks an RTC.

The upstairs bench is still buried in PII stuff, and the bench I was gonna have in a corner of the basement fell through like 6 times over 1 3/4 years, so I thought screw it, I'll just throw some folding tables up in that corner and see what this Patriot SL4100 does... nothing much right now. I grabbed nearest monitor though which was a 50/50 shot on that working. PSU spins up drives unplugged from board, seems to be okay. So plugged it into board and cpufan spins but no beeps, no video, on either onboard vid or vid card that was installed. Not sure of the monitor, so swapping that out, came up from the dunge to look up numbers, it's got P/N 11H4518 on the right edge of board, and COBALT Ver 2.3 IBM CORP. on the BIOS. Part number seems to go nowhere so far and Cobalt keeps sending me to Alaris Cougar stuff. Anyhoo, grabbing monitor I'm more sure of, POST card, and checking whether keylock was a disarm switch too.

Edit: Well this is going to be fun. I have no idea what POST code tables I should be looking at. The BIOS label says Cobalt which points towards the cougar boards with MR BIOS and AMI options, and the POST screen twitter dude posted says SurePath BIOS... which is a PS/2-ish IBM thing, maybe based on Phoenix codebase and done by them. I wasn't even sure I had right port set, tried 80, got FF, tried 90 FF, idk, 290? FF, 300 FF, wiggle and jiggle go back to 80, FF then 02, hmm okay, maybe it is 80. turn off, wiggle stuff again, reduce to single RAM module, get 08 now, IDK what it might be, RAM refresh or chipset initialisation??? Guess I'll just have to clean and jiggle and make the number go higher.

Normally higher integration results in a lower priced product...S3 Trio, CL 54xx, SB16 Vibra etc. I guess OPTi's marketing team didn't get the message.

@bitwrangler

If the board does not come to life - render me interested in obtaining the CPU.
Juuuust putting it out there ... : )

Good find - that document about bl3 voltages.

---

About ISA/VLB local storage controllers from the early-mid 90'ies.
You are right that most of the drivers for NT3.5x, NT4, Win3.1x, Win95 are not very good.

Don't know much about EISA/MCA myself.
From technology stand point they are older tech than VLB. Stuff that didn't take off really. Can not match good VLB hardware.
No idea what the market dynamics were at the time. I was too young to understand the bigger world.
I only remember when swapped dusty Trident 9000 ISA VGA with Cirrus Logic GD-5426 VLB on a 486DX-40 machine = massive boost in graphics performance.
That card had integrated EIDE on it = massive boost in HDD I/O.
I was so impressed. Anyhow.

@WJG6260 posted some hard metrics about EISA/VLB boards - the first time i saw such info (thanks), but his focus was mostly on VLB, so EISA is still a question mark for me.

---

Thanks for the notes and pictures WJG6260.
Very interesting stuff.

When i hear OPTi i think "jack of all trades" - compatibility/integration/cost first, performance second.
Not very interesting compared to other (leading) designs from the same time.
But quite interesting considering the several hybrid/unusual assemblies from Alaris, DataExpert, and others, who mix next-gen features with previous-gen processors.

When i hear EFAR (maybe ETEQ as well) i think "snake oil" - never sure what's in the bottle.

---

@Anonymous Coward
Good point about the CTCHIP executable.
Will give that a try with the Matra board and see if it has any effect on it - that will be a good hint about your suspicion about OPTi silicon inside.

@AnonymousCoward

Agreed. I don't know why, but it seems that OPTi always pushed integration as a "premium" marketing feature. I guess they had good sales guys haha.
---

@BitWrangler

I'm excited to see where this beast goes. Hopefully you can wake it up and scare away any evil spirits preventing POST. I am currently nursing an ASUS VLI in the same fashion. Jiggle and wiggle is always the way (until you jiggle too hard and have to break out the soldering iron) 😁

Cobalt was another name for the Cougar, but I think there could have been more than one Cobalt board. There were multiple Alaris Tornadoes/Leopards/etc. Maybe there's something under a name like Cobalt LX/VX? Just throwing it out there to see if anything sticks.

---

@pshipkov

I'm still trying to weave through the EISA web myself, to be honest. There are a myriad of EISA cards, but few of the good ones to be had these days. I have some experience but no hard numbers (well, I had numbers, but Excel crashing ate them 🤣) on EISA VGA cards. Unfortunately, I only have two of those, and one is the 1MB Compaq QVision that leaves lots to be desired. It's a nicely made card, but it's equipped with a mediocre DAC and 1 MB of VRAM, meaning that its Windows performance isn't too good. It's pretty close to an ET4000AX in DOS, and okay-ish in Windows. I can get you some hard numbers there. The real star would be the ELSA Winner 2000XHR, but I only have a 1000XHR. The 1000XHR is great, and it's plenty fast. The S3 928 is nice on ISA and very, very nice on fast EISA systems. Otherwise, there's no difference.

EISA is also kind of a mystery in that every implementation behaves a little bit differently. On the old Usenet archives, there are plenty of people complaining about incompatibilities with the various EISA chipsets. SiS's excellent 406/411 (in the SuperEISA) even gets some flak. I think it's not exactly the best implementation, as it doesn't recognize some of the more obscure EISA controllers I have, such as my DTC 2290 IDE card. I had hopes for the VIA-based board I'm working on and it's looking okay in that front, but EISA configuration software isn't playing nice on there. I think I need to repair the RTC and add an external 3.3v coin cell; thankfully, that's a small fix. It could also just be that it was the only such board with such a chipset. The other experience I have with EISA is in the form of the DataExpert 4409. That uses the HiNT CS8001/8002 Caesar Mini-EISA chipset. It lacks certain busmastering lines, so some hacks have to be used to get certain SCSI controllers going on it, but it actually seems fine otherwise. It's not a bad board, but very early.

Pentium + EISA is where things get interesting. The ASUS P/E-P55T2P4D is a dual Socket 7 behemoth that really has a wonderful EISA implementation. It's just too big and clunky. The older 430NX-based ASUS PCI/E-P54NP4 is cool, but it lacks integrated I/O. It also employs a clunky 3.3v PCI card that takes up one slot just to fit a large VRM and heatsink. The performance, however, is much better on these later and more mature chipsets. I wish I still had my numbers, but just to give the gist before I re-run and post them here, I did notice an appreciable difference with the S3 928 in EISA and ISA configurations. It was significantly faster in Windows, and faster on average by a smaller, but still notable margin, in DOS.

I'm glad to have been able to provide some random factoids 😀

It's always kind of fun to see what these designers were thinking (or not 🤣). I think that a great degree of the 90s innovation we lack today came from the willingness to dip our toes in insanity now and again. OPTi is definitely kind of shoehorned into that group and your on-point analysis is true. They were really a very compatible chipset manufacturer. I think that they did occasionally produce very high performance chipsets, but by the late 486 era, they were outgunned. The OPTi 495 is decent, but the 391/392 is pretty sweet. The PCChips M321 isn't a bad board by any means. The OPTi 495XLC is potentially interesting in that it supported Cyrix's proprietary write-back implementation before Intel's P24D standard existed. The 486S FasCache would be cool to test on one such board. I do think Alaris was a bit of a strange one. I still cannot figure out what their deal is; they appear to have gone by the name RYC as well. My Leopard and Leopard LX are marked RYC, as is my Cougar.

ETEQ had some really nice chipsets for the 386. Soyo has a board with the ETEQ Cougar that looks slick. It's only limited to 256k of L2, but it is equipped with two banks of 30-pin SIMMs, something seen only on premium boards like the SuperEISA. EFAR, however, loses me. If the Alaris Tornado 2 I shared really has an SiS471 clone with an OPTi 82C822 VL-PCI bridge (which should work, since the 822 is not chipset-dependent), that could be kind of cool. Otherwise? No idea. Snake oil is a good way of putting things. Best to reserve caution.

----

On a quick note: to all those out there with EIDE2300+ VLB controllers, I just dumped BIOS v1.16 from my card. I have a revision 3 board, with no tantalum capacitors (all radial electrolytics), a factory EEPROM with a 1996 BIOS date, and LGS PRIME 3B I/O chip on the front. I have no clue if this offers anything new, but figured it's worth passing along out there and preserving.

Got it POSTing!!!

Might be some work to go yet, Dallas is completely dead, it won't even save stuff for warm boot. CMOS setup is quite nice, can set the BL at 1x 2x 3x, both caches on and off, that's 9 speeds, plus whatever turbo does, ISA bus 1/3 to 1/6 clock, all cache and memory timings.

That's just about the coolest 486 splash screen I've ever seen. It's a pity IBM never really employed that font on many of it's actual CPUs. Only the Cyrix based ones got it, and they quickly dumped it for some boring Times New Roman in later models.

WJG6260 :
"I did notice an appreciable difference with the S3 928 in EISA and ISA configurations. It was significantly faster in Windows, and faster on average by a smaller, but still notable margin, in DOS."

Is the 2000XHR different than a standard 2000? Anyway, I traded my Winner 2000 a while back because I had no good use for it. Like you said, it's barely faster than the ISA version in EISA mode under DOS. I agree Windows performance was okay, but if you have a VLB slot then just about any VLB card would be way faster under DOS, and at least as fast in Windows (possibly with low quality though).
In hindsight I think maybe I should have kept it. It would have made a nice ultimate ISA card, or maybe an ultimate VGA card for an EISA 386....but I still have yet to find the EISA 386 of my dreams. There weren't many of them, and most of them are old and clunky (usually 25 or 33MHz)...and usually proprietary like a Compaq or AST.

Yah, I like the splash, seems "impossibly" cool for IBM, like management weren't around when they did it... only recent one I've had that I like near as much is a Red Fox PIII board. Wondering if this has a 499 because I can only see the one big chip with the IBM logo on, but they did weird stuff like put chips on the bottom of some of their other boards, and I haven't looked underneath yet.

I was getting 61 on the POST card and still no display off trident that was sitting in it, or another one that was handy, and 61 seemed to say it was trying to display something, so drilled out the blanked pin on the onboard VGA and hooked up to that, and lo and behold POST screen.

I knew 301 was keyboard but 8603 was worrying me, turns out it's complaining about lack of mouse I think https://www.ibm.com/support/pages/troubleshoo … ric-error-codes which wasn't high on my list of expectations, not many boards care. Not entirely sure how to get back into setup if it did manage to save, had told me variously F1 to continue (classic keyboard error troll) F2 to continue to setup on an error, and F10 for setup.. so yah, mash F keys then. F10 probably most likely.

Next I guess I'm figuring out the Dallas situation and taking the riser out for a good clean. I/O onboard, onboard vid maybe needs another 512kB chip, 3 ISA slots for goodies. But what? What are the coolest ISA cards in the world to go with the coolest splash screen in the world? 🤣 .... Not sure though that VGAs will test without disabling onboard, too many unknown jumpers, but I have a Stealth Vram (S3 911) and a Orchid Kelvin (GD5434) ... Thinking for sound of digging out one of those Aztechs with covox etc support, I think I've got one but not 100% sure.

Hi,

@WJG6260 (edit):
I used to have to 486-EISA computers: A compaq server and one self built. The second one I dismantled, but still did not put it back together yet.
When you interested in some benchmarks:
Re: The Ultimate 486 Benchmark Comparison
The compaq EISA has a Pentium Overdrive installed @83MHz (L1 WT only unfortunately), 112MB RAM (weird number I know)
and a Compaq Qvision EISA - so the same that you have, too.

More random tidbits found on search for hard info about my board.

Opti 82C465 Evaluation Board User Manual https://www.datasheetarchive.com/pdf/download … e%2520lightning potentially contains a couple of hints about implementing Blue Lightning support, also some insights into jumpers/switches that may have been left unimplemented or mysterious on other motherboards. Possibly has things that apply to whole 465-499 family.

465 focused technical note about turning on writeback for Cyrix DX2 may help with other implementations https://www.datasheetarchive.com/pdf/download … ation%2520board

Period piece about Cobalt AT 486BL (Cougar) https://jawawa.id/newsitem/ibm-goes-oem-with- … oard-1447893297

It occurred to me that IBM Microelectronics really isn't part of the Personal Systems division, though apparently had interaction with it for PS/1s, Aptivas and Valuepoints. So I'm trying that angle of attack, but so far just getting a bunch about their PowerPC boards.

@WJG6260

Thanks for the notes. Good stuff.
Looking forward to the weird Pentium boards you are lining up.

---

@BitWrangler

Needs bigger logo !
IBM marketing team was very proud of their product - that's for sure.

Joking aside.
Let's see where this guy goes from now.
Use the Orchid Kelvin (GD5434) please.
You said that BIOS (or maybe jumpers) allow you to set 3x multiplier. This implies BL3 processor.
The manual
Can you get it at 3x33 MHz ? Some documents suggest 3x is possible but they the "Evaluation Board User Manual" talks about 2x only.
It looks like clock generator supports 25, 33, 40 MHz only. So your best bet will be most likely the 3x33.

---

So, are we saying that the SuperEISA board is potentially the best EISA option for up to 486 / Socket 3 hardware ?
Or this is a loaded question given the very few sampling points we have about EISA hardware ?

BIOS setup settings where the 1x 2x 3x is are not sticking at the moment, it's defaulting to 2x, but possible breakthrough below leads me to think it IS a 3x, just gotta fix the dallas....

So I had a feeling I was missing something about IBM budget/mass-market systems, remembering IBM blue lightning ads from back in the day, though mainly from refurb and surplus clearance places. AMBRA that's what, a whole subdivision of IBM they buried. Kind of an arms length, this is the touchy feely friendly home division of IBM (That 1984 Apple commercial was making fun of the other guys, honest.) Ambra! So Blue Lightning featured heavily in their 93-94 advertising, and their systems specs lined up with what I know about this Patriot, cases were a tad different, floppy on other side mainly, anyway, they name their systems by processor class, 4, and clockspeed, so they had a D466BL which was a 66mhz BL and a D4100BL which was a 100mhz BL... 4100, so I think that Patriot is supposed to run at 100, good.... Also opens up a bunch of things to search for to find these systems and boards, maybe there's a Patriot SL466 SL433 etc etc

edit: dammit, thought I found the goldmine at https://www.elhvb.com/mobokive/eprm/eprmw/40.htm ambra files, but only drivers and fixes no info. Some of those things may be potential workarounds for BL weirdness with certain things. I might try the Cirrus drivers from there in case they seem optimised.

editII: chips.ibm.com the microelectronics site seems to be a dead end on archive.org can't bust into where they might have had motherboards and support, as is older ibm.com and pc.ibm.com aptiva aptiva no ambra, heh, there are 1996 support page captures but they hadn't figured out what to use the web for yet, they were sending you to AOL, Compuserve, Prodigy, PC BBS and their faxback service.

EditIII: Ambra Achiever 300 seems a likely candidate for a board match on specs, can't find pics of that though either. Several possibilities chased but turned out they were SX class 16 bit, or wholly intel 486 class based. Been attempting to synthesise common patterns of jumper use, out of aptiva and other boards which are "warm" i.e. opti chipset, 486DLxx, onboard Cirrus Logic etc.. so far have determined that the video disable jumpers might be somewhere near the video chip, though usefully more than once they used TWO jumpers to disable video, which is useful info as trying single jumpers at a time it might have driven me nuts if it turned out a second also had to be set.

While waiting on BitWrangler's next steps with his IBM Bl3 system i am going to post information for one more 386 motherboard.

Soyo 386DX/33(?) revision B, based on ETEQ "Bengal" ET82C4901, ET82C4902 chipset.

Most boards based on ETEQ chipsets tend to be wide PCBs with low level of integration - dont inspire confidence.
2-3 models have the right bones, including this one here, so level of interest was significant.

The motherboard came in really bad condition.
Covered entirely under tick layer of dust and dirt.
Massive battery leak in the upper left corner that marinated all components in the area over many years.
Damaged traces on both sides.
Up to half a millimeter tick crust on all metal surfaces.
Solder pads brittle and with compromised conductive properties.
Keyboard connector was a corroded mess.
Chances for positive outcome were minimal.
Obviously no signs of life upon initial power-on.

Threw everything i got at it.
Multiple passes of acetone and vinegar scrubbing.
Hot water showers and brushing.
Careful removal of the hard crusty bits with multiple precise tools.
Tracing and fixing broken lines.
Replacing the keyboard connector.
Replacing keyboard controller and BIOS sockets.
Still no life.
At some point realized that some of the pins and pads of the 206 chip were not electrically connected despite visual inspection suggesting otherwise.
Apparently they corroded badly enough.
Scraped the upper layer of corrosion and tried to solder them, but for my surprise no amount of flux would make the solder stick.
Ended up simply covering the pairs of pins/pads with blobs of solder. Does not look good, but did the job.
That last fix did it and the motherboard came to life. For my great pleasure.
Several layers of conformal coating hid the damage and greatly improved optics.

Btw, the build quality of this assembly is great.
The design of the RAM sockets is very interesting. They don't require pulling handles sideways to free the memory modules.
Everything needed is to simply push or pull the modules. The mechanism works in such a way that once inserted they are stuck there for good, but applying the right amount of force in the opposite direction frees them just fine.
Very handy.

For some reason had to keep PARITY CHECK turned-off.
Tried different types of memory modules but nothing worked.
At the same time all memory tests are happy, so not sure what's up with that.

Components used for testing:
256Kb L2 cache.
16Mb RAM, 50ns rated chips.
Cirrus Logic GD-5434 video card with 2Mb RAM.
Standard IDE controller with CF card.

Level 2 cache can operate in write-back mode if the corresponding socket is populated with 16k x 4 SRAM chip.
Performance improves further in this configuration.

--- AMD 386DX running at 45MHz, ISA bus at 15MHz

50MHz is possible, but system is quite unstable.
Tested at 45MHz.
Everything just worked. Nothing more to say really.
All BIOS settings on max, except:
AT BUS CLOCK SELECT = CLOCK/3 (best is /2)
DRAM WRITE CYCLE = 1 W/S (best is 0)
DRAM READ CYCLE = 3 W/S (best is 0)
SRAM WRITE CYCLE = 1 W/S (best is 0)

Intermediate level performance in DOS and Windows graphics tests.
Didn't bother with offline graphics tests given the above - it was obvious that it won't bring a surprise.

--- TI 486SXL2

For my big surprise the SXL2 processors didn't work at all on this motherboard.
POST does not always complete, random hangs and so on.
Tried hard to overcome the issues but without success.
First time seeing SXL2 not on the same page with 386DX.
Until now i thought they are 100% compatible.

--- IBM BL3 running at 90MHz (2x45), ISA bus at 15MHz

This motherboard simply loves the BL3 processor.
Stuff just works. No effort at all. Nothing more to say really.

All BIOS settings on max, except:
AT BUS CLOCK SELECT = CLOCK/3 (best is /2)
DRAM WRITE CYCLE = 1 W/S (best is 0)
DRAM READ CYCLE = 2 W/S (best is 0)
SRAM WRITE CYCLE = 1 W/S (best is 0)
PAGE MODE = DISABLED (best is ENABLED)

Performance is impressive !
Number 2 in Quake 1.
Number 4 in Doom.
Number 3 in the heavy offline graphics tests and PC Player Benchmark.

And that is a 45/90MHz system.
Will be looking for more of this ETEQ stuff in the future for sure.

---

The combined benchmark results have been updated accordingly.
Look for Soyo 386DX.

---

This is my first experience with ETEQ silicon and i like it. Lots of potential.
With that said, there are other motherboards based on ETEQ chipsets:
- 386 designs with low level of integration that just don't inspire confidence.
- Hybrid 386/486 builds with VLB slots. Some of them don't have FPU sockets. Hope to be able to try some of them one day.
- Pure 486 ISA/VLB systems that don't look very promising. Probably a pass.
Feedback will be welcome.

Nice, victory tastes best after the long struggles.

So not too much to report, pulled riser out for good scrub, it's got a set of holes for a backside ISA slot, bringing total to four, and the hole is in the back of the case, I don't know if I can resist. Got the Dallas eased out, was so stuck I thought I was gonna lose pins, but out now, and found something to hold a battery, so getting modded tomorrow. Tried a corner of the IBM label, but it's paper and not going to come off clean so it's staying on.

Finally though, I clawed this out of the belly of the internet with my fingernails... https://www.praxis8.com/ambra/default.htm with...
"cobalt board" jumper settings... https://www.praxis8.com/ambra/cobalt.htm which match the sizes and naming of the jumper blocks on mine... incomplete though, none of the processor config, dudes like "but you don't need a dx2" on another page. Yah well, I don't but might have wanted to compare. Anyway, looking at that video disable pattern, yah, that would have taken a helluva long while by trial and error.

I used to own this Soyo Sauce motherboard:
https://stason.org/TULARC/pc/motherboards/S/S … 6A-SY-019B.html

I think it's about the same vintage as yours. Late '91/early '92. I bought it because I wanted a 386 board that could accept and cache 64MB of RAM.
My experiences with the DLC type CPUs were largely the same as yours. I think I tried both DRx2 and SXL. Neither were stable, even with all BIOS timings set to the least aggressive.
DLC type chips supposedly don't play well on older boards that don't support hidden refresh...but these boards do, so who knows.
Maybe it's just Soyo specific. Maybe something was fixed in a later revision of the board/BIOS.

I continued playing with my Asus VL/I-486Sv2GX4 @160.

I set the VLB UMC controller to the fastest settings with the jumpers and used a CF flash card and get a bit more than 4MB/s in speedsys.

Annoying: The controller does not have an own BIOS, the the Asus mainboard BIOS maxes out at ~7.8GB. I used a 16GB CF card, so can use only half of it.
The parameters can be seen in the speedsys screenshot.
So the last thing to opimize is another controller with BIOS. Higher throughput would be welcome, too.
Or can somebody fix the BIOS to allow 30GB or something?

But the last instability of this system was resolved: Before I had problems to read my Windows 98 CD, there were errors sometimes. Now resolved: On the VLB IDE controller there is only the CF card, the CD drive (which is a DVD drive actually) is connected to an ISA Aztec sound card's IDE port. As a side effect there is now also sound.

I ran the DOS benchmarks again. Values were basically the same, but two are a tiny bit higher now:
doom max details: 1124 or 1123 realticks => 66.5 fps (before 65.9).
Wolf 3D got a tiny bit higher, too. Please note the new world record for 160MHz 486 😁

I guess a SCSI VLB would be good if you do not go above 40mhz anyway

@bitwrangler

These 100MHz BL3 chips are not exactly on par with 486DX2.
If i had to do meaningful work or play games for real, there is no way it will be on BL3 than 486DX2.
That's for sure.
So, much like you, i am not sure about that note from the linked web site. : )

@Anonymous Coward

This exact model was on Ebay recently.
Contemplated for a moment about it, but quickly moved on - the low level of integration didn't look powerful enough to my eyes.
I assume it can cache 64Mb RAM in write-through mode, correct ?

@CoffeeOne

Very cool.
Updated the Wolf3D chart accordingly.
So, what exactly made it even faster ?
Spell it out please.

As for the "limited" local storage.
Can you fill even 4Gb with 486 class software ?

@frankmonk

Ok, let's see. Some more detailed break-down from what i have seen so far:

XT
Never tried SCSI controllers there. No idea what to expect.

286/ISA, up to 22.5MHz, 0-wait state, CF card, Acard adapter for SCSI
With the fastest SCSI controller Adaptec 1542CF: SCSI > IDE
Many other SCSI controllers can easily be slower than standard IDE ones, it is a toss really: SCSI == IDE

286/ISA, 25MHz or higher, 0-wait state, CF card, Acard adapter for SCSI
Technically Adaptec 1542CF is still faster than IDE (by increasingly smaller percent), but unfortunately all SCSI controllers become unstable: IDE > SCSI

386/ISA, CF card, Acard adapter for SCSI
No overclocking, or modest overclocking: SCSI > IDE
Extreme overclocking - SCSI is still faster than IDE (by increasingly smaller percent), but all SCSI controllers become unstable: IDE > SCSI

386/VLB, CF card, Acard adapter for SCSI
This one is easy: EIDE > SCSI

486/VLB, CF card, Acard adapter for SCSI
This one is easy: EIDE > SCSI

Socket 4/5/7/8 class hardware
No idea.

Slot 1/2, CF card, Acard adapter for SCSI
No overclocking, or modest overclocking - it is a toss: SCSI == ATA
Extreme overclocking - SCSI controllers become unstable: ATA > SCSI

Later retro computing platforms
Don't have enough sampling points for a credible statement, but in general SATA > SCSI for both HDD and SSD. Sometimes by a lot.

present time
SCSI is no more, but there is SAS for datacenters ...

-

That is of course mostly from the point of view of a personal home computer, or a workstation.
Datacenter use cases are outside the scope of the above summary.

On a related note:
Performance is mostly limted by the data storage device.
Up until year 2000 SCSI HDDs were faster on average than IDE devices. This was decisive factor in favor of SCSI.
The arrival of the 10k rpm ide drives leveled the field.
Today, using a fast CF card or other solid state device eliminates the bottleneck for these old controllers and enables their interfaces to show the best they are capable of.

It's not really fair to compare EIDE and SCSI controllers on VLB in terms of maximum transfer rates, because to the best of my knowledge there were no wide SCSI VLB adapters, so capped at 10MB/sec (which admittedly was a lot back then). SCSI on VLB was kind of flawed from the beginning anyway.

Even if you don't use SCSI for speed, there are still other nice reasons to have it. But, if you didn't leave through the SCSI era you can easily be forgiven for not caring.