MikeSG wrote on 2024-09-20, 09:29:

It's worth choosing AGP cards that had 128-bit memory vs, PCI 64-bit memory.

With a low bus speed you would run low texture details anyway, but higher resolutions would be possible, or more 3D hardware calculations per pixel.

But not on a 486 ...though it would be very amusing to see an AGP card on a 486.

MikeSG wrote on 2024-09-20, 09:29:

It's worth choosing AGP cards that had 128-bit memory vs, PCI 64-bit memory.

With a low bus speed you would run low texture details anyway, but higher resolutions would be possible, or more 3D hardware calculations per pixel.

Well yeah and OEMs differentiated them by core clock, and memory type/speed, SDR vs DDR, SGR vs SDR etc etc, but it was more like they were deliberately nerfing the PCI card.

Well a 128-bit TNT2 Ultra could run Quake 3 at a playable framerate, where a 64-bit TNT2 M64 couldn't.

So it's an interesting test for a POD-83 if it was installed in a new 486 motherboard supporting 1x AGP.

Would I spend 1-2 years building & designing a new 486 motherboard to do this... no.

spoiler: a pod ain't going to quake3 good either. Rendering isn't the only killer here, there's also a bytecode recompiler constantly working and some very unoptimized sound mixing, and that'll bottleneck any theoretical benefit an AGP would bring to Socket 3.

I'm sorry to say that there are no fans of the 486 DX-2 in this particular way in the current VOGONS community that could be interested in improving the performance of a system based on this processor. For now, I consider the topic closed.

Tymo486DX2 wrote on 2024-09-21, 12:04:

I'm sorry to say that there are no fans of the 486 DX-2 in this particular way in the current VOGONS community that could be interested in improving the performance of a system based on this processor. For now, I consider the topic closed.

That's not right.
But it's crazy to seriousely think about such dreams that are economically garbage.
And I have tested it myself, seen that it is possible but absolutely nonsense.

I'm sorry for the other posters not taking this topic too seriousely but you have to commit you were in a tunnel too.

Disruptor wrote on 2024-09-21, 13:01:

That's not right. But it's crazy to seriousely think about such dreams that are economically garbage. And I have tested it mysel […]
Show full quote

Tymo486DX2 wrote on 2024-09-21, 12:04:

I'm sorry to say that there are no fans of the 486 DX-2 in this particular way in the current VOGONS community that could be interested in improving the performance of a system based on this processor. For now, I consider the topic closed.

That's not right.
But it's crazy to seriousely think about such dreams that are economically garbage.
And I have tested it myself, seen that it is possible but absolutely nonsense.

I'm sorry for the other posters not taking this topic too seriousely but you have to commit you were in a tunnel too.

It got as much seriousness at it deserved.

Say you've got 3 people, Chucky the 4 year old, Ron Average, and Mike the Tour de France champ. Chucky is our 486, Ron is our 1st gen Pentium, and Mike is our trained athlete 6th gen coppermine Pentium 3. Now Chucky has just learned to ride a bicycle without stabilizers, on two wheels. His parents are super proud, his rich but dumb uncle is super proud. Rich Uncle buys him a top of the line racing bicycle, kevlar this, titanium that, $10,000 worth... this is our AGP slot/card combo. Can Chucky ride it? Hell no, he can maybe figure out how to sit on the crossbar and coast a bit, if he kinda scoots off on a pedal and climbs on the crossbar, but he cannot pedal it. His legs are too short. If he was particularly acrobatic, maybe he can swing down one side, give a little bit of impulse to the pedal and swing back up again. On the whole though, the thing is oversized and ungainly for him. Chucky is faster on his kiddie bike. Chucky is getting good on his kiddie bike.

Ron average has an average used bike, it gets him around a bit, he can go faster than Chucky. If he borrows Chucky's new bike, he can manage to go a little bit faster, than on his own bike but he's not an athlete. Mike is super fast on his own bike which is pretty much the same as Chucky's new bike, he wins shit on a bike like that. He's faster than Ron is on Ron's bike, he bangs his knees on the handlebars on Chucky's original bike and can't pedal that very fast, but can make it go pretty good, faster than Chucky can, as can Ron, but they are both limited by it.

If Chucky pedals top speed on his original bike, a regular 1 or 2MB SVGA card, he can keep up with Ron on his average bike, a 4MB PCI 3D card, when he's just moseying along, or Mike the Bike on his AGP 16MB machine moseying along.... say it's a rolling start, the flag drops, Mike is just gone, Chucky stays on Ron's tail with a determined burst, but he gets away from him.

Say now, what if Chucky borrows Ron's bike? Well since it might have more saddle height adjustment and such, and less complicated gears, maybe if adjusted all the way down, Chucky can sorta pedal it, but has to fall sideways to get off or something, he can't provide full impulse to the pedals, but he gets a bit of push on top of the stroke... if he practices a bit, maybe he could match Ron's "just going to the store" speed. A little faster than full speed burst on his original bike.

Anyway, does everyone praise Rich Uncle for buying Chucky such an expensive bike he can't ride? No, they say things like "What the hell are you thinking torturing the kid like that?"

Tymo486DX2 wrote on 2024-09-21, 12:04:

I'm sorry to say that there are no fans of the 486 DX-2 in this particular way in the current VOGONS community that could be interested in improving the performance of a system based on this processor. For now, I consider the topic closed.

This is a quick back of the envelope calculation and does not take into account things like protocol overhead. Please feel free to correct me if I am wrong on any point here.

a) Pentium (socket 4 and later ) and Pentium II/III (slot 1 and socket 370 derivatives) have an external 64-bit (8 bytes) data bus. This external data bus runs at 66.6MHz on Pentium CPUs (faster on super socket 7 motherboards) and early Pentium II CPUs (faster on later ones). This gives a theoretical max external CPU bandwidth of approximately 8 * 66.6MHz = 532800000 bytes/second or about 520 MB/second (marketing material would express that as 533MB/second , using base10 instead of base2 for the byte to megabyte conversions )

b) A 486 CPU (socket 1, 2 and 3) has a 32-bit (4 bytes) external data bus. A 486 DX/2 CPU running at 66.6MHz runs that external data bus at 33.3MHz (on half o the core clock). This gives a theoretical max external CPU bandwidth of approximately 4 * 33.3MHz = 133200000 bytes/second or about 127MB/second (marketing material would express that as 133MB/second , using base10 instead of base2 for the byte to megabyte conversions )

c) AGP 1.0 running at 2X has a theoretical max bandwidth of approximately 533MB/second, AGP 1.0 running at 1X has a theoretical max bandwidth of approximately 266MB/second
(base10 byte to megabyte conversion)

d) Standard 33MHZ 32-bit PCI 2.0, as implemented on existing PCI 486 motherboards, has a theoretical max bandwidth of 133MB/second (base10 byte to megabyte conversion)

So, if a 66.6MHz 486 DX/2 CPU can only talk to the world at 133MB/second, which a PCI bus can provide, what would it theoretically gain, performance-wise, from a 533MB/second AGP interface that it would be unable to fully take advantage of ?

THIS is the question which I am asking myself (and I suspect most of us are asking something similar) ? This is not even taking CPU performance differences and actual bandwidth need considerations (AGP 2x was overkill even on much newer and faster platforms versus the much slower PCI).

AFAIU, you believe that an AGP interface could somehow improve the performance of a 66.6MHz 486DX/2, but you have never shared (unless I have missed that) an explanation as to HOW or WHY that might be the case. Unless that changes, there is indeed little to be added to the topic, IMHO.

IIRC benchmarks that I thought weren't totally full of crap of 486 RAM speed were giving something like 70 or 80MB/sec on a DX2, so it's not even gonna saturate PCI.

BitWrangler wrote on 2024-09-21, 19:06:

IIRC benchmarks that I thought weren't totally full of crap of 486 RAM speed were giving something like 70 or 80MB/sec on a DX2, so it's not even gonna saturate PCI.

To be fair, the design of the memory controller in the chipset might be partially to blame, but even in the best of cases, 133MB/sec will not be reachable.

Tymo486DX2 wrote on 2024-09-21, 12:04:

I'm sorry to say that there are no fans of the 486 DX-2 in this particular way in the current VOGONS community that could be interested in improving the performance of a system based on this processor.

There are, see this post by @pc2005 going all the way to testing PCIe GPUs under OpenGL 4.5 in Linux, although that was with an AMD 5x86-133 because you're trying to maximise the performance of the platform after all. Perhaps you could contribute by performing such experiments yourself, AGP-to-PCI adapters do exist.

Only "razor blade" mainboard counts.
There are no fans interested in manufacturing tha MB. The law of the commercial market decides here.

Tymo486DX2 wrote on 2024-09-22, 12:31:

Only "razor blade" mainboard counts.
There are no fans interested in manufacturing tha MB. The law of the commercial market decides here.

You say that so casually, like just knocking up a motherboard is easy or cheap let alone making the specialised ASIC to run it all and then refab the 486 to fit modern standards.

Yup this all easy and cheap and doesn't require a few decades or more of research, designing, prototyping, redesigning, more prototyping and then possibly fabbing a working engineering model or several of them till you have a fully working motherboard all the while eating into bucket loads of cash.

Yes this is something "Fans" of the 486 DX2 66 would want to build and finance.

Personally I think the DX2 66 is perfect just as it is on the hardware designed to support it, it games perfectly fine for the DOS games made for it.

Let's joke a bit more.
I read about overclocking 486 systems to FSB 50 MHz and running the cpu at 100MHz. The best solution is to mod the system on the FSB clock up to 66 MHz, which gives turbo for the 486 x2 processor, ultimately up to 133 MHz.
If the processor could withstand it in turbo mode, everything can be done even by the Ultra ATA-66.
I'm looking for the right version of the processor.

And no in topic but very close are two other processors
Pentium OVERDRIVE 133 MHz
Pentium Mobile 133 MHz
Host Bus 66 MHz as AGP 1.0
Data Bus Width is no acceptable 64 bit, VESA standard would be off.

Also, things get done, because the person doing it is personally extremely interested, and has a skillset they can use or extend to get it done. Sooner or later, when pointing your finger around and asking, "Can you do it?", "Can you do it?" ... you will realize that there are three fingers pointed back at you.

All of the enthusiasm and none of the skillset? Gotta start somewhere...
https://learn.adafruit.com/guides/beginner

Running a DX2 at 66 bus, you need one of two DX2s, an AMD or Intel with a 16kb cache that have been down marked from DX4 production. Well not really one, more like several, because not all of them might make the 2x50 step, then out of those, small proportion achieve 2x60, then an even smaller proportion 2x66

Tiido wrote on 2024-09-17, 23:41:

Common way is a specific 486 motheboard (I think it was a Biostar but my memory is hazy) and using Cyrix or AMD 5x86 with 2x multiplier. Finding a video card that takes 66MHz PCI is probably a challenge.

I have done this successfully on a LS-486E and a pcchips m918 (ali chipset), while running a cyrix at 2x60 running win98. I used a sis6326 (some dos compatibility issues but cheap) and later a much more dos compatible cirrus logic 5464 (better but more $$). Both of those use AGP chips pasted on a pci board. I've also tried with a PCI riva TNT2 modified with a 3.3 volt regulator as many lack this and wont run on a 486 (no 3.3 until later pci spec)($$$) however I wasn't able to get rid of blue screens and other issues, user Feipoa suggests in some threads that there are compatibility issues with most of the TNT drivers and 4/5x86 cpus so that's likely the issue, however since the card had other issues (RAMDAC) I didn't pursue that configuration further. I used a promise tx2 100 pci ide controller that was tolerant of the high pci clock (onboard gave io errors at that speed)

What can I say? You will need a GOOD board that will run at undocumented 60mhz, a fast 5x86, a bunch of cache chips so you can cherry pick ones that will work at that speed. And luck, lots of it, and possibly deep pockets.

The 486e showed minimal to no gains, it seemed like the board inserted a wait state to pci negating the faster bus clock. The m918 showed SLIGHTLY better scores, especially on vga memory speeds, and disk operations definitely felt faster in windows, HOWEVER here and there I would loose bios settings (probably eprom corruption) and after a few months that board started having weird issues like serial ports not working and hardware detection problems until finally one day it didn't work at 60mhz at all, I'm pretty sure it was damaged by prolonged use at that speed.

I obtained a second 918 board, however this one is unable to run pci at 60 no matter what I do, so definitely some random luck involved.

You can also run a pcchips m919 like this, however since it implements the pci divider automatically you have to switch a fsb jumper on the fly after boot to switch to 60 with a 1/1 ratio

bertrammatrix wrote on 2024-09-22, 15:54:

What can I say? You will need a GOOD board that will run at undocumented 60mhz, a fast 5x86, a bunch of cache chips so you can cherry pick ones that will work at that speed. And luck, lots of it, and possibly deep pockets.

You can try an AMD 486 DX4 120 SV8B in x2 mode to cherry pick. 2x 60 MHz.
Much easier than finding a proper Am 5x86 133 that works at 180 MHz with 3x 60 MHz.

Yes it's probably a lot easier to find DX4s with 2x than the couple of percent DX2 culls from those and hope they go high.

I feel like the Cyrix/IBM/ST/TI are ignorable though since I don't hear much about them overclocking past 100, though I think I heard something about the TI ones being made maybe a skooch better and maybe capable of 120.