Amiga 2000, 2MB chip-ram upgrade, and Genlock timing

[eeguru]

I have a couple dead and one dying Amiga 2000s - the later with extensive battery leakage but still works. I was planning on replacing the MB in the dead ones with the rescan/remake by Floppy209. However I'm making a few enhancements first - like using 30 pin SIMMs, ATX power connector, and a built-in scan doubler + DVI output.

I was also looking at the 2 MB chip RAM upgrade mod. I had started to incorporate that mod with a jumperable selection of 8732A + 4x 256K SIMMs or 8732B + 2x 1M SIMMs. The problem is the B version of the chip repurposed the XCLK input to be the A20 input. In later systems that support native 2MB chip RAM, the XCLK input was moved to Super Gary on the 3000/4000 and Budgie on the 1200. And those chips, respectively, handle the dynamic switching of 28 MHz to Super/Fat Agnus for local video timing generation. So to do that mod (I calculate), effectively breaks Genlock cards - including the Video Toaster - from working.

A) Is my assumption about broken Genlock correct? I'd like to eventually run a Toaster in one of the machines.

B) I noticed on the 600 schematic they direct muxed the 28M input to Agnus with the local oscillator and XCLK via XCLKEN. It seems reasonable to me that I can also do the same with the new board. Agnus generates all the video timing internally - and divides the 28M directly by 4 to generate all the CPU clocking - and by 8 to generate the color burst reference frequency. Anything I'm missing?

-A

 

[Chuck(G)]

Any reason to use period-more-or-less-correct DRAM? 512Kx8 SRAM in DIP isn't terribly expensive. Go to SMT and it's even cheaper. No worries about refresh or finding SIMMs. Heck, you could max the configuration out with SMT SDRAM for less than you'd pay for a stick of 256K SIMMs.

Just thinking aloud.

 

[eeguru]

Considered that. Amiga 'chip-memory' is poorly named. It's really graphics ram. If that is the only RAM in the system, it is also used for everything else. Agnus contains the blitter and copper and internally services Denise (GFX controller) through a dedicated output port. So you have to add chip/graphics RAM to the existing memory port on Agnus - which is setup for DRAM. So I'd have to have both 4 memory chips and the needed latches to demultiplex the row/column. Not hard and actually less logic than the current buffers and latches in the DRAM design. But ultimately I still want it to be 95% the original A2000. 256K SIMMs have the same chips as the current Rev.6 board but on an easy to pop-out carrier PCB. So it's a 1:1 swap and SIMMs are easier to find than the discrete RAMs and replace if one fails.

People have designed a drop in replacement boards using later Agnus chips (8375) that have memory + glue logic on-board with a compatible footprint. That's not really the issue. Even those solutions appear to break genlock board compatibility. I have a SuperGen-2000S and want to pickup a Toaster at some point. So keeping compatibility is a must.

I think I've worked out everything tonight. I've replaced a 7432 near Agnus with a 22V10 that combines the original logic with the XCLK muxing into the 28M input on Agnus + the RAS1/DRA9 demuxing for different RAM configurations. 4 layer boards of that size are not cheap. I'm trying to mitigate risk of infant mortality by keeping changes to a minimum.

 

[drdoc]

Have you considered the scarcity of 2MB Agnus chips? The A600 variants are still available here and there**, at exorbitant prices, but I've been looking for an 8372B for a year.

The A500++ and various A3000 reimplementations have rendered 2MB chipram almost unobtainium. It's so bad that there's at least one adapter that allows running a 1MB 8372A Agnus on a native 2MB A500++ motherboard.

Doc

** Well, OK. "here" is "Ebay" and "there" is "In your dreams, Baybee!"