For gaming? No. VESA local bus had at least one: http://old.vgamuseum.info/benchmarks/1208-cre … laster-vlb.html

Silicon Graphics Iris vision was the first ever dedicated 3D accelrator for x86 (~1991). That was ISA. And then theres stuff like the IBM PGC chipset (87-93 ish), which while not a dedicated 3D accelrator, does essentially have a coprocessor which can do vector and matrix calculations. There was a matrox version of this card called SM-640 which does hardware supported Gouraud shading (so must be 3D), which is ISA.

These are all CAD cards though, no Direct3D or OpenGL (Iris Vision does IrisGL, the precursor to OpenGL) so no games and the amont of programs that support include AutoCAD and could probably be counted on your hand.

The first 'gaming' cards were all PCI or VLB. Bandwidth for the large amounts of vertices to be uploaded each frame (eveything is immediate mode rendering) is a bit of a limitation for ISA.

Technically there were.
That is, we know of some very early prototypes of 3D accelerators which were ISA.
Also, depending on what you mean by '3D accelerator', there were various ISA Windows accelerators with very advanced acceleration features, which could render lines and polygons in hardware, suitable for 3d rendering (although in those days they were more aimed at CAD).
IBM also made 3 highly advanced videocards:
1) The Professional Graphics Controller, which had its own 8088 CPU onboard.
2) The 8514/A
3) XGA

Although 2) and 3) were MCA cards, there were at least some ISA clones of the 8514/A (such as ATi Mach32 and Tseng ET4000).
I don't think there was ever an ISA version of XGA though.

There also was Texas Instruments Graphics Architecture (TIGA), based around the TMS34010 and TMS34020 Graphics System Processors (GSP).
These cards weren't fast (in comparison to SVGA cards), but very powerful. Beeing fully programmable,
they could run programs on their own, render text fonts and perform complex calculations.
Windows 3.1 was the last OS who had a TIGA interface (still required vendor-specifc TIGA
drivers/runtime to be loaded).

The graphics processors themselves were used in several arcade games.
http://en.wikipedia.org/wiki/TMS34010#Video_c … nd_accelerators

At some point, there even was an 3D accellerator card for the IBM PC :

In 1987 TI provided the first demonstration of true real-time 3D games
with stereo sound effects on a personal computer (PC), using a small
TMS34010 adapter card (called "The Flippy").
The Flippy was designed as the basis of a game development system for
consoles and as a PC gaming card in its own right.

Jo22 wrote:

There also was Texas Instruments Graphics Architecture (TIGA), based around the TMS34010 and TMS34020 Graphics System Processors (GSP).

Yes used in the Matrox PG-1281 (TMS34010) And there was the TMS320 which predates TIGA, used in a few others such as SM-1024 (bigger brother of the SM-640 which is PGC compatible). All ISA-16bit cards.

Iris Vision (was available in MCA variant to) used Jim Clark's GeometryEngine (of course) which kinda paved the way for 3dfx and that whole Voodoo thang. The original (3) 3dfx guys all had experience with the Iris vision and its technology essentially inspired the Voodoo graphics (albeit for a different market to what SGI were placed).

spiroyster wrote:

Iris Vision (was available in MCA variant to) used Jim Clark's GeometryEngine (of course) which kinda paved the way for 3dfx and that whole Voodoo thang.

Isn't the GeometryEngine the exact opposite of a VooDoo?
GeometryEngine is more of a T&L unit, where VooDoo was just a hardware-accelerated inner-loop of a triangle filler.
It wasn't until the original GeForce that geometry was actually being processed anywhere other than the main CPU.

Scali wrote:

Isn't the GeometryEngine the exact opposite of a VooDoo? GeometryEngine is more of a T&L unit, where VooDoo was just a hardware- […]
Show full quote

spiroyster wrote:

Iris Vision (was available in MCA variant to) used Jim Clark's GeometryEngine (of course) which kinda paved the way for 3dfx and that whole Voodoo thang.

Isn't the GeometryEngine the exact opposite of a VooDoo?
GeometryEngine is more of a T&L unit, where VooDoo was just a hardware-accelerated inner-loop of a triangle filler.
It wasn't until the original GeForce that geometry was actually being processed anywhere other than the main CPU.

Yes, from that perspective ([EDIT: actually yeah I see what your saying). Both yielded TnL in this fashion, with the Voodoo relegating the TnL part to the CPU. Yes GE did what CPU did for Voodoo (TnL), RM did the 'rastering' (what Voodoo chips did), and DM did the 'displaying' (dunno what part of the Voodoo did that tbh o.0). That was my understanding anyhows.

wrt to TIGA, I don't think this had 'transformation' stacks, concept of texturing and certainly no concept of lighting (just raster and display?). I have the aformentioned Matrox cards and have been
dying to get something going with them, but have zero experience with TIGA so not sure as to 'how' programmable they are, or essentially how much I have to make the CPU do in terms transformation stacks etc. (documnentation on particular card models is scarce, I tried Matrox (UK) and even they are apparently clueless) o.0

As others mentioned there are 2 types of 3D cards....
CAD only without texture acceleration - these were "common" even in 80's. Many companies did them, but only few succeeded. Most famous are TIGA cards...
http://vgamuseum.info/index.php/cards/itemlis … chitecture-tiga

But youre talking about Voodoo 1, so i guess its about games, then i know only one card, but not sure that it powered 3D games....by looking at some pictures it could....
Its basically 3DO gaming console HW emulator for PC from Creative named 3DO Blaster...

More photos...
https://imgur.com/gallery/9qter

Console info - https://en.wikipedia.org/wiki/3DO_Interactive_Multiplayer

Ah yes, the 3DO Blaster... it's a bit of a special case though, isn't it?
As far as I know, it is a fully functional 3DO console, including the CPU, so it doesn't actually accelerate graphics for regular DOS/Windows applications, but only for software running on the board itself.

Yes the 3DO Blaster is basically just a 3DO console for use in a PC.
Also the ISA bus was some kind of a bottleneck if you take its max data transfer speed of roughly 15 Mb/sec (at 8 MHz) into account. Rendering a 3D scene with textures and framebuffering could easily exceed the ISA bus' limits. Not to mention that real 3D in SVGA usually needs 4 Mb or more. Manufacturers most likely asked themselves why the hell they should equip 4 or 8 Mb precious video memory on cards for a dead interface that couldn't even properly handle it by failing to provide a sufficient data transfer speed. Just made no sense by the time 3D got popular.

blurks wrote:

Also, the ISA bus was some kind of a bottleneck, if you take its max data transfer speed of roughly 15 Mb/sec (at 8 MHz) into account. Rendering a 3D scene with textures and framebuffering could easily exceed the ISA bus' limits.

It's still an accelerator though. You have to compare it against rendering every pixel over the bus with the CPU.
The accelerator relieves the CPU of having to fill every pixel, which effectively makes the bus speed independent of the actual resolution.
The CPU only supplies the geometry information, so it is only dependent on polycount. Given the low polycount of games in the early 90s, this geometry information would likely be smaller than an entire framebuffer.
Let's say you want 30 fps, and each triangle takes 256 bytes of data. At 15 MB/s you could have 15000/30 = 500 KB per frame. That would give you 2000 triangles on screen.

The accelerator will also have texture memory on-board, so textures only have to be uploaded once at initialization, so the actual bus speed is not relevant during texturing.
Likewise, the accelerator has the framebuffer on board, so that is also independent of the bus.

If anything, for an ISA bus it would make MORE sense to use an accelerator, not less.
Even better if that accelerator would also have hardware T&L, so you could also upload the geometry at initialization, and you'd only have to pass new matrices to the GPU every frame, giving you even more bus savings.

Scali wrote:

The accelerator will also have texture memory on-board, so textures only have to be uploaded once at initialization, so the actual bus speed is not relevant during texturing.
Likewise, the accelerator has the framebuffer on board, so that is also independent of the bus.

Depending on the complexity of a scene/game, what you call 'initialization on startup' could easily result in reloading textures regularly. Not the best idea, if you keep in mind, that ISA shares its data transfer speed with all attached devices, such as sound cards, ISA hard disk controllers and the likes.

I understand, that a 3D accelerator for the ISA bus could make sense under very certain circumstances but in 1996, ISA was already dead when the mainstream 3D API's began to skyrocket.

blurks wrote:

Depending on the complexity of a scene/game, what you call 'initialization on startup' could easily result in reloading textures regularly. Not the best idea, if you keep in mind, that ISA shares its data transfer speed with all attached devices, such as sound cards, ISA hard disk controllers and the likes.

Since that is a given, no software dev would be stupid enough to design a game that requires a texture swapping system over an ISA bus.
You design the textures so they fit inside video memory.
One could also argue that since this would be a larger bottleneck on an ISA card, that ISA accelerators would come equipped with more memory to begin with.

blurks wrote:

I understand, that a 3D accelerator for the ISA bus could make sense under very certain circumstances but in 1996, ISA was already dead when the mainstream 3D API's began to skyrocket.

Yes, but the fact that ISA was no longer relevant is not the same as saying that ISA accelerators would be useless.
There are valid technical reasons for using accelerators over ISA. By the time 3d acceleration took off however, the ISA bus was no longer economically viable anyway. Heck, even VLB was pretty much dead, given that virtually all 3d accelerators only targeted PCI.

spiroyster wrote:

I have the aformentioned Matrox cards and have been dying to get something going with them, but have zero experience with TIGA so not sure as to 'how' programmable they are, or essentially how much I have to make the CPU do in terms
transformation stacks etc. (documnentation on particular card models is scarce, I tried Matrox (UK) and even they are apparently clueless) o.0

You may already know these, but here a few articles from Geekdot that I found :

http://www.geekdot.com/hardware/miscellaneous/tiga/
http://www.geekdot.com/tiga-basics/
http://www.geekdot.com/tiga-programming/

Jo22 wrote:

You may already know these, but here a few articles from Geekdot that I found : […]
Show full quote

spiroyster wrote:

I have the aformentioned Matrox cards and have been dying to get something going with them, but have zero experience with TIGA so not sure as to 'how' programmable they are, or essentially how much I have to make the CPU do in terms
transformation stacks etc. (documnentation on particular card models is scarce, I tried Matrox (UK) and even they are apparently clueless) o.0

You may already know these, but here a few articles from Geekdot that I found :

http://www.geekdot.com/hardware/miscellaneous/tiga/
http://www.geekdot.com/tiga-basics/
http://www.geekdot.com/tiga-programming/

Yes thankyou. Familiar with those.

I got so far as

The 3 yellow(ish) levels are card-dependent, so without them, you’re lost and your shiny TIGA card just makes a nice paperweight.

but I don't have the heart to use them as a paperweight.

I might try Matrox Canada to see if they have any idea of their heritage.

spiroyster wrote:

Yes thankyou. Familiar with those.

You're welcome. 😀

spiroyster wrote:

I got so far as […]
Show full quote

I got so far as

The 3 yellow(ish) levels are card-dependent, so without them, you’re lost and your shiny TIGA card just makes a nice paperweight.

but I don't have the heart to use them as a paperweight.

I might try Matrox Canada to see if they have any idea of their heritage.

I've seen some TIGA drivers at files.mpoli.fi (Metropoli BBS) (1, 2, 3).
Ihink that was the same place where I got the software for a Number Nine TIGA card many years ago.

That beeing said, there's also a Matrox directory.
http://files.mpoli.fi/hardware/DISPLAY/MATROX/

I'm not sure, though, if it contains software for your model.
Perhaps the software is located in a different directory (if at all).

Edit: Links added.
Edit: OS/2 Museum also has a few more TIGA related posts..
http://www.os2museum.com/wp/have-you-seen-these-cards/
http://www.os2museum.com/wp/two-more-tigas/

It's worth having a poke around ftp.matrox.com as they keep all their old stuff (bit of an unstructured jumble though!) going back to DOS and Win3.x era on there.

ftp://ftp.matrox.com/
Just ftp.matrox.com doesn't work (in a www browser). 😉

The photos dont work anymore but I'll always remember when Vets found a boxed version of the Matrox Impression, only supported 3 games.

Bought these (retro) hardware today

http://old.vgamuseum.info/home/1172-matrox-im … ession-isa.html