Thanks guys,
I'll read that 132/168 thread tonight. Sound like an interesting project.
Adding the resistor bodge for 487SX should be quick and easy. Otherwise I would keep it as simple as possible to actually finish the project 😀 All the unfinished projects in my drawers taught me this, haha
I was expecting someone to tell me it's silly to go the extra mile for 100Mhz 😀 It's just for the hell of it. Not even sure it's doable but sure looks like its possible.
I'm already gathering parts for the next build with w95 and Pentium for games from -95 and forward. Thanks for the advice on the setup. Sounds perfect for that project.

I think it might be pretty difficult to do the SXL2 pinout on a PGA-168 to PGA-168 VRM module with the other features you talk about. There is a big unknown as to if it actually works. We'd want to confirm this can work before trying to add an option for it on an interposer with other features. I do like the idea to set Cyrix 5x86 registers though. I don't really like having to set it up in software though. There are really tiny DIP switches that would suffice which would offer sufficient options for the most useful registers. 487SX mode is a good idea.

Anonymous Coward wrote on 2020-04-23, 15:57:

Going from a DX/2-66 to a DX/4-100 should give you a lot more than a 10% boost especially if you're using an Intel DX4.
POD83 is borderline useless as a 486 upgrade, because the 5V systems that actually need it can't run it at full speed. Not to mention, it's only really fast at floating point operations. For things that need integer performance, AMD and Cyrix are better alternatives.

Thats interesting to know, is there much difference between an AMD 486 DX4-100 and an Intel?
Back in the day i built a DX4-100 system with an AMD processor, and it was just able to decode MP3's with the occassional "blip" if you were working it too hard.
yes the 33mhz bus was an advantage with a DX2 over a DX4, but not sure how much the extra clockspeed made up for it.

It's hard to compare the DX4s, because AMD came in three flavours:

-8kb write through
-8kb writeback
-16kb writeback (can work in write through mode too)

However, back "in the day" only the 8kb versions were common.

Intel came in writeback and writethrough flavours as well, both with 16kb cache.

The AMD chips were essentially just i486DX clones hacked to add a 3x mode, but Intel made several enhancements to their DX4 core. Some say Intel DX4-100 is equal to an AMD 5x86-133 if both have cache set the same way. I haven't done any comparisons myself however.

Some say Intel DX4-100 is equal to an AMD 5x86-133 if both have cache set the same way.

Intel DX4-100 is faster than 100Mhz 5x86, but not 33% faster.

Anonymous Coward wrote on 2020-07-17, 07:48:

It's hard to compare the DX4s, because AMD came in three flavours: […]
Show full quote

It's hard to compare the DX4s, because AMD came in three flavours:

-8kb write through
-8kb writeback
-16kb writeback (can work in write through mode too)

However, back "in the day" only the 8kb versions were common.

Intel came in writeback and writethrough flavours as well, both with 16kb cache.

The AMD chips were essentially just i486DX clones hacked to add a 3x mode, but Intel made several enhancements to their DX4 core. Some say Intel DX4-100 is equal to an AMD 5x86-133 if both have cache set the same way. I haven't done any comparisons myself however.

OK, thats interesting to know.
I have an Intel dx4100 overdrive CPU (DX4ODPR100 ), from what I can tell its 16KB, but not sure if its write through or write back.

All dx4 overdrive chips are write through, with fixed 3x multiplier.

Do you know what the pinout is for the voltage regulator socket?

I recently went thru this and using a multimeter and probing the pins (and a few bits of imcomplete information from the web, and a little leap of faith) I figured out how to wire up a generic VRM from Amazon. The generic VRMs are adjustable and it was $11 for a pack of two. This is what I used:

https://smile.amazon.com/gp/product/B07CVBG8C … 0?ie=UTF8&psc=1

On the subject of VRM.
Has anyone actually seen a Overdrive/POD voltage converter blow out from wear?

Iv had a few 486 boards (mostly cheaper luckystar ones mind you) that blow their VRM first thing before anything else. But iv never heard of a Overdrive dying , despite the fact that on a DX4 Overdrive anyway, the VRM isn't directly touching the heatsink

I'd really be curious to know if anyone has a dead OD from VRM failure.

Anonymous Coward wrote on 2020-07-17, 07:48:

Some say Intel DX4-100 is equal to an AMD 5x86-133 if both have cache set the same way. I haven't done any comparisons myself however.

I can kinda back that up. Granted im no master benchmarker, and i use DOOM in phill's benchmark pack for 90% of my testing.

I have a DX4-ODPR on a luckystar LS486. With ram and cache timings tightened as low as i can get them stable, the DX4-ODPR scored 30-32FPS in Doom, a 5x86-133 with the exact same settings, ram, cards etc, i only scored 34-35FPS in Doom.

Definitely a bit disappointing IMO. I kinda had it in my head a 5x86 would blow a DX4 outa the waters a bit. Iv seen write-back Intel DX4s do 40fps in Doom.

the DX4-ODPR scored 30-32FPS in Doom, a 5x86-133 with the exact same settings, ram, cards etc, i only scored 34-35FPS in Doom.

That doesn't prove anything, because Doom is severely bottlenecked by other factors.

According to Speedsys my DX4-100 is equivalent to a 586-133. Not sure how 'pure cpu' the test is, this is with a 50 MHz FSB and write-through cache.

EDIT: My system got a CPU score of 39.45 with 33 MHz FSB (and slow AF cache/memory timings), so it's not a pure CPU test.

The Serpent Rider wrote on 2020-07-17, 16:44:

the DX4-ODPR scored 30-32FPS in Doom, a 5x86-133 with the exact same settings, ram, cards etc, i only scored 34-35FPS in Doom.

That doesn't prove anything, because Doom is severely bottlenecked by other factors.

Curious! When i get the VRM on my board fixed, i'l give the 5x86 another try with more synthetics tests.
Besides speedsys, I personally try to avoid synthetic benchmarks as i don't feel they give as accurate "real world performance" results as benching actual games.
(Which unlike benchmark specific programs, arn't always as optimized). Maybe i'm looking at it wrong 😜

chublord wrote on 2020-07-17, 17:02:

According to Speedsys my DX4-100 is equivalent to a 586-133. Not sure how 'pure cpu' the test is, this is with a 50 MHz FSB and write-through cache.

Allmost the exact same scores as my DX4-100. Excellent! 😁

You can push Doom timedemo score on 486DX beyond 60 fps with FSB and PCI bus overclocking.

A 486DX? Surely you must mean a DX4. Any links on that one?

It's easier to get a 5V & 3.3V compatible board, they are common and cheap.

Any links on that one?

Re: The World's Fastest 486
Re: The World's Fastest 486

I think you're using the term "486DX" too loosely. Is the am5x86-180 (on a 60MHz bus) what you were referring to?

Yes, as you can see, CPU speed isn't main factor for Doom at this point.

A custom converter sounds great especially one with options for the Cyrix 5x86.

When going for speed always try to maximise settings and only then amp up the bus. A intelDX4(wb preferred) is a very good cpu at 2 x 50mhz fsb(~50fps in timed Doom demo3 for a write through one) and also hits hard at 3 x 40mhz. Latter is of course more economic on most boards and may be faster overall for most software.

There's a reason intel did not offer juicier 486s as they would have made their, very early, Pentiums look pretty bad back then.

Some time ago I made a direct 1:1 adapter because I needed to raise the CPU with a heatsink a bit. I saw the other thread with lots of bodge wires, but I created it by stacking and soldering top and bottom pin rows together (beacuse the pin count is the same here), installing one row at a time, from inside to the outside.

It is passive, there is no voltage converter, but you would just need to cut appropriate pins on the bottom of adapter, route the wires to the side of the board and add voltage converter. I guess that all the pins besides VCC pins are directly connected to the bottom in the original adapters, but that would need to be confirmed.

Parts are cheap and easily available.
round pin socket https://www.ebay.com/itm/10Pcs-Single-Row-40P … qEAAOSw2gxYtXUh (1 auction = 400 pins)
round pin header https://www.ebay.com/itm/10-Pcs-Gold-Plated-2 … x8AAOSw3qJZzxnq (1 auction = 400 pins)
protoboard https://www.ebay.com/itm/5Pcs-6x8cm-Double-si … FwAAOSwwgVemv6i (1 auction = 5 pieces)

BOM is 15 USD and you can make at least two adapters with that. Voltage converter module should cost not more than 5 USD. It may cost more if you want to put there something beefier, LCD/voltage monitoring etc.

If original adapters are double-sided PCBs (not multilayer), it would be easiest to desolder all components from one with rework station, scan the PCB and order some clones... or measure everything with multimeter, non-destructive way. Manufacturing 5 pieces of dedicated PCB would cost 8 USD shipped at JLCPCB.