This VGA Card does Polygon Fill, but is it FAST? 

I believe I have used that book, though there are a few out there that cover the 8514/A, so maybe this is one I didn't look at yet. Thanks for the reference.

Yeah they are a bit weird.

@@PCRetroTech I thought they were still called SIP even with the zig aag?

@@askjacob I'm not sure. I think you better ask @ropersonline about that.

That's the main reason I like retrocomputing so much. We can go back and reexamine all of these things more carefully and spend the time experiencing what we missed because it all moved past so rapidly. It was an era where things were comprehensible right down to the hardware level, even for an amateur, unlike the complex machines of today.

I had the 20inch 6fg, initially driven by a 4MB Diamond ViperVLB. I didn’t actually do CAD work, but had to be able to run AutoCAD to check inbound files before my software converted the drawings into machine control instructions for manufacturing (as I remember, the Weitek P9000 based ViperVLB had horrible AutoCAD drivers). I loved that monitor, kept it for 13-14 years, even though it ate a whole desk. Was my go-to for gaming, as a 20 inch NEC MultiSync was pretty badass, right through the days of the 3dfx Voodoo and RivaTNT. I think I was still using it as my primary monitor into the era of the GeForce 3 (true GPU’s with programmable shaders).

Thanks, I appreciate your comments.

Thanks. It's great to know there is an audience out there. And from the views and comments on this video, it seems that is certainly the case.

I wonder if there is already a driver that does this.

@@PCRetroTech I just checked it for the old XFree86. In the documentation about the ATI drivers is stated: "Mach8 series: Graphics Ultra, Graphics Vantage, VGAWonder GT (None of the 8514/Ultra and 8514 Vantage series is supported at this time)" And in the documentation a little further: "Also, given that IBM 8514/A and ATI Mach8 do not allow CPU access to their frame buffer, the driver will currently ignore these accelerators." So no IBM 8514/A support in XFree86 4.4.0 and i doubt that the support is better in the newer XOrg. I rather would assume that the old stuff is going to get thrown out. Your chances might be better with Windows 3.1 and Windows NT 3.1. EDIT: Or IBM's OS/2.

I guess unless you were a professional back in the day who needed one of these, it would have always existed as unobtanium.

@@PCRetroTech Yes very much so. Too rich for my 16-year-old self!

I bought a PS/2 model 70 on eBay just because I noticed it had a video-shaped connector in one of the slots, and decided to take a gamble that it would be something interesting like an 8514/A. Turns out, it was! Fully loaded with the RAM expansion even. Shortly after, I got hold of an XGA card, kinda by dumb luck as well. I’m glad to have them. They’re fascinating cards, if not super useful these days.

@@nickwallette6201 How one earth did you get that lucky!? The 8514/A cards were not popular back in the day. I guess spotting the video adapter could have meant something. But why not just a plain old VGA of higher spec than onboard?

@@PCRetroTech In a Microchannel machine with onboard VGA? :-) The odds were pretty good it might be 8514, or at least XGA. Now, having both land in my lap was a stroke of complete luck. If it makes you feel any better, the PS/2 appears to have been used in a kitchen, I think, and emits an aroma of garlic and frying animal fat while running. haha Perhaps it found a second life as the bookkeeping PC in a family restaurant, once its prowess was overtaken in whatever business shelled out for such a high-end 386 when it was new. Or, just a very good salesman.

I would have to reverse engineer the IBM driver, since it does it correctly. But yeah, I think that will be the only way unless there is someone out there with some code that does the right thing.

I hope so too

Thanks. Yes it was a bit of a disappointment, but all good fun though.

@@PCRetroTech it's all about the journey, man 😋

Yep. Kind of disappointing at the end. But to be honest, even if it was fast, you’d still be pretty limited in what you could do without any hardware Z-buffer logic. You could create a very fast Asteroids game, though. 👍

@@AftercastGames There'd be some limitation, but with Painter's Algorithm one can get fairly far. Or there is Binary Space Partitioning. Z-buffer isn't the only way. I'd have been happy with something Doom-like. That would be possible on this card I'm sure, but I was hoping for something really spectacular. Actually since fiddling with it I did think of something it definitely can do well. But more on that in a later video.

@@PCRetroTech Be interesting to see if the genuine IBM hardware had this same bug.

I'm glad people enjoyed this video. There'll be a short delay before any more videos on the channel though, as I'm currently moving country. But the channel will be back soon enough.

Yes, that seems to be the setup I'm still not super sure why so little for the framebuffer. It seems like if you are going to support high resolutions on the card for the 8514/A part, you may as well support them for straight VGA as well.

@@PCRetroTech Maybe I'm wrong and 1M is for framebuffer (768kB for 1024/768/256 mode) and 512k dynamic RAM is for accelerator. Vector operations not require much memory, just store functions code and endpoints, processed output image is stored in framebuffer memory for displaying.

@@adamw.8579 No, I believe you had it correct. The operations are actually stored in a very small FIFO buffer (8 ops in the case of 8514/A, 16 ops for ATI). The 1mb is for the pixels in the 8514/A modes. In the Mach 32 this was combined with the VGA framebuffer into a single framebuffer.

A Voodoo card requires a PCI slot, thus it will not run on a 386DX.

@@OpenGL4ever You are correct, Gravis was on my 386DX, but later I had 486DX4-100 on board with PCI and VESA local bus slot. This was near 30 years ago, my school age.

The best way to find that would be to do an FTP search for one of the files that you can see in the directory listing. Some of the larger FTP search engines should bring it up. Of course you need an FTP client to download it. I believe Chrome no longer has one built in by default.

Indeed, though I don't see any interlaced flicker with my card. But the original 8514/A did that for sure.

Yes, I noticed that you can run it as an 8-bit card. There is information in the documentation about doing 8-bit transfers. If the software is written for 16 bits it won't work of course.

Interesting suggestion, but this is just a 286 and I'm pretty sure the bus is not overclocked. I will look at the manual for the board and see if it has a jumper for that though.

@@PCRetroTech It's a standard AT motherboard, or a clone from taiwan? Taiwan clones used to allow faster 286s to be used by mangling the base clock. I remember seeing 12 and 16 MHz 286 machines at that time! (interesting, some taiwanese MBs used a 20MHz 286! Never saw one of that!)

@@LisiasToledo It's 12MHz. It does have a turbo button. Unfortunately I can't check it at the moment though as it is on its way to the UK.

I'm pretty sure back then there was no concept of overclocking the "ISA bus", it just ran at the CPU clock on earlier machines. Whether you card would work at higher frequencies was pretty much hit or miss. The idea of a standard frequency for the ISA bus only really came in once the CPU was on a "local bus".

@@TheUAoB Yes, you pretty much had to change the crystal and hope for the best.

The ones in the 8514/A Demo Diskette here: mail.lipsia.de/~enigma/neu/ibmpc.html ??

That sounds much more like it!

@@PCRetroTech it was quite fun getting ram for it as someone had removed it!!! 😢 Around 2004 I was upgrading my old IIci to full spec and I am so glad. It is one heck of a silly machine now and was my first internet (non bbs only) pc. I have a very old hub just to connect its old networking card to Ethernet and I love it. It has been maybe 4 years since I booted it up I should set it up to nostalgia bask.

@@hicknopunk That's the way!

There is a read mask, and for some bizarre reason this is rotated by one bit. But I fiddled with this to no avail.

@@PCRetroTech I assume you already ruled out whether the copy operation is using the current (foreground?) color as the background of the copied rect at the destination. That's the only other thing I could imagine producing that effect.

@@ScottLahteine I think so. The documentation of the blitter system and the write and read masks is a bit brief, so I could be overlooking something. But I tried a wide range of possibilities, including some that were suggested after reading similar documentation for other similar chips.

TIGA would be another good one to investigate eventually.

@@PCRetroTech That one should be more capable since there are actual 3D games that were made in the arcade using that chip :)

@@PCRetroTech yes indeed. Btw. it must have been mid 90s. I started with NEAT AT 386 16 MHz AMI Board. added a non Intel copro. Speedstar 24, MFM 20 GB on EIZO 9080i and STRKON CAD productive. The wealthy DiCAD customers had Compaq Deskpro 486 33 MHz and SPEA Gallery GA connected to Sony Triniton 20" SPEA labled fixed freq. Monitors. I still have my 486er wirh SPEA Hilite and Miro TIger 12. A cheaper alterative to the expensive SPEA Gallery cards. Think I do have all driver disks and manual in Original SPEA Box. Hilite needed an seperate VGA looped in for MS DOS. On Hercules or VGA monitor special extra programms could be accessed. I promise you, with this DOS based STRAKON system I am still able to create 2D CAD plans more efficient than with all contemporary CAD Systems. STRABAG International planned airport Basrah these days before it was bombed by the US Military.

Sorry, 20 MegaBytes 🖖

@@Dxceor2486 That I did not know. Do you have a reference for me (usual problem with links on YT).

I'm really not sure. I had a few theories along the way, but disproved them all. I did find one 8514/A card with 1.5mb of RAM though, so it seems to not have been too uncommon. My best guess is it is needed for even higher resolutions that the ATI card supports that 8514/A doesn't.

@@PCRetroTech it's just odd that they use two types of RAM. Could be some other explanation, but from a manufacturing point of view it's better to have just one type.

@@kuro68000 This might be the difference between the Ultra and the Vantage card. Perhaps the Vantage has just the ordinary chips and only the Ultra is upgraded with the fancy ones.

I figured it out. The 0.5mb is for the VGA core and the 1mb is VRAM for the 8514/A coprocessor. That's a pretty surprising design, but that's how ATI did it. They only did this with the Mach 8, not the Mach 32.

@@PCRetroTech so maybe there is a bottleneck there, either due to the video DMA or moving between those banks. If you did 10k polygons off screen it might be faster.

The 3000 probably not well. The 4000 was a big improvement and should be comparable if programmed the right way as I understand it.

@@PCRetroTech Being a 8-bit card I thought so. I only really use it because it makes it easy to hook up monitors to my 5160 XT.

@@ParallelSyntax For that they are very useful. They are nice cards for other reasons though.

It could be, but I wouldn't expect the slash if that were so. It does seem odd to use the same ROM for the Ultra and Vantage though, as it actually displays what it is when it starts up. But maybe it actually interrogates the other components to determine which one it is and the same code is used for both.

@@PCRetroTech That would be a reason to use an EPROM though, wouldn't it? You could have different ROMs for the different cards. It's just a matter of what version of the ROM you programmed it with. I could be wrong of course. It's easy (but potentially destructive) to find out. Is there a glass window for an UV eraser under the sticker?

@@TOTALANIHILATIONYEAH Probably. But I don't know how we'd check if you are right. It'd have to be documented somewhere.

@@PCRetroTech To find out, you could weigh a chip with the same dual inline package and pin count and then compare the measurement. The EEPROM version with the glass window should weigh a little bit more.

That's sort of what the ATI does as an extension. Technically there could be self intersecting, nonconvex polygons, so a list of left and right coordinates is not enough. It does have to find the edges of an outline in memory.

@@PCRetroTech An intermediate version would be reading the left/right pixels from offscreen and then writing them to onscreen memory.

@@thewhitefalcon8539 Yes, that's what the ATI extensions allow you to do that 8514/A doesn't.

The ideal case was actually three buffers, so that if you got another screen ready before the hardware was ready then you could just change which was the "pending" screen.

No, the only features I am using are the 8514/A features. ATI documents these separately and I also checked in other 8514/A documentation.

Thought a bit about this, and I've gotten a few potentially useful ideas. If this card is a 2-in-1 package, can you merge both video signals somehow? As in, what causes one chips output to be displayed over the other? If you can do that, depending on how the signals are mixed, you could use that for some funky blending effects. Maybe if you can change SHARE_CLOCK (IO 16EE.0) that could work. The repeared chars might be bad latching or bad memory on a specific area of VRAM? Maybe run some mem test. I'm guessing this card doesn't have fast XOR blitting, or at least doesn't use it in this case. Maybe it has some kind of microprocessor for these operations, but (being a microprocessor from the 90s) it's slow. Such an exclusion is understandable, if you're buying such a card you (probably) aren't expecting EVERYTHING to work at full speed, ATI has to cut costs somewhere, after all. Plus, if they got everything right the first time, the Mach 32 wouldn't have been necessary! Potential workaround: Request blitting of 1-high by whatever-wide rectangles so the fast blitter (in memset mode) can be used as much as possible. Also, if the blitter works on a "blit x, skip y" basis, that could allow for drawing fast parallelograms Also, you said the final part used ATI extensions, maybe you're being confused? About that part, you might have forgot to enable the ATI extensions, or something changed the blitter mode between you setting it and the Combined XOR and Blit happening, or... Actually, can I rant a little about the lack of documentation? I'm afraid that, if the info you have isn't enough, you'll have to reverse engineer the drivers and/or the card... Unlike CPUs, which are decently documented (mostly...), the graphics side seems to have almost no information for anything outside the standard CGA/EGA/VGA (if that, at times!) and _maybe_ some common clones. Almost nothing about low-level SVGA (aside from VESA stuff), even less that I could find about accelerators... no wonder GPU programming is considered dark magic if even the simpler, "old" stuff is just _gone_... Thankfully the wayback machine has archived some Mach8 drivers, so at least reverse engineering material _exists_... Maybe the OS/2 Museum has more information assuming their search 9 years ago went anywhere.

@@FavoritoHJS The 8514/A output is overlaid on the VGA output I believe. It is like the Voodoo card in that it is passthrough, albeit digital and not analogue. There are two chips on the board. One is a VGA core, the other an 8514/A coprocessor, which does lines, rectangle fill, polygon fill and blitting. As far as I know there is no XOR blitting, though one can select which bitplanes are written. However, the bottleneck is the polygon fill, not the blitting. The Mach 32 added fonts and other resolutions. I don't believe it is faster (I have one). I didn't use the ATI extensions yet. I used 8514/A documented functionality only so far. I did suggest that I *could* use the ATI extensions. They offer a combined fill/blit operation which should be faster. But the ATI extensions are in the other half of the 400 page documentation. I didn't find time to digest it yet. IBM deliberately didn't document the 8514/A at the register level. Numerous companies reverse engineered it and documented it, including Chips & Technologies, ATI and Western Digital. I was mistaken in my previous video about the IIT card. XGA is not register compatible with 8514/A (this means the Wikipedia article is definitively wrong by the way). The Mach 8 is documented well by the way. Well, with the exception of the blitter not being documented very clearly. But this is a common feature for all three clones I am aware of. Some of the books out there on the 8514/A also just plagiarize the ATI docs entirely, which is also amusing.

Correction: there is XOR blitting. But it is complicated. There's a notion of a monochrome source and a colour source (probably not what I would have named them). There's also a notion of a foreground and a background mix. Each mix can be set to an XOR of a source with the destination. There are various ways of selecting which pixels display the foreground mix and which display the background mix. The source data can be taken from the blitter source. The destination is then blitted from the relevant mix to the destination.

The RAM is split between VGA RAM and coprocessor RAM.

@@PCRetroTech Ah, that's possible. By the way, I remember having a VGA card that had some kind of acceleration (could be S3 Virge but I am not sure) that caused my friend to stare at the monitor with disbelief because the card could render a mouse cursor in hardware (it was supported by the OS) and the cursor could be drawn outside of the desktop area - I mean, when you move the mouse cursor to the far right, normally you can see only the tip of the mouse arrow but the whole cursor was rendered on the black border where in theory is no video ram ;)

@@tuomollo Interesting. I wonder if it could do other sprites as well.

@@PCRetroTech I don't know, maybe. I guess drawing something on the border could mess the auto calibration in the LCD monitors.

@@tuomollo I would expect that to only depend on sync.

3080 or 3090 dont cost that

Not too far off. But these weren't aimed at gamers, but semiprofessional I think.

Man, GPUs are _never_ going to get cheaper at this rate.

@@nickwallette6201 um they been dropping for weeks

@@primus711 No, I meant, if a high0end GPU was ~$3K in the 80s, and they're ~$1K today .... It was a joke.