Yes.
I was about to link a post i thought i made in the world's fastest 486 thread but apparently never did.
The only thing i found was the prelude to what i intended to post back then - Re: The World's Fastest 486
Too bad.
Quake 1 can tick north of the 22 fps, but never got the system fully stable.

One thing i noticed is that VLI handles processors somewhat differently than the late PCI 486 motherboards such as LuckyStar LS-486E revision C1/C2/D, Biostar UUD and the rest.
For example, BGC chips that never ever lit up on any of these boards actually work on the VLI at 200MHz, 4V and ... air cooling.
One of them can actually do it at 3.6V but L1 cache gets super fussy.
Pulled today a spare VLI board to verify it was not just my imagination back then.
Captured a quick video.
Was not prepared for 200MHz exercise, so didn't bother with installing L2 cache - PITA @ 200MHz.

So, if you have spare Am5x86 processors - give them a VLI try - some of them may surprise you.

---

jp20 controls cpu voltage autodetection if shorted, but not sure if 5v can be forced.

2.0 is more restrictive than 2.1 about cpu voltages - i think that's the only difference between the two versions.

Btw, how your jumper settings look like ?

---

If you go for AWE32 - consider the ct3980 model.

Hi pshipkov, completely missed this. Here is where I left the jumpers. Just change the bus speed jumpers from 40 to 50MHz. I need to order more bright colored jumpers, let me know if you need me to clarify any.

Thanks.
There is no bigger story here.
Wanted to check if we are doing the same thing since there are several different sets of jumper settings floating around.

Had an idea for securing a heat sink to one of these Lucky Star LS486 boards. I think this combined with thermal paste and maybe a few drops of thermal adhesive should hold the cooler in a tower case. I need something to tie the thread too for sure. And yes this board needs a deep cleaning 🤣. Is there any chance cotton/polyester thread would short the pins it’s touching on the CPU?

Chadti99 wrote on 2022-03-12, 13:15:

Had an idea for securing a heat sink to one of these Lucky Star LS486 boards. I think this combined with thermal paste and maybe a few drops of thermal adhesive should hold the cooler in a tower case. I need something to tie the thread too for sure. And yes this board needs a deep cleaning 🤣. Is there any chance cotton/polyester thread would short the pins it’s touching on the CPU?

Interesting approach... 😀

Rather than going through the hole in the board, what you could do instead; don't use the hole at all, just bind the heatsink to the CPU, i.e. place the CPU (pins down) on your workbench (apply the thermal paste etc, but you might want to give it a few practice runs before finialising), position not 1 but 2 wires underneath the CPU, both positioned on either side of the CPU to line up with the last square indent of your heatsink (you see how at one point the indents become round?). Then just wrap around the wires, pull tight, and tie knots so that the knots are on top, i.e. inside the heatsink indents. This will allow you to install or remove the CPU with heatsink from the socket.

PS: You could do this both horizontally AND vertically for extra strength (i.e. 4 wires in total).

Ghetto style always works.
Alternatively you can get rid of the thread and squeeze two small drops of Loctite in the opposite corners of the CPU - that will hold the heatsink + fan really well but without preventing you from removing them when you want by applying a bit of force to the heatsink.
I did that for the LSD based PC - worked great. If you think about it - that approach makes it easier to separate CPU from cooling.

---

Finally found the motivation to pull that IODATA PK-A486BL3-1 386 CPU upgrade module which sports Blue Lighting 3 CPU and give it a spin.

The story here is that all 386 CPU update modules i know of are problematic in some way.
They don't work well on fast/good/late 386 motherboards - plagued by FPU / L1 cache / other stability issues, or no lights entirely.
This makes them largely useless since all the benefits they bring get wasted by the slow/early motherboards they work properly on.

Which leads me to the motherboard in use:
Rectron Electronic Enterprises 386DX-33 PEAK DM
Based on Chips and Technologies FX82C351/5/6 chipset.

Simply put - it is a clunker.
Like the rest of those Electroglass, Peak and other mobos of the same type - wide PCB, low level of integration, slow chipset = very low performance.
The system cannot take more than 70MHz crystal oscillator, which results in maximum 35MHz CPU frequency.
There is a second crystal for ISA speed. It was 32MHz by default. Bumped it to 46.16MHz which improved performance. No lights past that.
The picture above shows 256Kb cache installed but had to run with 128Kb instead.
These type of boards require 64kx4 TAG chip to get you to 256Kb L2 cache.
I don't have 64kx4, so 128Kk it is.

ET4000/W32i and Cirrus Logic GD-5434 based video cards don't work on this board.
Used ET4000AX instead.

With BL3 CPU the system cannot soft reset and often requires 2 hard resets to boot back into DOS.
The IODATA's own drivers never worked.
Used Evergreen's REVTO486.SYS.
The /2 or /3 flags hang the system.
Used the module's on-board switches to establish 2x multiplier which results to 70MHz to CPU.
Tried to overcome the issue and get the CPU running at 3X, but without success.
Regardless of BIOS settings, even very low base frequency or presence/absence of FPU the driver hangs the CPU if /3 argument is used.
Not too bothered by that because higher base frequency with 2x multiplier is better than lower base with 3x multiplier anyway.
(Will fuss some more with this to see if /3 is possible for the sake of it. Any ideas ?)

Here is what LM6 and SpeedSys say:

Working L1/L2 caches.
Working FPU.

Couple of benchmarked numbers.
Compared against DTK-PEM-4036Y + TI 486 SXL2CPU running at 50MHz natively (shared info in this post.

1                 wolf3d     Superscape    PcPlayer    Doom
2Peak+BL3 @60MHz: 35.6 fps   30.3 fps      8.3 fps     14.8 fps
3DTK+SXL2 @50MHz: 65.4 fps   35.7 fps      9 fps       18.1 fps

(will try to swap this with the usual bars for better visual clarity)

Captured quick test run on video showing LM6, SpeedSys, Wolf3D, Superscape, PcPlayer Benchmark, BIOS settings.

Didn't bother with Windows GUI tests because of the problems with fast ISA GUI accelerators.
Didn't bother with other more complex tests. No need to waste hours to confirm that things are slow, despite the BL3 cpu helping a lot.
Didn't bother running tests with other cpus either.

---

This excersise was first and foremost to showcase a fully working BL3 cpu module since there is not much info around here about them apart from few posts by Feipoa.
Most of the shared stories are about problems and partially working stuff.
It is possible to run them properly, but as stated above, mostly on early/slow motherboards which makes the whole thing pointless.
In general, for those of us with 386 hybrid upgrades itch, Alaris Leopard/Cougar mobos seem to be a much better playground.

I highly recommend going through the entire Revto486 manual published by Evergreen. There are literally dozens of register options and commands with walk-throughs for various conditions. I tested my BL3's years ago and the information is no longer fresh in my mind.

I also have dozens of pages of hand written notes concerning the Evergreen Revto486 driver, caveats, and settings for the CTCHIP34 utility w.r.t. the IBM BL3. It will take some time to go through all of them.

I've pulled out some of my notes on the IODATA module.

SW1
ON = flush cache on every DMA and every I/O read/write
OFF = flush cache only on every DMA
Setting SW1 to ON completely kills the performance. L1 is approximately useless in this case. When SW1=ON, BL3 at 110 MHz approximately equal to properly configured SXL2-50.

SW2
ON = enable the 2x multiplier on the Evergreen PCB (not in BL3 die)
OFF = normal 1x mode

SW3
ON = co-processor disabled
OFF = co-processor enabled
there is mis-information for this setting online. Above is the correct settings.

SW4 = not used.

pshipkov wrote on 2022-03-06, 22:37:

Quake 1 can tick north of the 22 fps, but never got the system fully stable.

Would you be willing to provide hardware information, repeatability instructions, and general evidence of this in the World's Fastest 486 thread?

I have the available utils and documents from this forum and the few other places online. Plus some fresh info shared by somebody an hour ago. ; )
My experience is inline with your description.
0-1-1-X is how i use the CPU module currently.

Looking forward to see your VLB/BL3 contraption coming to life.
The ALI chipset is not exactly a performance beast, but is for sure much better than most other boards these CPU modules work reliably on.

--- VLI @ 200 MHz

There is not much to add really.
Use the mobo in standard configuration for 50MHz and 4x multiplier.
If the CPU has the chops there will be lights.
The good thing is that somehow more processors end up lighting up and don't even require active cooling. Peltiers don't make a big difference which is quite different than with late 486 motherboards.
But the system is inherently unstable. At least i was not able to achieve even decent level of stability.
Will try to get at least Quake 1 completing on a video. It does not happen very often which can be time consuming and frustrating.
I thought back then that stability can be improved on, but at some point lost interest and yet to get back to it for real.
Looks like @Chadti99 is hitting similar issues.

If you think these notes are worth it - i can post them in the other thread.

My ALi PANDA board does infact lighting up with my Buffalo BL3 and the world's tallest sandwich. I'll be working on it some this evening.

I found these in my notes:

To eliminate the A20 control issue with himem.sys, you need to add /machine:1 to your config.sys ling, e.g.
DEVICE=C:\Windows\HIMEM.SYS /testmem:off /machine:1 /v

Also, it is that we want XTOUT disabled, not enabled, to increase DOOM performance. The CTCHIP34 command line to set XTOUT=0 is CTCHIP34.exe IBM486.CFG /1000h:1=$8c

Also, the L1 enable in my particular BIOS sets MSR1000 byte 0 to 80h, which simply enables the CE bit (sets to 1), e.g. MSR1000 = 0000 9c80, with "byte 0" being "00" previously and changed to "80".

The PAL on my Buffalo unit doesn't adjust any of the BL3 register settings and I don't really know what the jumper on my Buffalo unit does, but disabling it causes a parity error. I have also noted that although I have the 87DLC on my Buffalo unit's interposer, I still need the FasMath inserted in the Co-Pro socket. This seems like a really bad idea in practice, but the FPU doesn't work otherwise.

The Evergreen driver does a lot of things and it doesn't really tell us what it is doing unless with look at the register readout.

Evergreen sets this up:
CR0 = 0000 FFF0
CR2 = 0000 0000
CR3 = 0003 0000
MSR1000 = 0000 9C92
MSR1001L = 0000 03FF
MSR1001H = 0000 00F0
MSR1002L = 0400 0000
MSR1002H = 0000 047F

And the Default configuration (L1 disabled in BIOS and no evergreen driver):
CR0 = 0000 FFF0
CR2 = 0000 0000
CR3 = 0003 0000
MSR1000 = 0000 9C00
MSR1001L = 8000 83FF
MSR1001H = 0001 F2FF
MSR1002L = 0400 0000
MSR1002H = 0001 047F

You can use the information I provided to dissect these bytes into binary to determine what each does. For example, you will notice that the Evergreen driver, with setting MSR1001H = 0000 00F0, the F0 is setting the cacheable range to 15 MB instead of 16 MB. I guess they are doing this because some hardware sets their ROM in this range. To make it fully cacheable, this would be set to FF. EDIT: Nope. I forgot that this cacheable range is to start from the 1 MB boundry, so 15 MB from 1 MB would mean the full 16 MB cacheable. The cacheability of the first 1 MB is controlled by a different register, 1001h bytes 0 & 1, or bits 0-15.

Another example is MSR1000 = 0000 9C92, the 92 sets CE=1, ASNP=1, CPC=1. This all will take a very long time to go through.

My results with the VLI @200MHz are consistent, stability wise, with the CPU I was using with air cooling at 4volts on other boards. If I knew for certain I could force the voltage to 5v I might pull my 200MHz stable CPU for further testing. I’m almost afraid to touch it as it’s running very stable on the Biostar UUD at the moment.

In other news Pshipkov’s hunch that the 0.1-0.2 fps difference between our Quake runs Was due to our differing DOS versions was absolutely correct. DOS 7 consistently scores 0.1fps slower in Quake benchmarks on both the LSD and UUD boards compared to DOS 6.22. I still need to test on the VLI.

At seemingly 100% stable settings, on the UUD, I see 21.8 vs 21.7 fps in Phils Quake Bench. This is at 200MHz 4x50 256k, fastest timings, and 1:1 CPU:PCI ratio.

Pic of build so far, note I had to drop the Tseng in favor of a Matrox G200 as it has issues at 50MHz PCI speeds. I’d really like to find an AT case in good condition with a triple digit LCD but I like the compactness of this case. I’m also tempted to mod a window as I really enjoy seeing my components and find myself running with the lid off.

pshipkov, I have some benchmark results for you to compare with.

CONFIGURATION
Daewoo AL486V-D w/BIOS 2600P v1.9, aka version 299 from Leading Edge Products
Buffalo BL3 at 75 MHz (3x25 MHz)
using the IBM BL3 registers as setup by the BIOS
256K L2 cache
Trio64V+ VLB with hacked 0ws BIOS
Promise EIDE2300 VLB
8 GB 133x CF card
16 MB Samsung LGS GM71C17400CJ6 memory, 3-chip
BIOS settings maxed out except DRAM Read set to FAST (Fastest results in parity errors)

BENCHMARKS
Coretest = 4333.6 KB/s

DOOM = 3740 realtics, or 19.97 fps

Quake = 2.3 fps

Landmark v2 = 172.35 CPU, 168.38 FPU

Landmark v6 = 221.9 CPU, 150.67 FPU

Cachechk v7
L1 = 52.0 MB/s read, 51.6 MB/s write
L2 = 38.0 MB/s read, 24.5 MB/s write
RAM = 18.1 MB/s read, 8.3 MB/s write

CHKCPU reports 78.6 MHz

3dbench 1.0 = 40.0

PCPBench /VGAMODE = 8.1
PCPBench VESA MODE 100 = 3.5

Wolf3D = 69.4

Results are OK for a 25 MHz FSB, that is 20 fps in DOOM at 75 MHz. By comparison, the SXL2-66 scores 20.6 fps and the SXL2-80 scores 24.3 fps. Unfortunately, it appears as if my BIOS or something else I don't fully understand is insisting my Buffalo BL3 stay in 3x mode. I tried to force it out of 3x mode by setting EDFS=1, DFSREQ=1, then clockmode back to 1x (000), then to 2x (011), then DFSREQ=0. No luck. stuck at 3x. I don't understand why because the Buffalo could run at 1x and 2x on other motherboards.

Fired up my cased VLI system this morning and popped the same cap on the Promise 2300 controller as I did on my spare. Pure coincidence or something else going on? I swapped in my repaired 2300 and all is fine atm. Both failed on different motherboards and power supplies. This guy had worked for months trouble free. Maybe something to do with my SD2IDE adapter? Maybe just old parts?

That looks to be a tantalum cap and they are know to pop some times with spectacular results. They tend to turn into firecrackers with age

@Feipoa
I experimented in the past with these registers but didn't get anything definitive out of it.
The two outcomes were - no observable changes in behavior/perf, or stuff goes wrong entirely.

Similar to how your BL3 is stuck at 3x i am stuck at 2x.
But that's fine. At least the modules work with their full set of features.
For once.

This PEAK DM motherboard is not a racing horse.
It can pull a cart maybe. A small one.

Your numbers are more inline with the Alaris boards, so let's see:

Coretest (Mb/s):
ALI Panda 3x25: 4.3
Cougar 3x25: 6.7
Leopard 2x33: 3.8
DTK SXL2 1x55: 4.5

Wolf3D (fps):
ALI Panda 3x25: 69.4
Cougar 3x25: 80.6
Cougar 3x33: 99.8
Leopard 2x33: 76.6
DTK SXL2 1x55: 68

PCP Bench VGA (fps):
ALI Panda BL3 3x25: 8.1
Cougar 3x25: 8.5
Cougar 3x33: -
Leopard 2x33: 8.1
DTK SXL2 1x55: 9.2

Doom (fps):
ALI Panda BL3 3x25: 20
ALI Panda SXL2 2x40: 24.3
Cougar 3x25: 20.9
Cougar 3x33: 24
Leopard 2x33: 20.1
DTK SXL2 1x55: 20

In general SXL2 and perhaps DRx2 (never tried one of those) are much better upgrades than BL3 modules.
Where BL3 modules come to play is on slow motherboards that cannot overclock well - there SXL2 is really held back by 25 or 33 MHz bus and BL3 comes ahead.

---

@Chadti99, this is a good looking case, both on the inside and outside.
Outside is clean and simple. Inside is spacious.
First rule of retro hardware - if a rig is working well - DONT TOUCH IT.
So don't. : )
Btw, no LAN ?

Hope this VLB EIDE / VLI business does not cause collateral damage and you lose the mobo.

@chrismeyer6
This is the second time he blows the fuse.
You can check few posts ago.

EDIT:
Forgot to add that DTK numbers with SXL2 CPU running natively at 55MHz.
The big story is that it is on par with BL2/3 in DOS interactive graphics, but outdoes them massively in offline compute.
EDIT 2: Added 3 screenshots for reference.

That’s awesome, had no idea you could run Doom at playable frame rates on such hardware. The comment about the horse and cart is cracking me up.

Thanks for the kind words, I’ll def heed that advice for awhile and just keep it as is. I’ve gone back and forth on network cards, it makes a lot of sense for moving files around but it seems like it adds 5 minutes to Windows startup. If I had the space I could see me having a 3-4 node setup for multiplayer Doom, Quake, Duke3D, etc.

The crazy thing with the capacitor issue is that this just happened on entirely different sets of hardware. I have two of these 2300’s that have popped the same caps within a couple weeks of each other. I’m looking at the VLI setup with the ARK1000 and Promise 2300 and they are chalk full of these tantalum caps. I better keep a fire extinguisher nearby 🤣. The wife was asking what that horrible smell was as I tried to air out my office, uh sorry honey, burned the toast.

Here’s a couple pics of the current state of my VLI build. I need to work on cable management.

"The comment about the horse and cart is cracking me up."
Is the retro comedy central.

So, after an incident and following explanation do you get one of "that guy" eye rolls ?
Joking aside.
It is curious that the EIDE2300 keep blowing up on you.
I think you use the standard black IDE2CF adapters.
So there is nothing out of ordinary.

What is the top most ISA card ?

Fancy Gotek. Haven't seen one with front display.

The performance differential between the Cougar and Panda was larger than I expected it would be. Do you have cachechk results for your Cougar at 75 MHz?

Read benchmarks:
cachechk.exe -t6

Write benchmarks:
cachechk.exe -w -t6

The -t6 stops it from running through all the memory on your board. Otherwise, it takes too long to run through.

I might try to solder on an IBM BL3 chip that I know will run at 110 Mhz onto the Buffalo to see if I can get 100 MHz out of my Panda system. In my notes, I have written that the Panda+BL3, on occasion, will turn on at 100 MHz, but not POST. Not sure if this is due to the IBM BL3 CPU itself, or some other factors with the MB design.

Updated the previous post with numbers for DTK-PEM-4036Y with SXL2 CPU running at native 55MHz - one more interesting data point that i forgot to add - assuming it is a ISA only board. It is on par with the VLB boys.

Will post cachechk screens soon.
Read through this post for additional info.
Cougar's memory metrics are mediocre at best.
Cachechk and other synthetic benchmarks are completely misleading about its properties.
In reality it is very well integrated assembly that excels at DOS interactive graphics and local storage I/O.

Brief summary of the fastest 386 motherboards i touched so far:

Alaris Cougar + BL3, 3x25, 3x33
Summary: Fastest DOS interactive graphics, fastest local storage. Slow FPU and offline compute. Stuff just works at 3x25. Not fully stable at 3x33.
pros: 512Kb L2 cache, VLB, on-board EIDE, great assembly
cons: BL3 CPU hardcoded to 3x multiplier at start = no chance for 2x40 upon post

Alaris Leopard + BL2, 2x33
Summary: Very fast DOS interactive graphics and offline compute. Stuff just works.
pros: VLB, great assembly
cons: 128Kb L2 cache only, clock gen limited to 33MHz

DTK-PEM-4036Y + SXL2, 1x50, 1x55, 2x31
Summary: Very fast DOS interactive graphics. Fastest FPU + offline compute. Solid at 1x50. At 1x55 some ISA cards start getting flaky, needs magical FPU ( IIT 40MHz). 1x50/55 > 2x31.
pros: 256Kb L2 cache, great assembly
cons: BL3 CPU modules don't work

In my opinion, outside the Alarises and considering what a problematic pain in the ass these bl3 modules are, the best chance for peak perf is with SXL2 running at 2x40 on a very fast board.
Your SXL2-66 can prove that "theory".
Also, try the DTK Symphony that is sitting on the side.
Notice how mine is matching the VLB systems despite running at "only" 50/55MHz. In other tests not considered in this conversation it outdoes them handily.
Very high chance it spanks them with 80MHz CPU.

If one day we see bl3 module running properly at 3x33 on a fast board I will be proven wrong.
But until then I am inclined to bet against them.