Amiga PC Bridgeboards...

  • Thread starter Thread starter furan-
  • Start date Start date
  • Replies Replies 73
  • Views Views 9512
I've been looking to Japan for more snap-on accelerators. Note the topline part below is just a sort of demo of their dummy chips, and their chip carriers that snap on to the BQFP(winged) 386sx.
http://sandy.blue.coocan.jp/Interposer/386_upgrade.html

Of some humor, is this guy who soldered down an IBM 3.6v chip and bent all of the power pins up, to attach to a 3.6v regulator.
http://kenji.ram.ne.jp/pc/nse_ibm.html

I learned a few random things about the 386sx form factor - I had been wondering why it has those wings - this is a Bumpered Quad Flat Package. The "bumpers" are so the chip can be protected while inside the plastic carrier. The chips have a weird pitch that hardly anything else has - 0.635mm. This was done to be exactly 25 mils when placing pads on a PCB - to make it easier.

On the 286 side:
I now have 2 different kinds of 286 upgrades on the way that I got cheap (one from Yahoo Japan), for my a2286, as well as 2.54mm header risers so it will fit. One is the KINGSTON SX/20LCM upgrade - this includes onboard cache, but a slow cpu. I'm thinking about swapping the cpu out for a Yamaichi BQFP100 socket (also on the way), the other is the IO Data PK-VF486 / 87SW2. These both have slower chips but they have an onboard oscillator and cpu there, so upgrading them is not a big deal (may have to disassemble the software that comes with, though).

According to the service manual for the a2286, the ISA bus is fully exposed at the bottom and the amiga-side (where the translator chips are) is exposed at the top. It is conceivable that a different sort of daughterboard could be fitted that just puts the amiga translators on the ISA bus. The only drawback I know so far, from disassembling the rom payload that the software sends to the card, is that the memory map is checked out and custom I/O are used with the controller chip (which would need to be removed for this to work). This could allow a fully loaded PICMG SBC to take over (consuming an extra slot), assuming it holes out all the right I/O, and can run the amiga payload as a boot rom.

There's also the possibility of upgrading it from 1mb to ~16 as the controller supports it and enough I/O lines are wired through to the daughterboard - a sim saver/sim carrier could be wired in.

Another possibility is that, with the whole ISA bus being brought to the daughterboard, the jumper pins on the bottom of the board could be removed and replaced with pins with longer legs, so that a second daughterboard could be stacked on top, which converts things to PC/104 (where a CL VGA and ESS board could be stacked on top, leaving the ISA slots free).

Photos of the A2286AT for reference: http://amiga.resource.cx/exp/a2286at

Really I just want to make my own bridgeboard that can speak to the original software but support VGA on both sides over zorro III.

Also picked up a SpeedStar64 after reading up on it. Searching for FTUISA5434A brought me to cheaper prices.
 
Last edited:
Very interesting plans!!
I hope a few of them work out. :) I'm having a lot of trouble finding a snap on for the 386 - I know a few companies made them but I can't find them in single quantities for under $1000.... so I may just focus on the 286.

Edit: I tried to go one step faster with the Diamond Speedstar 64 by buying faster RAM. Here's a table from the tech ref (http://www.vgamuseum.info/index.php/cpu/item/download/1330_a6d67b2ab41f1663f52232a86798dc90):

Q8fKDGd.png]


As you can see, the 5436 adds EDO support but drops ISA support. The 5434 is still heckin fast.

My Speedstar 64 will have 70ns DRAM - it looks like I can't go any faster on the 5434 without EDO support (and at some point we're limited by ISA bus bandwidth, but I haven't done the math).

gscp3K6.png


I have an ARK1000VL VL-Bus card that has 45ns interleaved RAM. Shame I couldn't use that.
 
Last edited:
Amiga PC Bridgeboards...

It’s probably premature but I ordered an ET4000 W32i - with 2 mega installed. It will interleave access between the first and second meg of vram, and the chips are 45ns. It’s fast as a GUI accelerator but whether it helps with DOS non-VBE accelerated modes, is an open question. It would be fun to compile our own performance list but I don’t currently have matching bridgeboard hardware - I would need to upgrade one of my A2386SX boards in the same way (and even then since you have a golden gate it might not be a fair comparison).

I’m about halfway to figuring out how, through disassembly, to enable all the debugging error strings in the bios and in the a2386sx.bios payload as video text mode output (that is the path the code takes for them when enabled) - still digging into it. If I can sort this out it may be possible to diagnose these cards more easily in cases where a battery leak isn’t outright shorting something out but something about the configuration just “isn’t quite right.”

One more thing -specific to the A2386sx: some time ago I found a utility that reprograms the clock divider on the SCAMP (VLSI 82C311) controller to overclock the CPU - I can't find that utility anymore, but I thought the SCAMP controller documentation was really interesting. I've uploaded the datasheet to my dropbox: https://www.dropbox.com/s/55536rttki75d0b/VLSI82C311.pdf?dl=0

Of interest:
1) The SCAMP clock generator is documented as increasing the clock when video DRAM A0000-BFFFF is accessed.
2) The CPU clock is generated as /2 from that 32/40/50mhz oscillator on the board
3) Bus clock is generated from the 16mhz chip.

The divider for the CPU can be modified to attempt an overclock, or to simply have a 1:1 with the CPU clock oscillator and socket that. It looks like the bus clock can be modified as well. I wish I could find the Amiga utility that did the first part. But this looks like some simple register programming. So we could set a different base clock (higher or lower), set the CPU internal multiplier, maybe to get a slightly higher overclock. And modify the bus clock to get a little more bandwidth.

The interesting stuff starts at page 43.

8kF4fVJ.png


I am on vacation and a little obsessed with this project, sorry for all the spam.
 
Last edited:
I'm on vacation out of town, so I haven't been able to use the upgrade board on my a2286at yet... Now have a VIPER Jet (Japan) on the way that is much like the Cyrix snap-on upgrade, just need to swap out the CPU on top for a socket.
 
You will get the pictures soon mate!

And very good job, thanks for sharing the informations!! :)
 
I've acquired what I'll call a "reference platform" for the upgrade I want to do on the A2386SX, and to try out various settings (apparently the mem read/write in graphics range is set by a register that the A2386SX bios might not set). It's just a motherboard with the same SCAMP controller - VLSI82c311. Also received ET4000W32/i (apparently no faster in DOS despite interleave), and Diamond Speedstar 64. Also a CLGD5429 w/50ns ram.

I've been out of town for some time but when I catch up I'll post something. :)
 
I’ll be interested to know if the Speedstar 64 works. Mine arrived yesterday (sold as working) and doesn’t appear to work with the GoldenGate. I get no startup beeps from the bridgeboard and just a black screen on the monitor. Can’t enter the gg setup program either. There are two closed jumpers on the board (7 and 8 ) which I haven’t tried opening yet as I haven’t had time so far. Going back to the ati Mach64 and all works ok.

i think one jumper handles zero wait states and the other irq’s, but references online do not seem to match what I’m seeing in my board.
 
Last edited:
Man I'm on a roll...

So I tested out the 286 bridgeboard I got, works fine. Jammed the kingston CPU upgrade into the 286 socket, between the board and the daughterboard (carefully), worked fine.

So now I'm adding a riser - I need to add bypass caps still, but this allows the upgrade to fit, the daughterboard to expand into the zorro-only side of the board, and allows using zorro cards behind both "expanded" slots as there's space.

The upgrade board is a 386-20 - I'll be socketing the onboard clock and adding a Yaimachi IC149-100-028-B5 to socket the upgrade CPU.

Here's what the socket looks like (on order, will take a month or so apparently):
x5yl4me.png


Here's the board in its current state - still making things happy again...
JhteH2M.png


Another dream when I get it back up and running is to find a way to have a PC-104 card (video card, sound card, whatever) mount in between the daughterboard and the card.
 
I've somehow broken the a2286AT by adding these headers. I cut them down thinking it was noise, to a level that would just fit the upgrade, but no dice, even with the old CPU. Still debugging :(
 
Issue was that my removal of the upper connector delaminates the board so I’m having to run some jumper wires. Had I thought of it at the time I would have ordered female idc headers with extra long legs, not removed the existing male headers for new ones!!!

Note to self: these are antiques!

Edit: board fixed and upgrade card works (but the upgrade card is currently a 386sx/20)


Sent from my iPhone using Tapatalk
 
Last edited:
Well, I know I'm alone on this thread lately, but I figured I'd post a milestone. I upgraded a 286 to 386 upgrade board that was for a 386sx/20, to a 3.6v, 5v IO tolerant IBM 486SLC2/50 (50G6950). It's all hacky and I have a 50mhz crystal that will fit on order, instead of the one deadbugged in the PLCC mathco socket, and I haven't booted an OS past BIOS, or enabled both caches and tested yet, but I think it's gonna be awesome. :lol:

zdf40cV.png

iIb4mJr.png
 
Last edited:
You are not alone! I am reading with interest. And wondering if similar could be done to a GoldenGate?
 
Yay, glad you're still here

I lucked out with the Kingston upgrade board for 286 -> 386, and that the 286 board I bought actually works. I'm not sure how much benefit the "l2" cache will provide, and if having that cache and the onboard cache will lead to coherency issues, but will see.

One thing I noticed about these cirrus logic cards is they have their own crystal matching the isa bus speed - you might try removing yours (such that you can put it back) and see how it boots. I'll try and dig more to see what that crystal is used for. I am having problems with cirrus logic(specifically) card stability as well, and I think it might be because the ISA bus doesn't have good termination (at least on my board).

The 286 to 386/20 upgrade boards (which have cache) are listed here. They bench close to a 386/20. Cost around $60.
https://www.ebay.com/itm/KINGSTON-S...ures-ASIS-Because-I-cant-Re-Test/173715214963

I should also mention - I'm not booting my Amiga to workbench - I have all disks disconnected and am using the test pins on the card (shorting 5v to pwrgood on the test header) to take the bridgeboard out of reset.
 
Last edited:
On video card stability - with this 286 board, the Speedstar 64 sometimes hangs at boot, as does my Mach64, as does all of the cirrus logic boards in general, that I have. I started looking into this:
http://www.os2museum.com/wp/the-isa-osc-mystery/

I measured the clock speed on my bus as 6mhz, then 8mhz after a bit, vs the 14.318mhz crystal on the board. If you have a scope you might try measuring as well. I have a 14.3181mhz oscillator on my bridgeboard, so I will replace the bumpers (capacitors) near the oscillator, and remeasure at the oscillator. I suspect the chipset ultimately controls the bus speed (as does the a2386sx).
 
I replaced the bus crystal and bumpers/resistors on my board (which had a bus and dma output, of which I had already selected to use the bus clock) with a 14.318mhz oscillator and now get bang on 14.318mhz shortly after the bus comes up. It first starts at 8mhz, and then switches over, and I suspect that's based off the oscillator for the 286 / 2. My a2386sx doesn't have this problem. When it powers up, the bus immediately gets the right speed. Maybe it's the change in speed on the bus?

More interesting stuff - the Diamond Speedstar64 has a pin on the fingers for the bus clock, but not osc (14.318mhz). I ended up getting it to work rock solid by disabling 0 waitstate, even though 0ws on my et4000w32i works fine.
 
Last edited:
Today's gets:

(A2386SX work)

OlPbTa2.png


VIPER-JET 386SX to 486SLC upgrade from Japan. Snaps on top of the 386sx. Jumper switches between the onboard CPU and the upgrade.

I am considering putting my "fast" IBM 486slc2 that does 66mhz on it, and messing with the multiplier MSR to see if I can get 75mhz based on 25mhz clock. Given the cache, and that zero wait state would work, this might end up making an A2386SX the fastest it can get (...for now). I'll have to move the series resistor and bypass caps around the CPU to the other side of the board to make it fit. The pins on the IBM CPU I have are fragile from resoldering/desoldering, so I am not sure it will make it. If not I will try to source another one....

(A2286AT work)

Bhg0GE6.png

NC-830big.jpg

I also got this oddball pc/104 VGA board with an assortment of PC/104 boards I bought for cheap. It's an Advance Logic ALG2032+ (has UNIVBE drivers), with 2 35NS EDO DRAM that has a 64 bit memory bus (from board examination it looks like 32). With some effort, I can remove the headers and sandwich it between the A2286AT daughterboard and the board inline with the 8 bit bus riser pins (and then jumper the 16 bit wires). It also has a spot that looks like it's for a ROM, but it's actually for a disk on chip up to 144MB. Once that's done, all there is is memory, which is going to be a pain because of the way the banks are set up - the chipset supports 16mb, but is undocumented. Folks don't seem to like this chip, but all I care about is DOS - maybe it's really fast at DOS. It has good UNIVBE support.

When I'm done the A2386SX will probably go in my A4000T and the a2286AT will go in my A3000T.
 
Last edited:
Back
Top Bottom