Chadti99 wrote on 2025-09-05, 12:03:

Completely missed this thread, congrats on the Quake score, wasn’t sure we’d see 23+ fps on a 486, nice work!

Do we have the bios and jumper settings documented for the 23.4 fps Quake score?

Fastest dram and cache settings and I’m assuming 1:1 pci at 66x3? Screenshots of bios settings would be awesome.

Thanks.

Yes, I can get back to you with bios settings, I'm still using this thing almost daily with no changes to the setup.

All of this at a 1/2 pci (30 or 33mhz). This is primarily driven by me being lazy to make the board switch to 60 on the fly and also partly by the decision to use this particular graphics card in the build. Since I'm going for the best playability of early directx (outlaws, jedi knight etc - which when accelerated will run fine at 640x480 on this) this card makes sense to me as it offers the best acceleration on this platform - HOWEVER I don't think it worked at 60mhz, and any of the cards that did work at that frequency may have scored better at SOME benchmarks (doom or 0.5 fps better in quake) but overall were much slower as actual accelerators.

analog_programmer wrote on 2025-09-05, 13:50:

But, but... PCChips M919... "PCChips = cr*p" 🤣 Actually pretty nice system and impressive results 😀

Is the passive cooler simply placed on top of the CPU or there is some kind of retention?

Imho the problem with pc chips is the chipsets used are often great but bios was not (older revisions often being better then newer ones, strange inconsistencies with caching, memory speed timings etc) AND pcb quality was generally low.... leading to unreliability that's often hard to pinpoint. I have 2 identical boards and I couldn't say which one is "better" ...the one that happens to work with all the parts I throw at it 🤣

The cooler just sat on the cpu during bench setup, now it is permanently affixed with 4 dabs of epoxy on the corners and thermal paste in the center. That's my "standard procedure". It's not heavy enough to worry me, especially since there is no fan vibrating on it.

Chadti99 wrote on 2025-09-05, 14:15:

🤣, the m919 board I have was a total pain to work with, kept losing its bios settings on top of other issues. Tempted to pull it back out or find another example. And I agree with OP on forcing the 1:1 after each boot being a pain.

I’m also worried to move my 200Mhz capable 5x86 around too much in fears of killing it from ESD or otherwise.

Interestingly, none of my 919's has ever done that (2x v3.4b/f and one older v1.5 revision board) sort of failure BUT I had a m918 board give me similar bios issues AFTER EXTENDED USE AT 60MHZ PCI. This eventually made the board unusable, though initially slowing the board back down made the bios settings retain again. I tried swapping a bios chip from an identical board but it didn't fix it. I wonder if it could be some damage to whatever onboard memory holds bios settings (is it a little eprom or something?) due to some out of whack bus speed (or just age??)

I totally understand your worry. On one side, it's a cool thought to swap a bunch of hardware for a few more comparative benchmarks, but on the other side sometimes I've taken old hardware out only for it to not work again when it went back in 🤣🤣🤣 it's like a lottery every time

bertrammatrix wrote on 2025-09-05, 18:18:

sometimes I've taken old hardware out only for it to not work again when it went back in 🤣🤣🤣 it's like a lottery every time

I dread taking my completed systems out of the closet for this reason. It will either work just fine, or I’ll spend half the night trying to diagnose the issue. We certainly aren’t collecting baseball cards here.

Chadti99 wrote on 2025-09-05, 12:03:

Completely missed this thread, congrats on the Quake score, wasn’t sure we’d see 23+ fps on a 486, nice work!

Do we have the bios and jumper settings documented for the 23.4 fps Quake score?

Fastest dram and cache settings and I’m assuming 1:1 pci at 66x3? Screenshots of bios settings would be awesome.

Here, nothing special...fastest for cache and most memory and slowest for anything ISA related...to give a chery picked sound card the best chance of working properly. Note also turning off any unnecessary i/o and ide (I use a promise pci controller). Period reviews point especially to the secondary IDE as being problematic/causing errata during use even at stock speeds.

analog_programmer wrote on 2025-09-05, 13:50:

But, but... PCChips M919...

When compared to simplicity of LuckyStar LS-486E C2 overclocking, it's still kinda meh. Nowadays you could say that LuckyStar is for "casuals" though.

The Serpent Rider wrote on 2025-09-06, 21:48:

When compared to simplicity of LuckyStar LS-486E C2 overclocking, it's still kinda meh. Nowadays you could say that LuckyStar is for "casuals" though.

I think M919 board is more easily obtainable. But for the "fake cache" revisions (all boards different than rev. 1.xx) it needs a proprietary PCChips COASt module.

I'm not into PCChips hater's club 😀

Thanks for the screenshots! Can anyone remind me what the jp3 jumper block positions are for 60Mhz and 66Mhz bus speeds? Thanks!

I got it!

Closed, open, closed = 66
Open, closed, open = 60

Trying to find stability at 66x3. Does your CHKCPU look like this? Specifically does it mention L1 in Write Through? Noticed my bios also has L1 listed as write through and grayed out. Any chance you could share your jumper settings?

Okay, each boot gives me different results between L1 being enabled or disabled , L2 being WB or WT. Looking at the 1MB pancake cache module I had forgotten it mentions it’s for 3.3v CPU’s only. To get my 5x86 to run at 200MHz it needs the juice at 5v. So it could be a combo of a flaky board and overvolting this cache. The fastest I saw was 22.6 in Quake but that was with 313 cache and 3ws read and write I think and L1 reporting WB mode.

Chadti99 wrote on 2025-09-08, 14:25:

Okay, each boot gives me different results between L1 being enabled or disabled , L2 being WB or WT. Looking at the 1MB pancake cache module I had forgotten it mentions it’s for 3.3v CPU’s only. To get my 5x86 to run at 200MHz it needs the juice at 5v. So it could be a combo of a flaky board and overvolting this cache. The fastest I saw was 22.6 in Quake but that was with 313 cache and 3ws read and write I think and L1 reporting WB mode.

Right, so the cache module (1mb panacakepuppy) ALWAYS gets the cpus regulated voltage (whether the board jumper sets it to 3.3 or 4 volt or in between ) REGARDLESS if the cpu is set to 5volts with the 4other jumpers. So, if you are using 5 volts feel free to leave the jumper in the 3.3 position and that is all the cache will see

If you are not able to run fastest cache timing try different video cards, some affect this more then others, and mainly RAM. All of us that run these speeds seem to have been through quite a bit of it to find the perfect stick

I recommend video card that is able to run the quake benchmark from dosbench in 640x480. Now, obviously this is a lot slower HOWEVER I found this to be a fairly good indicator of stability. If it crashes there windows would be unreliable.

Chadti99 wrote on 2025-09-08, 14:25:

Okay, each boot gives me different results between L1 being enabled or disabled , L2 being WB or WT. Looking at the 1MB pancake cache module I had forgotten it mentions it’s for 3.3v CPU’s only. To get my 5x86 to run at 200MHz it needs the juice at 5v. So it could be a combo of a flaky board and overvolting this cache. The fastest I saw was 22.6 in Quake but that was with 313 cache and 3ws read and write I think and L1 reporting WB mode.

You should try to vary voltage. I ran in to some CPUs that would misbehave after a certain point if the voltage kept going up past a certain point. You can use the 5 volt jumper block to " insert " whatever voltage you want from a regulator board, or, I have used a pair of 4amp diodes across the terminals the iumpers connect for 5 volts to lower the voltage by the diodes 0.7 volts or whatnot. Throw a big schottky in there (from a power supply) and its 0.3. Note, this isn't "the proper way", but it does work

Also not sure what you have for ram, but at this speed don't try and use more then a single stick.

Thanks for all the tips bertrammatrix!

I ended up sourcing another board and this seems to be quite a bit less flaky, although there are still some inconsistencies between boots. I think my l2 gets disabled from time to time requiring me to power cycle. At least my L1 is remaining in write back mode though.

I’ve been able to match your bios settings with a stick of Samsung EDO 50ns memory. 200MHz@66*3. It’s stable enough to run all of dosbench on this new board.

Can you elaborate more on this?
“ So, if you are using 5 volts feel free to leave the jumper in the 3.3 position and that is all the cache will see”

Are you saying there is a way to run the CPU at 5volt and the cache at 3.3volt?

Chadti99 wrote on 2025-09-16, 20:50:

Can you elaborate more on this?
“ So, if you are using 5 volts feel free to leave the jumper in the 3.3 position and that is all the cache will see”

Are you saying there is a way to run the CPU at 5volt and the cache at 3.3volt?

Yes, you are running it that way already. Again- panacakepuppy 1mb cache modules draw power from the motherboards own cpu voltage regulator. It produces 3.3 volts, or 4, or with my voltage mod resistor anything in between, depending on what the jumper is set to. Your CPU draws voltage from that regulator only if it is NOT set to 5 volts. Once you switch the 5 volt jumper bank over, the cpu ONLY gets power directly from boards 5 volt rail, but, the regulator is still supplying whatever voltage to the cache. Grab a voltmeter and poke at it, you will understand

Chadti99 wrote on 2025-09-16, 20:50:

Can you elaborate more on this?

I figured I'd take a few pictures that will explain it better then I can.

Voltage selection jumpers on m919:

Quick and dirty diode step down on m919. This is some random pc power supply double diode (wired so both internal are parallel), iirc the resultant voltage was somewhere around 4.6 volts with this

Fancy Amazon adjustable voltage regulator hack on m919. I think this thing can handle 6 amps and produces very little heat. 10 of them were 12$. Note I am not using all the pins (they are all parallel), also note regulator stealing ground from pc speaker pin. Obviously I'd tidy this up if it was a permanent solution.

bertrammatrix wrote on 2025-09-17, 02:14:

Chadti99 wrote on 2025-09-16, 20:50:

Can you elaborate more on this?
“ So, if you are using 5 volts feel free to leave the jumper in the 3.3 position and that is all the cache will see”

Are you saying there is a way to run the CPU at 5volt and the cache at 3.3volt?

Yes, you are running it that way already. Again- panacakepuppy 1mb cache modules draw power from the motherboards own cpu voltage regulator. It produces 3.3 volts, or 4, or with my voltage mod resistor anything in between, depending on what the jumper is set to. Your CPU draws voltage from that regulator only if it is NOT set to 5 volts. Once you switch the 5 volt jumper bank over, the cpu ONLY gets power directly from boards 5 volt rail, but, the regulator is still supplying whatever voltage to the cache. Grab a voltmeter and poke at it, you will understand

confirmed. happens quite often actually.

Hot dog, this is an interesting mod. So the "3.3 V M919 cache modules" always draw power from the motherboard's VRM, regardless of the jumpers? And the goal is to down regulate the CPU from the motherboard's 5 V rail using an external VRM (or CPU interposre)? This happens quite often? It's the first I've heard of this. I've been running my CPU and L1 from the MB's VRM at 4 V, but I should be running the cache at 3.3 V and the CPU at 4 V with different regulators.

Isn't the switching reg from your photo a 3 A unit? That's what I wrote on my unit. Was I mistaken; is it really 6 A?

EDIT: Does anyone know the current output limit of the VRM on the M919, assuming a dual transistor model (some MB's only have one)?

feipoa wrote on 2025-10-01, 13:37:

Hot dog, this is an interesting mod. So the "3.3 V M919 cache modules" always draw power from the motherboard's VRM, regardless of the jumpers? And the goal is to down regulate the CPU from the motherboard's 5 V rail using an external VRM (or CPU interposre)? This happens quite often? It's the first I've heard of this. I've been running my CPU and L1 from the MB's VRM at 4 V, but I should be running the cache at 3.3 V and the CPU at 4 V with different regulators.

I ran some measurements on 3.3 V 256K, 5.0 V 256K, and 3.3 V 1024K modules.

The 3.3V 256K module takes DATA voltage directly from the M919's VRM, regardless of whether or not the CPU is jumpered for 3.3V, 4V, or 5V. The module takes TAG voltage from the PSU 5V rail.

The 5.0 V 256K module takes DATA and TAG voltage directly from the PSU's 5 V rail, regardless of whether or not the CPU is jumpered for 3.3V, 4V or 5V.

The 3.3 V 1024K module takes DATA and TAG voltage directly from the M919's VRM, regardless of whether or not the CPU is jumpered for 3.3V , 4V, or 5V.

feipoa wrote on 2025-10-01, 13:37:

Isn't the switching reg from your photo a 3 A unit? That's what I wrote on my unit. Was I mistaken; is it really 6 A?

Seems that many of these modules look similiar, with some being 3 A and some being 5 A.

feipoa wrote on 2025-10-01, 13:37:

Does anyone know the current output limit of the VRM on the M919, assuming a dual transistor model (some MB's only have one)?

The regulators on the M919 have markings C3420. If it is the same as the 2SC3420, then 5 A is the absolute maximum current per transistor. https://www.datasheetcafe.com/c3420-pdf-18669/

Has anyone tried 8 ns SRAM on the 256K M919 module to see if we can achieve 2-1-2, 0ws/0ws at 180 MHz with more than just a special 16 MB EDO stick? Ideally with a 64 MB EDO stick. For those not aware, there's no 64 MB cache hole when using the 5/6/1996 BIOS.

bertrammatrix's switching regulator mod inspired me to make a similar unit, primarily so that I can run the L2 cache at 3.3 V and the CPU at 4 V. Like bertrammatrix's unit, the mod is entirely connectorised. I used wire with 24 AWG.

When running your Am5x86 at 180/200 MHz with the switching mod connected to the 5V input, how high were you able to raise the voltage to the CPU while at load? I could only take the switching VRM up to 3.72 V, implying there's about a 1.2 V dropout. I didn't see any mention of a drop out voltage in the datasheet. Maybe there's another analogous term for used for switching regulators instead of dropout.

Because I could only get up to 3.72 V with Vin=5V, I resorted to connecting the switching reg to the 12 V rail. I ran it 10 minutes at 3.85 V in Win95 without any trouble, however the switching reg was getting warm, or 63 C. I'm not sure how hot it will get after 2 hours. I guess I need to add a heatsink or fan to the VRM module. It might be easier (and with less heat generation) to use an ultra low drop-out linear regulator with Vin=5V and ditch the switching.

Maybe there's other packaged switching regulator modules which have less drop-out than my LM2596S. I'm curious to know if the switching regulator IC on bertrammatrix's buck module can reach at least 4 V when tied into 5V...?