RCBus Kits

357 views
Skip to first unread message

Sergey Kiselev

unread,
Feb 27, 2025, 2:37:37 PM2/27/25
to retro-comp
Hi,

I listed some a few RCBus kits for a couple of projects designed by Tadeusz Pycio on Tindie:
I also have several Z80-512K kits of my own design. The Z80-512K module includes Z80 CPU, 512K RAM, 512K Flash ROM, Zeta SBC V2 style, RomWBW and FUXIZ compatible MMU, clock generator, and an optional reset / CPU supervisor circuit. It is a nice little module that will save some slots on your backplane:
Thanks,
Sergey

Steve Cousins

unread,
Feb 27, 2025, 4:38:04 PM2/27/25
to retro-comp
Hi Sergey

That's great news. I've added links to these products on my website: 

Note: the documentation links on the Backplane 4 Tindie page point to the Compact Flash documentation.

Steve

Sergey Kiselev

unread,
Feb 27, 2025, 5:24:22 PM2/27/25
to retro-comp
Thanks Steve!
I updated the documentation links for Backplane4

7alken

unread,
May 14, 2025, 9:50:32 PM5/14/25
to retro-comp
hi Sergey,  here this MPGSEL_0,1,3,4 is correct? isn't this rather 0.1,2,3 ??
https://github.com/skiselev/Z80-512K/tree/main

peeked into docs to get how page/bank mapping for RomWBW/FUZIX is done best, so this is standard? looks flexible, nice;
few days ago I tried to find this in RomWBW or supported hardware manuals, without success ...

Petr

Sergey Kiselev

unread,
May 19, 2025, 1:14:31 PM5/19/25
to retro-comp

Petr,

Apparently I can't count ;)
Fixed... it should be MPGSEL_0 - MPGSEL_3 ;)
That's what RomWBW uses:

As far as the best page/bank mapping, I am not sure if there is a "best", but my 16 KiB * 4, initially implemented in Zeta SBC V2 seems to be fairly popular now. It made it to several RC2014/RCBus memory modules (Scott Baker's RC2014 compatible memory board, Spencer's 512k ROM 512k RAM Module), and, in addition to Zeta SBC V2, I've used it in Easy Z80, Z80-512K, and Tiny Z80.

7alken

unread,
May 19, 2025, 9:56:27 PM5/19/25
to retro-comp
thanks Sergey, its that Z2 MMU ya? I know there is default 32/32, Alan likes that 48/16, Bill has its ZRC, I didnt peeked yet into RmWBW source to fetch all, had busy by collecting design notes in timeline and details ... will post thread right now, yesterdays fulltime work, rather aggregation of many months and clearing ideas/notes by slow drawing; I need something modular but little-bit smaller and ... and more KiCad, and Verilog challenege, something mostly in Bill Shen CPLD steps; ... that Z2 looks nice, I want to try also adress that already fill 8bit IO and mapping also this on different machines and with low/high rom-vectors; will see... U2 ))
Petr

Alan Cox

unread,
May 20, 2025, 8:43:37 AM5/20/25
to 7alken, retro-comp
On Tue, 20 May 2025 at 02:56, 7alken <antos...@gmail.com> wrote:
thanks Sergey, its that Z2 MMU ya? I know there is default 32/32, Alan likes that 48/16,

It really depends what you are running. For MP/M 48/16 is fairly essential to get a reasonable TPA.
Given the choice I'd like what the old machines had (16 x 4K pages) but the 74F410 has been unobtainable for many years.

Alan
 

Sergey Kiselev

unread,
May 22, 2025, 1:48:54 PM5/22/25
to retro-comp
Yes, Zeta SBC V2 MMU... I think Wayne and Tadeusz refer to it as Z2 MMU. I am fine with that ;)
There is a bit of history of the design given in my post here

Generally speaking, more granular paging allows for better flexibility. As Alan said, 4 KiB pages would be nice. But that would require larger page table, and therefore it is more difficult to implement. Even the 74LS670/74HCT670 are getting expensive now...
For the Z80 based projects, 16K x 4 seemed like a reasonable tradeoff to me. It allows 32/32 KiB split, 48/16 KiB split, or really any arrangement of 16 KiB pages, if one wants to do it (I believe FUZIX does that)

Alan Cox

unread,
May 22, 2025, 2:45:31 PM5/22/25
to Sergey Kiselev, retro-comp
On Thu, 22 May 2025 at 18:48, Sergey Kiselev <skis...@gmail.com> wrote:
Yes, Zeta SBC V2 MMU... I think Wayne and Tadeusz refer to it as Z2 MMU. I am fine with that ;)
There is a bit of history of the design given in my post here

Generally speaking, more granular paging allows for better flexibility. As Alan said, 4 KiB pages would be nice. But that would require larger page table, and therefore it is more difficult to implement. Even the 74LS670/74HCT670 are getting expensive now...

Might soon be time to use 4 HCT573s and a 139 unfortunately. Tempting as a project because it could then all be AHCT which would let me run the 65C816 at a higher speed.

Alan

7alken

unread,
May 23, 2025, 2:06:09 PM5/23/25
to retro-comp
thanks guys, as I am going CPLD-way (fingers Xed), I am only scratching what all is/was used, asking the machine for history of 16x 4kB, she mislead me also to Zeal8bit, where I stayed shocked almost, whats going on there ... well, new OS, new video, interesting; but reading the docs, its 4x 16kB also ... so I in fact didnt found yet any example of usage of that 16x 4kB (and this considering for 8bit 64kB addresable); for 16bit systems having usually at least A19 so 1MB or A23 so 16MB (for me, almost maximum for any kind of retro) ... as there are SDRAM chips larger, there still can be some banking for fast ramdisks (say, poor mans SSD in retro)
so are there any bankings for these larger systems? there can be also 16x 64kB (fast thinking) with 256 banks so 16M total, or say 64x 256kB so 256M (total max) even for largest 32bit system, still considerin 24bit addres as total practical limit, anyway ... if interested, start separate thread about banking or point me to existing one, if it exists ... to continue .... now I found for example this interesting slides (... and nobody here of course want to swap wirtual memory to disk, I am pretty sure ))
https://pages.cs.wisc.edu/~shivaram/cs537-sp23-notes/vm-tlbs/cs537-vm-tlbs.pdf

(as side of Zeal8bit reading, I was already before that specifying simple continuation of Bill Shen VGARC  to allow 256 colors and 8 tilesets of 128 chars (with own ink/paper palletes) for 64x48 8x8bit concept, having color/tileset attributes, so 16kB dual-port sram (chars, attrs, tilesets+palettes) - something simple enough to allow as fast as possible syntax coloring editing similar to Sinclair QL / SMSQ/E "ED" - for me the squared(!) fonts are VERY nice - amd I also ramdomly found SAA7105H video encoder which has "parallel byte per pixel" strem input (chip handles 256 colors LUT of true-colo 24bit RGB, but I am not sure if it can pass-through also VGA output - I will post my spec below to 7xsys - you know, I am not targeting games, but mostly that as fast as possible syntax coloring editing and also windowing/glyphs, some virtual LEDs, may be even blinking, in sparse cases ... fascinated here by 256color Aurora QL SMSQ/E, still  ))  


Reply all
Reply to author
Forward
0 new messages