Model3 annotated board photos

[Bart]

I just dug out an old breadboard I had sitting around with an aborted Sega Genesis 6-button pad de-multiplexer circuit that would have read out the button states and presented them on a parallel interface suitable for hooking up directly to my Model 3 I/O board. The idea was to be able to play Virtua Fighter 3 (albeit still without sound because I have no idea what sort of audio signal comes out of the I/O board pins and how to connect it to speakers) with a Genesis game pad.

I'm going to try to get it working this week.

But then I realized that I can use the input pins to send over program data. I had discussed this with Charles MacDonald years ago but thought it too hacky. Ideally, we would also tap the board LEDs to use as outputs for bi-directional communication but I don't want to solder stuff onto the board. However, it probably won't be too difficult to write some code to monitor the input pins for incoming data and copy that to RAM. Sitting in a tight loop and using one or two pins for clock signaling, the PowerPC could monitor the lines for incoming data. Then, I can transfer programs quickly without ever having to touch the EPROMs again!

That will take me a bit longer because I'll have to find an Arduino board (I do have one that I really like working with, though) but it should really make testing a lot faster than ever before. Printing out values to the screen will probably be more than enough for our needs but we could also investigate how to get the serial port working. Doing that via trial and error with EPROMs would be too time consuming and would likely destroy the sockets sooner or later (they're already in bad shape from all the pull/re-insert cycles I've put them through).

[model3man]

Sounds like a neat idea Bart. Are the LED also usable as input? I will post in your other thread with some part numbers to make a harness.

I was wasting time the other day looking at 90s workstation graphics cards. Lots of weird and crazy stuff (HP's visualize fx6 used six PA-RISC cpus as geometry processors....)
But the main thing I noticed was that some of these cards used the exact same 3DRAM and CDRAM used on the Model3 graphics system!


Here is a partial list of what I've seen with these parts.

Sun Microsystems several variants available. Creator3D and Elite/Expert3D.
AccelGraphics AccelECLIPSE (rebranded/sold by E&S)
Evans & Sutherland Lightning 1200
NEC TE3V/TE4E and derivatives (very rare)

https://www.okqubit.net/boards/e&s.html
https://web.archive.org/web/20000824035 ... eet_lx.PDF
https://support.express.nec.co.jp/wpape ... s/te4e.pdf

[Bart]

I found this hilarious video on Youtube of an HP Visualize Fx6. I think Richard Stallman is the narrator?

Thanks for the reply to the other thread! I'll order the parts tonight All I'm getting is the PCB stack in the metal case with I/O board. No filter board or anything else as far as I know. My other two Model 3 PCBs are essentially like this as well.

I spent some time a couple weeks ago trying to track down folks who might have worked with a Real3D Pro-1000 but no luck yet on finding one. I've got a few more emails to send, though. Doubt we'll ever find one. But there just has to be a unit hiding somewhere....

[model3man]

I found a bit more information about the very odd and unique Mitsubishi 3DRAMs used on these boards.

In 1994, Sun was developing the next workstation visualization card, and the lead architect correctly identified the ram bandwidth problem, especially when drawing zbuffered polygons with transparency or AA which necessitates a read-modify-write cycle. His solution was to exploit cache locality inside the DRAM outside of the bus. So the pixel blending and span caching happens on the DRAM itself.
Sun and Mitsubishi collaborated to produce these custom rams, with Sun getting an exclusivity period. As a result, this 3DRAM was used on all Sun workstation cards from about 1995-1999. Further, they also developed a successor, 3DRAM64 which has even more bit depth and capacity - which was used in the XVR-1000 and XVR-4000 cards which were geared towards very specialized molecular modeling applications.

I really am curious why they didn't solve the bandwidth problem with a tiled renderer which could be easily parallalized and wouldn't incur such a steep penalty with their available geometry processors.

Looking at the Sun graphics cards, the ICs were also fabbed in Hong Kong with the same package marking format as Sega's. There sure seems to be a lot of cross pollination.

[model3man]

I now know that the Real3D chipset as used on Model 3 was fabbed by LSI in Hong Kong. Many, many things from Sun, SGI, and others used this fab.

Edit: Just tested the UART inputs to the MAX238 rs232 interface on the CPU board. Unsuprisingly, these signals don't seem to be driven and have no activity.

[Bart]

Meaning there is no chance they are mapped anywhere in CPU space? Or that the controller hasn't been activated in software?

[model3man]

They are probably mapped somewhere still. Either all the uart stuff was compiled out, or the uart peripheral in the sega asic was turned off for the final build.
I'm sure it's entirely possible to populate the pcb with that MAX238 and get some comms going if you knew how to talk to the uart asic, but I didn't see anything.

Now I have Sega Rally 2 and Lost World so i can try also checking the uart on those as well.

[biggestsonicfan]

Sorry to possibly necrobump, but has more work been done to understand the RS-422? I've built my own converter using an FTDI RS-422 to USB converter and crimping JST-NH pins to the wire end and attaching a 6 pin connector, but I can't get any data out of the RS-422 on the filter board's connector. Typically the LEDs on the usb show activity but as the arcade board pumps data to the RS-422, nothing.

[Bart]

Sorry to possibly necrobump, but has more work been done to understand the RS-422? I've built my own converter using an FTDI RS-422 to USB converter and crimping JST-NH pins to the wire end and attaching a 6 pin connector, but I can't get any data out of the RS-422 on the filter board's connector. Typically the LEDs on the usb show activity but as the arcade board pumps data to the RS-422, nothing.

Unfortunately not I would *love* to have this working. It would make it really easy to write firmware that can load programs directly from a PC. No more burning EPROMs (the sockets are wearing out on my VF3 board).

The problem here is that because no game seems to use it, we have absolutely no idea where the registers are mapped. We can probably guess a rough set of address ranges for where it might be but that's insufficient to make progress. However, I did just have a long shot idea. We have the Real3D Pro-1000 firmware. Maybe it also had an RS-422 connector for debugging and perhaps its firmware still talks to it? The Pro-1000 was basically like the Model 3. We don't know what the boards looked like but it also had a PowerPC, obviously the video chips (probably on their own board), and communicated with Windows NT PCs via SCSI cable (probably using the same SCSI controller present on Model 3 Step 1.x boards). OTOH, the existence of the SCSI connection would have made RS-422 redundant, unlike Model 3 arcade boards.

[Bart]

The manual for the Pro-1000 says there is an RS-232 interface ("External RS-232 Interface). Not 422. There is also a second port on Pro-1000 described as "RS-232 Syncs for Host".

EDIT: re-read the thread. RS-422 is indeed the MIDI interface which is exposed on the filter board, so this is indeed relevant!