486 class motherboards with Intel Pentium Overdrive P24T (POD100) processor running at 100MHz (2.5x40)

--- motherboards

Used the best representatives of this class hardware (as well as some commoners), which gives good sampling range and level of confidence for any drawn conclusions.

Motherboards that worked with POD100 processor:
Asus VLI-486SV2GX4 revision 2.1
Biostar MB-8433UUD-A rev.2 with 1024Kb L2 cache
Biostar MB-8433UUD-A rev.3.1 with 256Kk L2 cache
Chaintech 486SPM M104
Chicony CH-471B (not fully stable)
Chicony TK8880F / 2066A2
DTK PKM-0033S E-0
DTK PKM-00395 (0039S)
ECS UM486V revision 2.0 picture
FIC 486-PIO3
Gigabyte GA486AM/S revision 2.2
LuckyStar LS-486E rev:C2
LuckyStar LS-486E rev:D
PC-Chips M912 V1.7
PC-Chips M918i revision 1.2
PC-Chips M919 V3.4B/F
QDI V4P895GRN/SMT 1.0
Shuttle HOT-433 version 4
Soyo 4SA2
TMS PCI-400-4
Young Micro Systems mb-sw486g-3vl-v11
Young Micro Systems vs486g-3vl (Vega Green PC)
Zida 4DPS version 2.1

Motherboards that worked with POD83 only:
Abit PB4 rev 1.3
Abit PB4 rev 1.5
ASUS PVI-486SP3 rev 1.22
Chaintech 486SOM M102
Chicony CH-881A revision 1.0
ECS VIP something something
Luckystar LS-486E rev:F
Soyo 4SAW2
MS-4144 ver 1.5
Unknown brand TDVIP-2

Motherboards that don't have support for P24T, or didn't work with it at all:
Addtech Galaxy II 485 rev 1.0
Chicony CH-498B revision 3.0
Edom International Corporation 486VL3
FIC 86-VIP-i02
Octek Hippo 12 VIP
Opti-495SLC 3406 rev 1.1
Pine PT-429

Skipped on bunch of ISA only and some early VLB boards.

--- test setup

-- all motherboards

Complete stability implied (with one small exception for M912).
Made sure systems are fully optimized - taking the most out of it.

L1 cache in WB mode delivered better performance for all motherboards.
M919 and M912 are exception since they do not support P24T cpu type. Only P24D, which goes by L1 cache in WT mode.

Used the maximum possible amount of L2 cache (except for Young Micro Systems mb-sw486g-3vl-v11)
L2 cache set to WB or WT mode depending on which one delivered better performance for the given motherboard.

Used 32Mb RAM as a common denominator, considering PCBs with 256Kb of L2 cache in WB mode.
Used EDO or FPM RAM type depending on which one delivered better performance for the given motherboard.

Used the fastest video card that produced best results on the given motherboard. At the end only 3 cards were used - Ark1000VL VLB, S3 Trio64 VLB and MGA PCI.
Didn't run accelerated interactive 3D graphics tests like GlQuake since SIS based PCI motherboards work with Voodoo3, UMC PCI based ones can go up to Voodoo2 only, VLB boards are out of luck entirely. There is no common ground for comparison. Also 3D graphics are entirely CPU bound on this class hardware, so performance will be more or less reciprocal to the findings below.

Used on-board IDE controllers if available.

Used 2Gb Transcend x133 CF card with 512Mb FAT16 partition.
Some late 486 motherboards can achieve better IDE metrics with other types of CF cards, but this one is a good common denominator, so it was used.
For example the Biostar MB-8433UUD-A rev.3.1 SpeedSys screenshot below shows IDE speeds in the range of 5Mb/s, while with Sanddisk Extreme CF it can go +1Mb as shown in the original POD100 post (link above).
Leaving the subject for another time. It requires proper testing of different mechanical HDDs, CF cards and SSDs to find the upper limit for each motherboard/controller.
At this time i go by the common denominator as a basis for comparison.

(UMC based motherboards only) IBC DEVSEL# DECODING = MEDIUM (best is FAST), otherwise POST does not complete.
Unless specified below, BIOS timings and jumper settings are set for best performance.

-- per motherboard

Asus VLI-486SV2GX4 revision 2.1 with Promise EIDE2300 Plus VLB EIDE controller

Biostar MB-8433UUD-A rev.3.1 with 256Kb L2 cache with IDE driver version 3.2 and /D0:16 /MM0 /F0 flags.

Biostar MB-8433UUD-A rev.2 with 1024Kb L2 cache
L2 CACHE WAIT STATES = 3-1-1-1 (or higher), otherwise BOOT does not complete.
SpeedSys hangs on "YEAR 2000 bUG".

Chaintech 486SPM 104

Chicony CH-471B
L1 cache in WT mode - motherboard does not support P24T with L1 cache in WB mode. Can be forced with JP29 in position 2-3 but system becomes unstable.
L2 cache in WB mode.
Appian ADI/2 driver.

Chicony CH-498B revision 3.0
Inherently unstable, but was able to capture a SpeedSys screen at least.

Chicony CH-881A revision 1.0 with Promise EIDE2300 Plus VLB EIDE controller (driver settings on max - /T /M0:8)
L1/L2 caches in WB mode.
System kind of works but is not fully stable. What works is mostly simple DOS interactive graphic tests.
Few numbers for reference: Wolf3D: 125.5 fps, PC Player Benchmark: 29/8.7 fps, Doom: 56.7 fps

Chicony TK8880F / 2066A2 with Holtek VLB EIDE controller and driver with /D0=2,2 flag (fastest).

DTK PKM-0033S E-0

DTK PKM-0039S with Holtek VLB EIDE controller and driver with /D0=2,2 flag (fastest).

ECS UM486V revision 2.0
L1 cache in WT mode since motherboard does not have proper support for P24T. Only for P24D (L1 WT).
Promise EIDE2300Plus controller with driver. All settings on max.

FIC 486-PIO3
L2 cache in WB mode with timings 2-2-2.

Gigabyte GA486AM/S revision 2.2
L1/L2 cache in WB mode. No IDE driver used.
Chadti99 tested briefly this board. Details in this post.
Later on i was able to stabilize the motherboard here, but only with 128Kb L2 cache. This particular board didn't age well it seems.

LuckyStar LS-486E rev:D (LuckyStar LS-486E rev:C2 is basically the same)

PC-Chips M912 V1.7 with Holtek EIDE VLB controller and EIDE driver with /D0=2,2 flag (fastest).
L1 cache in WT mode since motherboard does not have proper support for P24T. Only for P24D (L1 WT).
It was last on the list and extremely fussy about L2 cache. Tried hard to overcome that, but was already running out of patience.
I feel it is possible to find the desired combination so the board can handle well the Windows OS and related tests.
Leaving it for another day.

PC-Chips M918i revision 1.2 with the only know IDE driver for ALI M1489/1487 - ALIHDD.SYS and /T /S20 /PIO:4AAA /IOW:YNNN /BKM:YNNN /LBA:YNNN flags.
L2 cache disabled for best performance. This was covered extensively in previous posts.

PC-Chips M919 V3.4B/F with IDE driver version 3.2 and /D0:17 /MM0 /F0 flags.
L1 cache in WT mode since motherboard does not have proper support for P24T. Only for P24D (L1 WT).
Ark1000VL didn't work at all.
S3 Trio64 was the second best VLB video card that produced better results in some of the tests than MGA PCI.
When 1Mb L2 cache module is used DRAM READ/WRITE STATE must be set to 1, otherwise complex compute tests start failing.

QDI V4P895GRN/SMT 1.0 with Ark1000VL and Promise EIDE2300 Plus

Shuttle HOT-433 version 4 with IDE driver version 3.2 and /D0:15 /MM0 /F0 flags.

Soyo 4SA2
There are multiple BIOSes available for this motherboard and its cousins (this is the most advanced PCB of the series).
For PCI graphics the best BIOS seems to be SA-0730, for VLB graphics - IO-B1
Tested with MGA (PCI) and Ark1000VL (VLB). Ark1000VL must be set to 1 wait-state or the system does not POST. Even with 1 WS this is the fastest VLB card for this board.

Young Micro Systems mb-sw486g-3vl-v11
256Kb L2 cache was used (support for up to 1024Kb). Above that size leads to major instabilities, especially with tight timings. Tried hard to overcome that. No luck.
SpeedSys hangs on "YEAR 2000 bUG".

Zida 4DPS version 2.1 / TMS PCI-400-4
TMS PCI-400-4 is less stable than Zida 4DPS version 2.1 in general.
Screenshot shows no L2 cache, but it is actually there. Delta is too small for the limited vertical resolution of the graph.
Performance is slightly lower if L2 cache is disabled from BIOS.

--- benchmarks

For clarity - the WinTune2 test is for Windows accelerated GUI.

Few things to notice.
VLB > PCI for interactive DOS graphics. Not a big surprise here. It was already established by Am5x85 based tests in previous posts. Now confirmed and with POD100 CPU.

Small format SIS based boards are consistently last in line.
It was the same with Am5x86 processor.

Notice the Chicony guy.
One of many with Am5x86 at 160MHz.
Did well in the DOS tests with Am5x86 at 180MHz, but completely failed in Windows and complex computation.
Didn't work with Am5x86 at 200MHz.
But with POD100 - surprise surprise. Who would know.
Top notch interactive DOS graphics and Numero Uno in the most complex test - LW3D rendering.

Also, notice DTK's PKM-0039s.
Surprisingly good performance. Suddenly VLI is not that far ahead in DOS graphics and IDE speed.
Best accelerated Windows GUI with VLB hardware to date.
While numbers look great, here is the bad - it does not support Am5x86 CPUs.
In fact it does not support Intel 486DX4-100 CPU, only up to i486DX-75.
Worse, it has an issue with 1024Kb L2 cache. Frequently enough it will refuse to boot until L2 cache is disabled from BIOS, then restart, then enable it again, then restart for complete boot.
This is not related to BIOS timings or other settings - just BIOS gets in a weird state and either needs to be cleared, or the above exercise has to be performed.
So, cannot recommend that board really.

Otherwise the story repeats itself - Asus VLI-486SV2GX4 is again the highest Socket 3 place, this time with POD100 processor.

There will always be new iterations of CPU, RAM size, cache size, benchmarks, motherboards, other hardware, etc to add to any roundup. The feeling of being completeness is only fleeting.

Were you not able to get the UMC FIFO/MasterMode driver working at 40 MHz?

Were you able to confirm that with the POD installed on all those motherboards that L1 write-back cache was indeed working? That is, if you set L1:WT you notice a substantial performance hit?

Of course.
Did I make it sound too absolute ?
I certainly don't see it that way.

Some of the UMC based motherboards can handle IDE drivers with high speed settings and BIOS timings on max when running at 40MHz base frequency.
Outcome differs based on used data carrier - CF, mechanical HDD, or SSD. Also, there is different models of them.
Finding "exact facts" is and will be convoluted.
At this time used the on-board controllers as-is - as an indication of raw performance.
The whole IDE/EIDE/SCSI thing needs a dedicated study.
Leaving it for another day.
Volunteers wanted.

Yes.
Captured data for L1 and L2 caches in WB and WT modes. Shared best outcome only.
I invite you to cross check any results - just in case.

Were the m919 v3.4 results with L1:WB properly enabled, or in WT mode?

L1 in WT mode since there is no proper support for P24T type CPUs. Only P24D.
Added that note to the original post.

Also, will capture and share 2-3 more SpeedSys screens for completeness tomorrow.

Some really good info here, thank you for sharing these results.

I am curious why your Gigabyte GA486AM/S revision 2.2x wouldn’t run a POD at 100MHz? I have one that does just that at the fastest cache and memory timings. Might your board need a recap or a different bios or perhaps the UMC chipset is a different version to mine.

He already recapped his AMS board.

Are you able to indicate which boards were able to run with L1:WB and which had to resort to L1:WT ?

feipoa wrote on 2021-08-21, 09:06:

I've never been very good at the wick. Maybe I have junky stuff.

2-1-2 at 60 MHz would make for a nice Cyrix 5x86-120-S1R3 system if all the selected PCI cards worked at 60 MHz. If you have another rev.D, it would be interesting to see if 3-1-3 is completely stable with 15 ns. I suspect it is.

I can confirm 3-1-3 is completely stable at 60MHz for me with stock 15ns chips on Rev.D. Looking into sourcing the 8ns chips.

feipoa wrote on 2021-09-28, 08:04:

It might be helpful for you to get a socket 3 interposer module and setup a trimmer on it so that you can have variable 3 to 4 V adjustment on any system.

Where might I find one of these, would I need to source it from another 5x86 upgrade package? I really like this idea since it will keep me from having to modify the mainboard.

I really think I have a shot at running an AM5x86 stable at 200MHz on air cooling. She runs great at 3.45v at 180 all day. She needs more voltage to boot at 200MHz on air cooling, even makes it to the dos prompt. If I could just dial back the voltage from 4v, which I’ll try, it just might work out.

@chadti99
Thanks.
Can you run the common benchmarks above on your GA AMS board and share the numbers with clarification about l1 and l2 cache modes ?
I will look one more time at why this was a problem here.

About LSD, yes 3-1-3 is fine with 15ns L2 chipswt 180 and even 200 mhz.

The most magical AWD cpu here can boot to dos at 200mhz with 4v, but is inherently unstable.
Surface temperature needs to be below 7-10C for DOS to be ok. It needs to be below 2-3C for complete stability.

@feipoa
The only two boards that ran L1 in WT mode where m919 and m912. Will post today the rest of the speedsys screens and add notes accordingly.

Updated the post above with more all speedsys screenshots, also adjusted layout for better readability.
M918i and HOT-433 still need screenshots with applied UMC IDE driver.

pshipkov wrote on 2021-11-28, 16:35:

@chadti99 Thanks. Can you run the common benchmarks above on your GA AMS board and share the numbers with clarification about l1 […]
Show full quote

@chadti99
Thanks.
Can you run the common benchmarks above on your GA AMS board and share the numbers with clarification about l1 and l2 cache modes ?
I will look one more time at why this was a problem here.

For sure, I’ll find some time this week.

feipoa wrote on 2021-08-20, 12:38:

Soldering on 9 SOJ-28 without solder paste was not fun. My solder paste had expired. Anyone know how long solder paste is still usable past the expiration date when stored at 20-25 C ?

BitWrangler wrote on 2021-08-20, 13:27:

For hand work and rework it doesn't really expire, the fluxes evaporate off a bit, so if it gets too stiff, just add a touch of liquid flux or a couple of drops of isopropyl alcohol. What expires is it's ability to have precise properties necessary to perform in an automated PCB production scenario where it is stencilled onto PCB, leaving "just exactly the right amount" on the pads, too sticky there's too much, too wet it bleeds under stencil, then stick the components down such that the vibration of the line doesn't send them skittering, and melt precisely on schedule in the reflow oven. If you're not doing a process that relies on any of that, just use it whatever the date is, re-wet it if it dries out.

pshipkov wrote on 2021-08-20, 18:23:

As for how quickly the soldering paste goes bad - in my experience - if not properly packaged - few months.
By proper packaging i mean - put the syringe inside (deflated from air) plastic bag with zipper and store it in a drawer.

feipoa wrote on 2021-08-20, 22:11:

OK, I guess I'll pull out my expired paste and see how it acts.

Yesterday I pulled out my expired solder paste. It had never been opened. It has a mfg date of 25 July 2018 and is of formula SMD291SNL. That stuff was way too viscous to be used with fine tipped syringe tips and fine pitch SMD components. It works, but because I had to use fatter tips, it puts too much paste on QFP pads, doesn't adhere to the PCB well (in the un-heated state), and causing bridging.

Now you say I just have to add some flux to the solder paste? I have CHIPQUIK CQ4FL no-clean liquid flux and CHIPQUIK SMD291 no-clean paste flux. Do you propose I open up the SMD291SNL syringe and add the liquid flux or the paste flux? If so, which of those two options (liquid or paste flux) and how much of it? Then stir it with a tooth pick I gather.

I also experimented with some solder paste I had ordered a month ago. I bought a different formula this time because it supposedly has a longer shelf life. I bought TS391AX with mfg date of 24 Aug. 2021. I had no problem using the ultra fine tips on the syringe with QFP pads with the new paste. It comes out a lot more liquidy.

Some experimental notes here - if the solder paste is too old and you have to apply it thick and grobby like, I was able to reduce the amount of bridging by using an extra fine tipped iron and touch each pad one at a time, that is, opposed to using hot air. This is slow and you still sometimes get bridges.

Something feels off about how liquid the paste you have is.
Mine has tooth paste like viscosity. But when it goes old/bad it thickens and gets ... brittle. It becomes difficult to squeeze it out of the syringe.
If you already soldered something in the same area of the PCB it can get hot/warm enough so the next round of putting paste on it makes it instantly go loquid.
But if i keep the amount just right (minimal) even if it gets liquid when touching warm/hot surface i rarely get crossing upon heating it.
Also, it always sticks pretty well. Never had to add flux.
Currently i using 247-LOWTEMP-15 type (Sn42/Bi57/Ag1).

The liquidity of the new paste looks pretty much the same as all the youtube videos I've seen. It is maybe a tad less viscous than the green sensodyne toothpaste. While my old set of paste is more like cookie dough, or perhaps epoxy that is starting to cross-link. I can't work with it. Thus I am curious about BitWrangler's comments to add flux (liquid or paste?) to soften it up. Have you tried this yourself with old solder paste?

Yup, I know what you mean by the paste going on smoother if the PCB is already warm, but that wasn't the case when I described the viscosity of my two pastes (the new and the old). I was, however, wondering if I can still use the old solder paste by first heating the syringe to the right temp to soften it up.

I ran some tests, and it looks as if his idea to mix in flux might be valid. I ran two tests, one with liquid flux (from a flux pen), and one with flux paste. The pen and tube are shown, along with my dried out solder paste.

The round nub is the plunger removed from the syringe. You can see how it is cake like. When I mixed the liquid flux with the caked solder paste, the solder paste seemed to disappear - like the volume decreased by 75%. It also didn't get more creamy, but rather splotchy.

On the other hand, when I mixed the SMD291 flux paste with the solder paste, the consistency of the mix made it appear new again. See photos. The paste shown on the toothpick was picked out of the syringe and is the cakey old paste, same as the plunger, as shown.

I will try to mix in some flux paste into the syringe and mix it up, then solder some PLCC chips and report back.

I never had any that got to the crumbly dough stage. I think you're getting the same problems as are experienced in trying to re-wet clay or mix eggs into flour. It either needs to be done a little at a time, or mechanically agitated (like a blender) very vigorously if you just put it all in at once. Another way might be similar to making smooth gravy, in that it's easier to work with a small quantity, getting it worked into solution well, like the flour in a cup of water, then mixing it with the larger quantity.

GA486AM/S ver 2.2 AMD180 (60x3)
1024k L2 15ns
32MB 60ns FPM Samsung
ARK2000PV
3222 1rw
L1 WB L2 WB pci 1:1/2
(Couldn’t get past post at pci1:1, even with PCI IDE controller)

JP15: 1-2 open 3-4 close 5-6 open = 60MHz FSB
JP15: 1-2 close 3-4 open 5-6 close = 66MHz FSB

Dosbench
1. 3dbench 11.1
2. 3dbench fast 102.2
3. Chris 240 80.5
4. Chris 480 23
5. PCP 240 30.1
6. PCP 480 11.7
B. Doom Max 1260
C. Quake 20

GA486AM/S ver 2.2 POD100
1024k L1 15ns
ARK2000PV
32MB FPN 60ns Samsung
2111 0ws
L1 WB L2 WB pci 1:1

1. 3dbench 0.0
2. 3dbench fast 97.6
3. Chris3d 240 77.2
4. Chris 3d 480 23.2
5. PCP240 29
6. PCP480 11.3
B. Doom max 1320
C. Quake 26.9

Excellent. Thanks Chadti99.
Will add your data to the charts and notes in the original post above (along with info for another board that i missed).