Real reason for PCI is cost. VLB video cards is getting harder to find and expensive.

Cheers,

PCI > VLB anyday.
My memory seems to recall early PCI performance might be a *little* slower in benchmarks, but the convenience alone out ways the losses.
You can pickup a cheap fast PCI Trio64 for dollars these days, where as with VLB you'd be waiting and paying an arm and a leg for a good card like a TSENG Labs.

VLB has its charms, but it also has its quirks, especally if you plan on overclocking, or using multiple VLB cards at once.
Something PCI tends to handle alot better.

pentiumspeed wrote on 2020-03-31, 18:34:

Real reason for PCI is cost. VLB video cards is getting harder to find and expensive.

Cheers,

but even then 486 PCI boards are way more expensive than VLB ones :p

Exactly, this is why I went with Pentium using a TX chipset ATX style motherboard. Much more flexible according to others using setMul utility that now I know which one to use that has broad support of PCI cards I can choose from. Another choice is HX board but it has less support on PCI video cards.

Cheers,

SodaSuccubus wrote on 2020-03-31, 23:07:

VLB has its charms, but it also has its quirks, especally if you plan on overclocking, or using multiple VLB cards at once.
Something PCI tends to handle alot better.

One of those "quirks "is the fun to be had in getting the damn things properly inserted and having them stay that way . I don't know what the VLB connector spacing tolerances were supposed to be, but they seemed to be off by a wide enough margin to be a problem in practically every VLB build I every worked on back in the day .

PCI was a godsend .

pentiumspeed wrote on 2020-03-31, 18:34:

Real reason for PCI is cost. VLB video cards is getting harder to find and expensive.

Cheers,

Also the s3 cards are the most compatible and cheap pci cards for DOS.

darry wrote on 2020-04-01, 01:27:

PCI was a godsend .

VLB can be tough. When I'm trying out old cards, I've noticed that my Trident 9440 & Trident 9400cxi like the farthest slot, the CL-GD5429 seems to like the middle slot. If I throw in a VLB IDE card, then it's trial and error until I find a pattern that they can coexist in.

Back in the day, PCI was a lot easier than VLB, but from the retro computing stand point, I was surprised that things get picky when you mix generations. I have two 486 PCI motherboards; Via82C505 & 420EX chipsets.

These cards work in the 486 mothatboards: S3 Trio64V+ (1995) , Tseng 4000w32p (1995) , & Matrox Mystique (1996)

These cards work in my 440BX system, but don't produce a signal from the VGA port on the 486's: Radeon 9250 (2004), MX4000 (2004), Radeon x1300 (2006)

The 486's will boot, but the monitors don't see any signal. Not sure how to dig any deeper to figure out what is wrong, but my guess is that the cards require PCI 2.1 compliance.

I also have a Permedia 2 (1998) card that used to be my go-to troubleshooting card, but after 15 years in a box, it isn't producing a syncable vga output on anything these days. I can sort of make out a picture, but it is way distorted, striped vertically like someone took the monitor and twisted the right side of the screen down a dozen times. Maybe the crystal is shot ?

Some PCI cards might be better for 50/60 MHz FSB speeds in boards without PCI dividers, but even 50 MHz doesn't play nice with some mid 90s era cards.

One of my Diamond Stealth 2000s (Virge 325, 2MB) freaks out (palette errors and crashes) in 256 color mode or at least VGA 13H at 50 MHz FSB on an Acorp SiS 496 board I've tried. ISA cards were fine though, so not a CPU/cache/RAM issue at 50 MHz.

I have a second Stealth 2000 I haven't tried and I'm pretty sure most or all AGP-capable GPUs will also handle 66 MHz PCI mode (so Rage Pro, Riva 128, Voodoo Banshee, Virge GX, among others ... probably Rage IIC as well, maybe some II+ models).

The S3 Trio and Virge cores tend to benchmark a lot better than ATI's Mach 64, but that might not matter much for actual game performance, especially with 50-66 MHz FSB. (and even without max cache timing at that bus speed, the DRAM timings might be tighter: the actual read/write intervals mind you, as DRAM wait state settings will have to be greater, but depending on the chipset, the actual DRAM cycle timings might be closer to maxed out at higher clock rates: best RAS/CAS/etc/etc cycle timing; though if the chipset DRAM controller runs at the same speed and timing regardless of CPU FSB setting, the gains will be less to none: just a matter of alligning wait states and cycle times ideally then; but for chipsets that run at the FSB clock rate it'll make a difference for that and potentially for DMA/FIFO speeds too ... and hurt stability on chipsets not able to cope at the higher clock rate)

OTOH, with the right wait state settings, some VLB cards are OK up to 50 MHz, but would I doubt any would work reasonably (or at all) at 60 or 66. (but also be pleasantly surprised ... and I suppose some might just be DRAM speed limited, so swapping in faster RAM could allow that)

One of my Diamond Stealth 2000s (Virge 325, 2MB) freaks out (palette errors and crashes) in 256 color mode or at least VGA 13H at 50 MHz FSB

That's typical for all S3 Virge and Trio cards before GX2 (66 Mhz AGP capable), but only in VGA 13H. Matrox cards will work fine. Yes, even original Millenium and Mystique.

a lot better than ATI's Mach 64

Mach 64 family have limited VGA and subpar VESA performance, but decent GUI speed.

but would I doubt any would work reasonably (or at all) at 60 or 66.

VLB boards don't have 60-66 option.

These cards work in my 440BX system, but don't produce a signal from the VGA port on the 486's: Radeon 9250 (2004), MX4000 (2004), Radeon x1300 (2006)

You need 5V PCI compliant card, which was dropped at this point in time. Although some professional stuff like Quadro 4 NVS retained such compatibility.

the difference is one sounds worst than the other but both sound bad.

it really differs from person to person, for me its extremely important because 3d cards are all in pci form, so i don't collect any non-pci 486 boards.

for me its extremely important because 3d cards are all in pci form, so i don't collect any non-pci 486 boards.

You don't want to play 3D accelerated games on any 486.

Baoran wrote on 2020-03-19, 10:03:

I seem to get better performance on older vlb/isa motherboards compared to pci/isa motherboards with same cpu and pci/vlb/isa motherboards seem to be even slower. That does of course not mean that it is because difference in video cards because I don't think neither card is fully utilized in dos.

This is just based on the limited amount of different motherboards I have tried.

In my experience it's the exact opposite. I do have some decent VLB cards (S3 Trio 64, Cirrus CL 5440, ATi mach32) and motherboards, but (some) PCI cards always yield a bit more FPS in Quake and PCP bench. Fastest PCI cards I've tested on a 486 (late model SiS and UMC chipset motherboards) are the Matrox Millenium II 4MB, Cirrus Logic CL 5446 and S3 Virge PCI.

From benchmarks I have seen posted PCI motherboards have a about a 15% improvement over VLB using the same processor.
EDO ram and faster motherboard bus and better memory management , faster video with PCI motherboards.

and S3 Virge PCI

S3 Trio 64V2/DX should be slightly faster.

MKT_Gundam wrote on 2020-04-01, 01:35:

pentiumspeed wrote on 2020-03-31, 18:34:

Real reason for PCI is cost. VLB video cards is getting harder to find and expensive.

Cheers,

Also the s3 cards are the most compatible and cheap pci cards for DOS.

Without going to endless debates about stuff I agree ! S3 based VLB or PCI on a 486 are the most compatible in DOS in my experience too.

for dos performance, i surely have faith in my riva128 and gf2mx.

I sort of prefer the VLB for their specific 486 oriented idea but as said the costs are not worth the point of having a VLB while the board has the PCI even if I suppose the differences might be minimal. The real differences should be the better expandibility early PCI cards would help on Windows with better internal accelerations, drivers, etc.. without instead obviously using more modern cards that had heavy drivers, maybe incompatibilities with the chipset/cpu I suppose like I don't know a late Voodoo3 or something like that sound useless.
But a card like a Matrox Millennium PCI with WRAM on a late 486 would be something I hope to see build sometime to see how well that card should help the cpu working in a "heavy" o.s. like Win95.

386SX wrote on 2020-07-30, 09:38:

(...)
But a card like a Matrox Millennium PCI with WRAM on a late 486 would be something I hope to see build sometime to see how well that card should help the cpu working in a "heavy" o.s. like Win95.

My PCI 486 has one. It is working like a champion in dos and Windows 3.11 now.
The Intel Ninja board is mainly rock stable but not quite a performance beast... Just avarage. Running an Intel dx4 100MHz (33x3) and 64MB ram should make it suitable for running Windows 95 or 98.