VOGONS


First post, by mkarcher

User metadata
Rank l33t
Rank
l33t

I recently got my hands on an EIZO branded graphics card. While this company is best know for their monitors, they also sold graphics cards. For example, EIZO had the AA40, which is displayed as one of the C&T 82C480 based 8514/A clones. That card seems to be a straight-forward implementation of the basic 82c480 design idea. The card I recently saw seems to have no photos on the internet yet, so it's time to fix that!

The attachment front_wo_memory_board.jpg is no longer available

This is the successor, the AA41. It differs from the AA41 by being twice as good (measured in PCB area). No, just kidding. It differs from the AA41 by actually extending the architecture. The 82B484 chip responsible for sending the VRAM data to the DAC has been replaced by a custom ASIC with a "spectragraphics" copyright, which is named FFC ASIC by EIZO. This is an abbreviation for "flicker-free cursor". It provides a hardware mouse cursor. This function is integrated in many more sophisticated DAC chips, like the Bt485, but if I understand the documentation correctly, the 6288 16K x 4 SRAM next to the "FFC ASIC" can contan a 256-color 64x64 mouse cursor, whereas the usual "hardware cursor" feature is usually limited to a three-color, or two-color/invert cursor.

Furthermore, there are two further ASICS. This card is claimed to have a local video memory bus, called the GS bus. The "IF" ASIC implements the ISA bus interface, and adds some extra features, like a deeper command word FIFO. The original 8514/A had an 8 word FIFO, and this card is claimed to extend that to 80 bit-blit commands, each of them consisting of multiple words. The second ASIC is called the "DE" ASIC (for "data engine") and implements a drawing engine. While the 8514/A already has an internal drawing engine, the DE ASIC is claimed to provide an arbitrary 3-operand drawing function implementation, which is not present in the 8514/A. This has possibly been added to aid with implementing the MS GDI architecture.

The card is delivered with two feature-connector cables, one for VGA cards with a pin-header feature connector, and a second one for VGA cards with a card-edge feature connector.

The attachment feature_cable.jpg is no longer available

The for chips in the center are the main video RAM. The 8514 architecture accesses words of multiple pixels at the same time. All the pixels inside a word are collectively called a nugget. A 8514/A nugget can consist of 4 or 5 pixels, and the video memory is organized as 256 nuggets per line. This card has 4 VRAM chips, each storing one of the pixels in a nugget, so it uses 4-pixel nuggets, and a screen width of 1024 pixels. There is a 5-pixel-nugget variant of the card, the AA51, which supports a screen width of up to 1280 pixels. (See later post for correct info on the AA51) The RAM chips on the card are 4 bits wide, so the card is limited to 16 colors (at the full resolution), as long as the optional memory upgrade module is not installed. The memory upgrade module adds 4 bits to each pixel in the nugget, upgrading it to a 256-color card at 1024x768.

The attachment MemoryUpgrade.jpg is no longer available

I also got the 5.25" floppy disks and the software installation manual of that card.

The attachment floppies.jpg is no longer available
The attachment ManualFront.jpg is no longer available

I'm sorry for the excessive sharpening artifacts on the floppy drive labels. I didn't notice the over-sharpening of the mobile phone camera when I took those shots.

See the next post for the contents.

Last edited by mkarcher on 2024-04-21, 09:17. Edited 3 times in total.

Reply 1 of 7, by mkarcher

User metadata
Rank l33t
Rank
l33t

These are the contents of the three floppy disks. Due to mechanical surface damage (a dent in it), I was unable to read one of the sectors of Disk 1 (Utilities and Drivers) with a correct CRC. The bad sector is part of AI.EXE, the standard 8514/A compatible software interface driver (the "Adapter Interface"). Luckily, that file is not compressed, the error only affected the last 64 bytes, and many read attempts were close to the correct content. With the aid of reverse engineering and comparing the read attempts in which the MFM clock was not lost, I am confident that I restored the bad sector correctly.

Reply 2 of 7, by mkarcher

User metadata
Rank l33t
Rank
l33t

Chapter one, containing a short hardware description is attached as a ZIP file.

The pages about the configuration utility GACONFIG for DOS and OS/2 are

The attachment GACONFIG_DOS.jpg is no longer available
The attachment GACONFIG_OS2.jpg is no longer available

The 4 pages about the Adapter Interface driver for DOS are attached as a ZIP file again.

Reply 3 of 7, by Disruptor

User metadata
Rank Oldbie
Rank
Oldbie

It seems like here has been listed a similar card:

imi wrote on 2020-02-12, 17:46:

a very intriguing EIZO AA41 (CAD accelerator?) card

www.vogons.org/viewtopic.php?p=825315#p825315

dr.ido wrote on 2020-02-13, 10:03:

Interesting - It doesn't look like a typical TIGA card. The piggy back ISA connector is likely for an optional EGA or VGA card - This card probably doesn't do any standard PC modes or DOS text modes.

derSammler wrote on 2020-02-13, 10:20:
I've got Win3.1 drivers for Spectragraphics cards. Can't attach anything here, however. […]
Show full quote
imi wrote on 2020-02-12, 20:15:

unfortunately I do not have any drivers for the EIZO card, and I didn't find a lot of info about it in general, but I did get the perfect card to go along with it eventually a while ago ^^

I've got Win3.1 drivers for Spectragraphics cards. Can't attach anything here, however.

dr.ido wrote on 2020-02-13, 10:03:

Interesting - It doesn't look like a typical TIGA card. The piggy back ISA connector is likely for an optional EGA or VGA card - This card probably doesn't do any standard PC modes or DOS text modes.

Why do you even think this is a TIGA card? I'm pretty sure it's a VGA card supporting the DGIS api, because that's what Spectragraphics was known for.

Reply 4 of 7, by Predator99

User metadata
Rank l33t
Rank
l33t

Nice card, many thanks for sharing the drivers!

Does it work?

Reply 5 of 7, by mkarcher

User metadata
Rank l33t
Rank
l33t
Predator99 wrote on 2024-04-20, 08:45:

Nice card, many thanks for sharing the drivers!

Does it work?

In an initial test, it worked perfectly. But that's just an initial test - the initialization code in the BIOS works, and GACONFIG works, including some diagnostics and a test screen. I tried to run AUTOCAD with that card booting from floppy and using EtherDFS for mass storage. This broke in many ways. I managed to get Autocad running in VGA mode, but it always crashed on me in 8514/A mode or with the EIZO drivers. I'm confident that's a software problem, though.

Reply 6 of 7, by mkarcher

User metadata
Rank l33t
Rank
l33t

Further research revealed that I was wrong about the AA51 operating in 5-pixel nugget mode. While it is true that the AA51 supports 1280 x 1024, and it is also true that the 5-pixel-nugget mode of the C&T 82c480 has been specifically designed to support 1280 x 1024 with minimal overhead, that's not how the AA51 works. Instead, it has two banks of memory instead of a single bank, each bank being organized as 256K x 16 (without memory upgrade board) or 256K x 32 (with memory upgrade board). This design still stores 4 pixels in one memory word (4 16-color pixels at 256K x 16 or 4 256-color pixels at 256K x 32).

The very elegant (but inflexible) design of the 8514/A adapter is that one row of the VRAM chips equals to an integer amount of scan lines, so the "magic VRAM operation" to load one row (aka page) from the video memory array into the dedicated video data shift register only has to happen at some time during the horizontal blanking, relaxing the timing requirements drastically. The original 8514/A used 4 banks of 64K x 4 VRAM chips, that contain a 256-entry shift register. As there are 4 pixels per nugget, and each shift produces the next nugget, this dictates the horizontal scan line length of 1024 pixels. In 640x480 mode, there is an unused area at the end of each scan line. The AA41 (and other C&T 82C480-based cards) use more moden 1M x 4 VRAM chips, which have a 512-entry shift register, but this capability is possibly not used at all (it would allow the shift register to be reloaded every 2 scanlines instead of each scanline). Adding a fifth pixel to the nugget is not a standard 8514/A feature, but a C&T addition, which was quite common, because it allowed cards optimized for 1280x1024 with exatly 1.25MB of video RAM. On the other hand, this way of adding extra memory means that the line still consists of 256 nuggets, and at the same horizontal scan rate the design requires the same nugget rate. The result is that 1280x1024 in 5-pixel-nugget mode is limited to scan rates that require interlaced display. And likely that's the reason why EIZO went a different route for the AA51.

The 82c480 in two-bank mode interleaves the two banks, so the line length doubles from 256 nuggets to 512 nuggets (2048 pixels). Due to interleaving, each bank only sees half the nugget rate, relaxing memory timing requirements. Another important change on the AA51 compared to the AA41 to facilitate high-performance display is the use of a high-performance DAC instead of the Bt478 "standard" DAC. The high-performance DAC in this case is a Bt458 (not a Bt485, though!). This DAC is able to input a nugget at a time (even 5-pixel-nuggets) instead of a pixel at a time. I suspect the Bt478, that is also present on the card, is only used in VGA passthrough mode. Another possiblity is that the Bt478 is used up to 80MHz dot clock - the code in GACONFIG flips a bit if the dot clock exceeds 80 MHz, and that's also the upper limit of the best speed grade of the Bt478.

This also explains why the memory upgrade board on the AA41 I posted earlier has 4 empty spots: Those are only populated on the AA51 memory upgrade board (which uses the same PCB), but has to extend two banks to 32 bits (8 bits per pixel) instead of of only one bank.

Furthermore, these cards are also said to exist in MCA versions, named AC41 and AC51. There is software support for those cards. The Microchannel card ID is identical for both cards, it's EF7F.

See https://www.ebay.com/itm/154623015291 for pictures of the AA51. Disclaimer: I bought this card right now. This post is not an advertisement for that auction, and has only been posted after the transaction happened.

Reply 7 of 7, by imi

User metadata
Rank l33t
Rank
l33t

cool, thanks for posting the drivers and manuals and @disruptor for bringing this to my attention after all these years ^^