Good info.
Thanks for the pic of the interposer. Makes it easy for others to follow.

So, the slick looking on picture devices turned out to be a bit flaky ? Okay. That's not great. Are you sure you didn't miss them ?
Asking this, so we don't contribute negative feedback to their product. You know, the Internet is a very small place.

Looking at the findings and results i am not confident the buckgen brings enough value, but your ending comment is positive.
Did you try to hit the same results using the out-of-the-box CPU voltage values that the motherboard offers.
Can you expand on this, if you have any further notes of course.

Next update.

By coincidence, on the HOT-433 rev. 4, I was able to run STABLE another Intel 486DX4-100WB at... 150 MHz using...5 Volt!
Call me crazy, but it really works! So, for this exemplar, the voltage-interposer is not necessary, as 5 Volt is provided by the mainboard 😉

This should be probably the fastes Intel-486-CPU in Windows 98SE ever...Please correct me, if I am wrong.

Sadly, at FSB 50 MHz, my L2-modules can still do only 3-2-3 (not even 3-1-3), so the performance in GLQuake and Quake II drops compared to the i486DX-WB at 133 MHz (FSB 66).

On the other hand, using 32 MB EDO-RAM with 50 ns, for FSB 50 MHz it was possible to have a DRAM Read Wait State of 0 (instead of 1 for FSB 66 MHz) in BIOS.

The results (still using GF2MX-400):
- GLQuake: 21,8 FPS (best one: 22,0)
- Quake II: 12,9 FPS

pshipkov wrote on 2024-07-22, 16:36:

So, the slick looking on picture devices turned out to be a bit flaky ? Okay. That's not great. Are you sure you didn't miss the […]
Show full quote

So, the slick looking on picture devices turned out to be a bit flaky ? Okay. That's not great. Are you sure you didn't miss them ?
Asking this, so we don't contribute negative feedback to their product. You know, the Internet is a very small place.

Looking at the findings and results i am not confident the buckgen brings enough value, but your ending comment is positive.
Did you try to hit the same results using the out-of-the-box CPU voltage values that the motherboard offers.
Can you expand on this, if you have any further notes of course.

Well, I tested all 5 pieces under exactly the same conditions (the same power supply; as well different power supplys). 2 of them are NOT working (every time they show exactly the same Vout, regardless of the trimmer-position).
For me, they are defective (maybe the trimmers can be replaced, if I have some time). The range of Vout shown on the display is flacky, even my measurements stay constant (e.g. the display switches live from 4,6 to 4,5 V, but my measurement shows constantly 4,6 or 4,64 V or similar). Believe me, they are flacky. No idea about the reason. I know, this sound like a "product tester" - but as this is my very first time using a buck-generator, I feel obliged to share my experience. If someone knows more stable models - an info would be very nice.

The big advantage of this voltage-interposer is to test a CPU witch is not stable at a desired frequency using "stock" voltages of a mainboard (3,3/3,45/3,6/4,0/5,0) - but maybe the same CPU can run the desired frequency using some voltages "in between". This was for me the way to found out the possibilty of one i486DX-WB-100 to run at 133 MHz (witch CPU was only stable at 120 MHz at 4,0 V, but was not able to boot at 5,0 V).

For example, I remember, feipoa shows many years ago somewhere here in the forum how to do a directly voltage-modification of a mainboard (a 486 from Biostar I think) for having voltages for a Cyrix 5x86 of about 3,7V (if I remember correctly) - using an external voltage-regulator for Cyrix-CPUs (I will try to build in the near future), there is no need to risk a damage of your mainboard...

Of course, I have some CPUs here running well at 4,0 or 5,0 Volt - for them, no voltage-interposer is needed.

So, it looks like one has to make the conscious choice how to approach the overclocking.
The higher labor - lower cost approach that you outlined above. Basically more manual work but potentially fewer CPUs needed, assuming the more flexible voltage control that will enable some of the CPUs to show their best.
Or the lower labor - higher cost way, where more $ are spent on more CPUs, then bin them to find the ones who survive the more rudimentary voltage controls of the motherboards.

Getting i486DX-WB-100 running at 133MHz is great. I cannot remember every going beyond 120.

Btw, linked your post with the interposer in the first page dictionary so it is more discoverable.

pshipkov wrote on 2024-07-24, 05:39:

Getting i486DX-WB-100 running at 133MHz is great. I cannot remember every going beyond 120.

One exemplar runs at 150 excellent, too, see above. This was initially not intended - it was stable at 133 MHz @ 4,2V/4,5V/4,8V. After that, I wondered, if it can do 150 (3x 50) - at 4,8V, a permanent restart during the boot of Win98SE occurs, so I had just to increase the voltage in little steps up to 5,0 V (still using the interposer) - and it works (so it is a "collateral success"). Run it for about 2 hours (ever 10-15 minutes a new test of GLQuake and Quake II), after that I tried some "fresh" restarts of Windows to test the stability - no problem for this CPU.

I think, the very big advantage of the voltage-interposer is for anybody to optimize the CPUs already present in the own box (without the absolutely need to buy more of them), using a little bit more voltage than the "usual" 486-CPU-values can maybe do up to +20 or + 30 MHz more for some CPUs, if lucky. As I already wrote, one of my Am486DX5-133 does now stable 180 MHz at 3,70 V - this CPU was not able to boot at 4,0 V (standard voltage provided by the board), so in the past only 160 MHz was possible at 3,6 V.

In addition, there is no need of any mainboard-modification regarding the voltage (it maybe works on mainboards with defective onboard-voltage-regulators, too, BUT I am not sure about this, because pin J1 is still connected to the onboard-CPU-socket).
Another advantage is, if a mainboard is not stable at higher FSB, just take another one, and try it again! An easy swapping of CPU+interposer from one board to the next is now possible! 😀

Additional info - for the voltage-interposer, I finally decided to buy a new unused socket for few euros (as feipoa means this, too), this was the better choice, I think.

Agreed with your points.
I may splurge for one of these contraptions.
Hopefully for a working one. : )

A few things I don't understand (maybe I missed a few key lines of text).

If the i486DX4 is stable at 150 MHz and 5V using the interposer hack, couldn't you just run the CPU on the motherboard directly and set it for 5 V? Or was it all those filtering capacitors that help make it stable at 150 MHz / 5 V ?

Another approach which may be more convenient for people: use a standard socket 3 interposer/VRM unit and replace the voltage set resistor with a trimmer. This will give you up to 4.3 V with most linear LDO's, but if you swap the swap the LDO for an ultra low drop linear regulator, this should get you up to 4.7-4.9 V output, depending on your PSU.

pshipkov wrote on 2024-07-29, 07:57:

Agreed with your points.
I may splurge for one of these contraptions.
Hopefully for a working one. : )

Yesterday I found out, that the trimmers have a very huge range of rotation number to adjust the Vout - the trimmers can be rotated about 25 (!) times.
At least for one of the "defective" pieces I was able to adjust the appropriate voltage, but still not sure if it works fine all the time.
Anyway, for me the trimmers are still a bit flacky (and the displays are not very correctly about the Vout, too, as shown about).
So if you (or somebody else) have some better models, please tell us.

feipoa wrote on 2024-07-29, 20:05:

If the i486DX4 is stable at 150 MHz and 5V using the interposer hack, couldn't you just run the CPU on the motherboard directly and set it for 5 V? Or was it all those filtering capacitors that help make it stable at 150 MHz / 5 V ?

This CPU is stable directly on the onboard-socket at 5 V (without the interposer), too 😀
I am very surprised about this result, too: + 50 % higher CPU-frequency 😀

Of course, in this case the interposed is not necessary (but the interposer was important to check the voltage-stability of this CPU in advance in little steps between 4,0 and 5,0 V)

feipoa wrote on 2024-07-29, 20:05:

Another approach which may be more convenient for people: use a standard socket 3 interposer/VRM unit and replace the voltage set resistor with a trimmer. This will give you up to 4.3 V with most linear LDO's, but if you swap the swap the LDO for an ultra low drop linear regulator, this should get you up to 4.7-4.9 V output, depending on your PSU.

Yes, of course.
But, if you need to buy a standard interposer (not already having one), it can cost about 30-50 (maybe more) euro. After that, it still must be prepaired for use as you describe it (sadly, sometimes with a chance to destroy it during the preparation - "shit happens"...).

On the other hand, to build the "new" interposer above (starting from Zero), you have to invest about 10-15 euro only (162-pin-socket plus buck-generator plus capacitors), and maybe 1 hour (more or less) time to build it.

pshipkov wrote on 2024-07-24, 05:39:

Btw, linked your post with the interposer in the first page dictionary so it is more discoverable.

pshipkov, maybe you should rename the topic about the voltage-interposer on the first page e.g. to

" buck generator + interposer for precise voltage control of 486-CPUs made by Intel and AMD" or similar

Otherwise, somebody could use the interposer for 486-CPUs made by Cyrix, too - with a danger to destroy them and/or the mainboard.

good point.
did.

The same system as above (HOT-433Rev4/32MB-50ns/FSB 66/i486DX-WB-133) using 512 KB L2 @3-2-3 (instead of 256 KB)

This configuration is until now NOT fully stable in Win98SE, sometimes they are errors by starting the software in Windows (eighter the board can not handle 512 KB @ FSB66, or the L2-modules itself can not handle FSB 66 MHz).

Anyway, I was able to get results in GLQuake (25,0 FPS) and Quake II (14,3 FPS).

By the way: the Vout of buck-generator measured by myself sometimes vary during the tests by about 0,05 V (but I do not think, this is the reason for the errors above). This is maybe caused by the buck-generator, or by the used PSU (= Vin)

IBM PS/2 Model 30 8530 P-Planar

Rules exist for a reason, and breaking them has consequences.
In the realm of computing, I've found it beneficial to maintain a distance from major North American brands like IBM, Compaq, DEC, Apple, Dell, HP, Gateway 2000, and so on.

IBM's Model 30 boards were once so common that they can still be acquired inexpensively today. Decided to examine one, specifically the Model 8525.
It features a non-standard power connector, which mostly aligns with the AT standard except for one pin that needs to be changed from +5V to GND.
Notice how on the brown board, this pin is bent, and it is entirely missing on the green board (it broke off).
The 8525 board never functioned correctly; I couldn't test anything on it, and it eventually stopped working altogether. I didn't attempt to repair it because I didn't know where to start. It is now with another forum member. Hope he resurrects it.

I acquired a second board, the green 8530 mentioned above. This one worked well initially. Early tests showed good structural and functional integrity, but it didn't overclock well, which is important to me.
At approximately 11MHz, the keyboard controller would hang the system during boot. Tried different KBC chips, but the assembly only accepted type 5 controllers, regardless of brand. This was unusual but manageable. I found a few different chips from AMI, Intel, and NEC, but none could overcome the 11.68MHz limit set by the 70MHz oscillator with a frequency divider of 6. The board accepted the NEC V30 CPU without issue, which can exceed its 16MHz rating, but running it at mere 11MHz frustrating.
Eventually, I realized the system clock oscillator (in the upper right corner) could be hot-swapped after booting to DOS. I used a frequency multiplier from Migron that fits a can-4 socket. This resulted in 24.535*3.33 = 13.68MHz. Better, but far from great.
The next logical step was to see if the 150ns on-board RAM chips were too slow. It was unlikely but who knows really. Replaced the four 41464 chips with 80ns versions, and the board died. Hard. Couldn't revive it. Cannot identify what and where went wrong.

I'm decently skilled with tools and electronics repair, so I know it wasn't due to carelessness. I was particularly cautious because, with two boards, I noticed how poorly constructed they are. The PCB is flimsy, bends, bubbles and melts even with modest heat gun use.

Acquired a third board, the brown one above.
Interestingly, it didn't power up with the NEC V30 CPU, which was unexpected. After probing with an oscilloscope, I determined the issue was likely related to the 72X8202 chip. Notably, these chips differ between the two boards.
Initially, the Japan-made SLA6330J was on the now-dead green board.
Swapped the chips, which caused the brown board to fail too. This was baffling. Probing with a multimeter confirmed all pins were properly soldered, but some signals were apparently incorrect. I had some flux residue around the swapped chip and between its pins. This had never been an issue before, so I initially dismissed it, but eventually, I decided to clean it off. Hot water and a brush do the job. After drying the board and plugging it in, it worked immediately. This was perplexing.

The brown board with the NEC V30 CPU was still limited to no more than 13.68MHz. I captured results from standard benchmarks at 10MHz (normalized) and peak 13.68MHz, but performance was meh. Decided to carefully upgrade the on-board RAM. After doing so, the board failed, just like the green one.
Very frustrating, but got the memo and decided to stop here.

So, let's summarize:
8525: Odd power connector that maps to AT. No MCGA connector, only on-board pins for it. DIP on-board RAM, making it easy to upgrade with faster chips.
8530: Standard AT power connector, easy to use. Proper MCGA connector, easy to use. PLCC on-board RAM, difficult to upgrade.
Poor PCB quality.
Limited to one type of keyboard controller.
Some boards don't support NEC V30 CPU upgrades.
Sensitive signal processing; even flux presence disrupts signal integrity.

I usually respect and appreciate these vintage electronics and the engineering behind them, but the IBM Model 30 motherboards are just too bad. I hesitate to call them garbage, but will refrain from doing so. Anyhow.

Tested with:
NEC V30 16MHz
Diamond SpeedStar 24 (ET4000AX)
640KB RAM (first 128KB are 150ns, the rest are two sticks with 80ns chips)
XT-IDE controller with CF, DOS 4 and 6.22 (no performance difference)

Performance is unimpressive for an 8086/V30 system.
benchmark results

---

EDIT: If somebody wants these motherboards - PM me and they will be yours for a shipping cost.

These IBM Model 30 boards sound troublesome and sluggish. I'm surprised you persisted with 3 sample motherboards. Perhaps there were some subtle differences between your 150 and 80 ns memory which caused the failure. I remember once using some memory modules on an M919 board which killed it. I later replaced the northbridge on that board and all was well.

Regarding your flux residue causing issues - was it no-clean flux? Apparently, no-clean flux should be fine to leave on the PCB. Check your solder roll to see if it says 'no-clean' anywhere on it. If not, then maybe this is why you had to remove the flux.

I'm surprised you also used the same board-killer RAM on your 3rd Model 30 motherboard. Did you also try putting the 150 ns RAM back on the MB to see if the board came back to life?

Regarding - Some boards don't support NEC V30 CPU upgrades - Did you swap BIOSes between the boards to see if the issue was BIOS related?

The first one was flaky. That was long time ago, so cannot even remember what was exactly wrong with it. Gave it to another forum member to struggle with it.
Yes, there is something off with the 150/80ns memory chips. Sounds similar to your experience with M919.

I use LaCo regular soldering flux. It is old, but want to finish it up before opening a fresh one.
So far it never gave me trouble, but going forward i am going to watch it.

It took a third dead board before it clicked with me that the RAM is killing it, or so it appears to be.
Reverted back to the original 150ns chips, but it was too late.

Swapped BIOSes between the boards. Also downloaded some microcodes, but it is not the BIOS it is one of the boards not licking the V30.
This one feels like quality control failure.

This is the final configuration of my 486-system using an Intel 486DX4-100 @ 150 MHz/5 Volt at FSB 50 MHz (this CPU is already shown above):

- Shuttle HOT-433 rev.1 with 1 MB L2-cache @ 3-2-3
- 32 MB FPM-RAM 60 ns
- 2D-VGA: Permedia 2 from Diamond (Fire GL 1k), 8 MB SGRAM
- 3D-VGA: Voodoo 2 (8 MB EDO-RAM) from Diamond, too
- Creative Soundblaster 16 CT4170

The results:

- GLQuake 800x600 in Win98SE: 23, 6 FPS (Voodoo 2)
- Quake II 800x600 in Win98SE: 9,2 FPS (Voodoo 2)
- 3DMark 99MAX: 292 3DMarks, 402 CPUMarks (Voodoo 2)
- Quake I in DOS: 17,3 FPS
- Doom in DOS: 58,67 FPS (1273 realtics)
- PCPBench /vgamode in DOS: 30,9
- 3DBench 1.0c in DOS: 87,7

Remarks:

- the RAM-stick of 32 MB/60 ns is able to do stable Read Wait State = 1 and Write Wait State = 0 at FSB 50 MHz (I had to try some different sticks to find a stable one)

- this board can NOT handle EDO-RAM correctly: as shown above, this exemplar does NOT run well using EDO-RAM in Windows (random errors, hangs and BSODs), so FPM-RAM must be used

- about the Permedia 2: it is in 3D OpenGL MUCH slower than the Voodoo 2, in addition in 3DMARK 99MAX (DirectX) the picture is really terrible - missing structures and textures! So it is used in 2D only

- maybe because of the Permedia-driver (have tried some different ones), no shutdown of Win98SE was possible (black screen). To solve the problem, the option "Video BIOS Cacheable" in BIOS must be set do "disabled"

- the Permedia 2 is not recognized well in System Speed Test 4.78 (only 512 KB of VRAM)

- maybe because of the high FSB of 50 MHz, maximum two (2) PCI-cards can be used at the same time (this problem occurs on other HOT-433 boards I have, too)

-the onboard-IDE-controller is flacky (as known for its type) - maximum two (2) IDE-devices at the same time can be used, and for FSB 50 MHz the PIO-Mode of 4 is not stable (so it must be reduced to 3)

So, with regard to all restrictions, the system works now pretty fine and stable at 100 %.

next pics

next pics

and the last pics

Looks great. Didn't think i will see Intel 486DX4-100 running stably at 150MHz.
What is the local storage ?
Did you consider using an extension IDE card like Promise Ultra-100/133, or other maybe SCSI one ?

I am not 100% about the EDO RAM issues with this board. How confident you are about not being a flaky memory module(s) but the board itself ?
Asking this, because the V4 of this motherboard can handle EDO RAM at 66MHz just fine.

Updated the link in the first page dictionary to point to this final round/post here.