ISTR Virge was a bottom tier device (inexpensive), while the Matrox was considerably up in price range.

My memory of it was that the S3 was the only real device worth having at one early stage but was outpaced quickly enough if you had the money to buy better. I didn't. It certainly was my only viable option, and there was a time starting off my business when I couldn't afford one.

mrmazda, that is how I remember it as well. I got this Virge for $20-30 at the time (1997-ish), while at the same time other options by ATI, Diamond, STB, et al were considerably higher, and Matrox and Voodoo were shockingly higher. I still remember my excitement when I could finally afford a 4MB Voodoo1 pass-through card for $100 - the same Voodoo I am still using in this pc.

business_kid, S3 cards are great budget cards with high compatibility with 2D software. At the Vogons website dedicated to retro computing and hardware, they frequently recommend S3 cards for DOS gaming; but the problem is that many manufacturers used the same chips and made minor modifications. The result is inconsistent quality across cards based on a particular chip. My Virge/GX card, while already a "bottom tier" card despite having 4MB SGRAM, was probably also affected by the cheap manufacturer's flaws in quality.

By the way, I solved my problem with freezing in 16bpp modes. It seems the Matrox card is more particular about PCI slot order. My clue was in OS/2, when booting with the Matrox driver it would hang when loading the nic driver. If I disabled the nic driver or reset to VGA it would boot fine. I swapped it with the nic so the Matrox is in the highest slot and now all is well

I remember those quirks. Some motherboard BIOS would allow reversing device enumeration order so that physical swapping might not be necessary to solve such sensitivity. Are you using an old OEM driver, SciTech's, or its progeny Snap, for your Matrox on OS/2?

I don't recall seeing such an option in my mobo's Award 4.51 bios 1.3S. In OS/2 I used both the GRADD/SciTech driver and the oem driver from the cd. After making it work it is currently using the oem driver. I was able to replicate this quirk in MS-DOS when enabling the nic driver in AUTOEXEC.BAT. I didn't check in Slackware because I hadn't setup networking yet. The nic is a D-Link DFE530TX+ (RTL8139)

That option might not have appeared anywhere until after Y2K, ancient stuff, taxing my ancient memory.

All my OS/2 boxes have Intel E100 NICs (SB8255[7,8] chips) using E100B.OS2 and E100BEO2.NIF, so that I don't have to fumble through bewildering MTPN setup when cloning. Plain text config files are one of the pleasantries of having moved from OS/2 to Linux as primary OS.

All I remember of those days is that in order to force an IRQ assignment in windows 95 I had to physically PULL OUT hardware. Only to put it back in after the cards that needed certain IRQs got them. And yeah, which slots things went into mattered. Seems that's still an issue these days, although getting a PC with slots OTS is a bit harder now.

I think you could control it somewhat with slot allocation, as that affected the order things were looked at. Even getting a PCI slot these days might be difficult. In those days I was using a Realtek 8139 for coax to a 10MB nic. It wanted IRQ 11, or it wouldn't go. W95 was hopeful on pnp, but that one sort of worked. The issue I had was linux, where I lost irq 11 to the video card occasionally. I think I sorted that in /etc/interrupts, then some other way (/etc/modules.conf?). I liked that - loads of options you could set. So of course they took it out and replaced it with something that didn't work - the early stages of the present arrangement.

Thinking about Win95 and PnP resources still makes me cringe. I think that was when I first began losing my hair Indeed, plain text config files is one of the specific reasons I moved to Linux on my main box.