x86 card revival thread

[micheledipaola]

thanks @evanboonie for all your work. The cache mod looks really promising and interesting, and the component list you shared is less than 20€ from digikey, so definitely worth a shot. I am totally scared by the soldering work, so I should probably ask for some help by some skilled friend, but I am putting this mod on my to-do list for sure... looking forward to the PCMCIA one too on my... OrangePC 2 (let's start using the correct naming you provided) but... I leave you to go on and explore it before

 

[evanboonie]

I finally got around to testing a 3.3v AMD Am5x86 in my OrangePC 2 and it works! Well, mostly. First off, it just refuses to POST when set to the 3x multiplier. I believe that the issue isn't so much with the multiplier itself, but the fact that the bodge wire enables write back cache when in 3x mode. I didn't feel like moving the wire around to confirm since it works in 4x though. There are also some BIOS issues with the 5x86: The system will only ever report it as a 75MHz 486 regardless of the bus speed. HWiNFO, System Information 8.0, and TopBench 3.8 also believe that it is a 75MHz 486. However, Speedsys 4.78 does actually report it as an Am5x86 running at the correct clock speed. I even got a successful Speedsys run with a 40MHz bus and the CPU running at 160MHz with accurate looking benchmark results. I haven't gotten around to installing Windows on it yet, so I'll report back when I get that up and running stable.

I also received the other CL-PD6710 PCMCIA chip that I ordered from China and this one checked out okay on the multimeter (good VCC+GND pin locations). This one was a "Basis Communications" part instead of Cirrus Logic, so if you go looking for one, that is my recommendation at this time. The Basis part is also the one installed on the PCMCIA PC/104 board I have which I was planning on harvesting as a backup option. After soldering the new CL-PD6710 on to my OrangePC 2, it still boots. I don't have any software on it to test the slot functionality yet, but I hope to install Windows 95 on it this weekend and will give it a test then.

 

[evanboonie]

I have finished my journey with the Orange PC 2 hardware upgrades and everything is working great! For those just tuning in, my card started life as a "220" with only serial/parallel. I added L2 cache and PCMCIA to give it all of the features of a "290". I even got write back cache working on my Am5x86 at 4x by moving the bodge wire to always enable it. Running at 160MHz with WB enabled, I scored a 60 on Speedsys. I also got Windows 95 installed and running stable on this configuration. PCMCIA works within Windows 95 as well! Below are some photos of my modded card. I put an oversized heatsink on the 3.3v regulator as it does get quite hot. I had to cut a fin off of the heatsink to make it fit. I also trimmed back the PCMCIA cage to make room for socketing the Decl ROM after I dumped it.




Eventually I would like to try hacking the BIOS on this card to tighen up timings and see if I can squeeze even more performance out of it. I'm also waiting patiently for the PCMCIA PicoGUS to give it sound. I'm currently moving to a new house now though, so this will have to wait for future months to complete.

 

[evanboonie]

I've also finished modding my original Orange PC. The clock doubling and 8k of cache works perfectly on the TI486SXLC2 I installed. I tried overclocking it, but the OrangePC application reports errors on start even when I run it at 30MHz, so this card is basically not over-clockable. The 25MHz (50MHz clock doubled) speed and additional cache still result in almost 2x performance vs the Cx486SLC that came installed in the card. Speedsys locks up with this configuration, but SI8 and Topbench report scores between a 386DX-40 and 486DX-33. I also used an ISA riser card to install a low profile sound card which I wired up to my Quadra's CD audio input and it also works perfectly! I really like the audio quality and the fact that line level audio is exposed on headers for the OPTi 82C931 card I'm using, but loading the drivers do make MS-DOS 6.22 take 5-10 minutes to boot... If anyone else tries this, I'd recommend using a different ISA riser than I used. The "riser" I used was designed for pin headers, not another ISA slot; so tedious bending of pins on the edge connector slots was required to make it fit. You can't see in the photo as it is below the sound card, but I also attached a heatsink to the CPU as it gets quite warm vs the original.

 

[NJRoadfan]

It shouldn't take 5-10 minutes to boot MS-DOS. Give UNISOUND a shot in configuring the PnP soundcard.

 

[evanboonie]

It shouldn't take 5-10 minutes to boot MS-DOS. Give UNISOUND a shot in configuring the PnP soundcard.

Thanks for the tip, I'll give that a shot. It booted in seconds before I installed the OPTi driver I found on Vogons.

 

[powermax]

I also took some better quality photos of these Orange386 and OrangePC 290 cards.
View attachment 45833

Is the above Orange386 card from the OrangePC 200 series?

I assume it's a Nubus card. The question is where the Nubus interface is implemented. I see a Xilinx FPGA (U44) and a large IC right above it (U67). Unfortunately, the sticker "90 ATC 01020" hides the chip markings. It would be great to know what kind of chip sits there.

Every Nubus card requires a DeclROM to be recognized by Mac OS. U82 looks like the DeclROM. Did you try to dump its content?

 

[joevt]

If the ROMs are working, then try SlotsDump or SlotsGrab from https://github.com/joevt/SlotsDump to dump them. It might be interesting to see if those dumps match an externally dumped version.

 

[evanboonie]

My previous post #137 on this thread has dumps from the original OrangePC (200.bin) and the OrangePC 2 (220.bin). When I say "Original OrangePC" I mean the one with a single 16 bit ISA slot and 386sx/486slc. The OrangePC 2 is the 200 series. I desoldered the DeclROM chips and put them in an EEPROM programmer to get those dumps.

 

[joevt]

My previous post #137 on this thread has dumps from the original OrangePC (200.bin) and the OrangePC 2 (220.bin). When I say "Original OrangePC" I mean the one with a single 16 bit ISA slot and 386sx/486slc. The OrangePC 2 is the 200 series. I desoldered the DeclROM chips and put them in an EEPROM programmer to get those dumps.

They just contain product info. No drivers or BIOS or anything. Everything else is zeros.

They both have:
Board:
OrangePC
Orange Micro, Inc.
1.0

200 has:
CPU:
CPU_80386_OrangeMicro_OrangePC

220 has:
CPU:
CPU_486_OrangeMicro_OrangePC

 

[Byrd]

I even got write back cache working on my Am5x86 at 4x by moving the bodge wire to always enable it. Running at 160MHz with WB enabled, I scored a 60 on Speedsys.

On your card and CPU did WB give a reported boost over WT cache? I've a OrangePC 290 card with an Intel DX4-100, unsure if the WB-enabled version but keen to try if worth the effort.

 

[evanboonie]

They just contain product info. No drivers or BIOS or anything. Everything else is zeros.

Yeah, everything else is placed in the resource fork of the OrangePC application. The application is where the BIOS lives and is passed into the card when you start it. What is in the DeclROM looks like just enough to get the mac to know that there is a NuBus card there so it can report it to the OrangePC app at startup.

 

[evanboonie]

On your card and CPU did WB give a reported boost over WT cache? I've a OrangePC 290 card with an Intel DX4-100, unsure if the WB-enabled version but keen to try if worth the effort.

Yes, it did provide an increase in benchmark scores. The speedsys CPU score increased by 5-10 points and the "move" memory test was significantly faster. A DX4 with WB should be enabled by default on a 290 if it is set to 3x multiplier and has the bodge wire on the back of the CPU socket. If the wire is missing, you can add one in the location it is placed in my photos to enable it for all CPUs that support it (not sure if this causes problems for CPUs that don't). If your 486 supports WB it will have an "&EW" etched into it, the WT models just have "&E".

 

[powermax]

My previous post #137 on this thread has dumps from the original OrangePC (200.bin) and the OrangePC 2 (220.bin). When I say "Original OrangePC" I mean the one with a single 16 bit ISA slot and 386sx/486slc. The OrangePC 2 is the 200 series. I desoldered the DeclROM chips and put them in an EEPROM programmer to get those dumps.

I wonder what function the Xilinx FPGA (U44) implements. Judging by its position it does something related to the Nubus interface.
The XC3000 datasheet tells us that XC3042 is a FPGA. That means that it needs a configuration program stored in a EEPROM/SROM to be operational.
Do you have any idea where its configuration is stored? It's often a specialized SROM in a small 8-pin package.

Anyway, it would be great to find that ROM and dump its content like it was done for a Digidesign Nubus card, see here

What kind of IC is U57? I can't see its markings because of the sticker on it.

 

[Bolle]

The Xilinx bitstreams are stored in the OrangePC application. It gets configured when the card is initialized when the software is launched.
The chip likely has something to do with putting video data out through Nubus as there are different bitstreams for different video modes.

 

[powermax]

The chip likely has something to do with putting video data out through Nubus as there are different bitstreams for different video modes.

Sounds interesting. You assume that FPGA provides a data path that pushes video data generated by WD90C30A to the Macintosh frame buffer over Nubus, right?

In the meanwhile, I started on hacking the OrangePC application in a specialized fork of the DingusPPC emulator.
I created a mockup for a Orange PC 200 card that exposes the DeclROM provided by @evanboonie here to the Slot Manager.
OrangePC sees my virtual card and attempts to communicate with it:



The appalication quits very soon because little HW is emulated now:

 

[powermax]

I discovered the following HW details so far:

At $FS040000, where "S" designates the Nubus slot the card is inserted into, seems to contain some kind of HW register.
The OrangePC application accesses it at the very beginning. It's read two times and written once with $106C. It must implement at least 14 bits.

At $FS060000, some memory aperture is located. The app uses this aperture to initialize some on-card memory. It writes 32K of zeros there, then verifies that the memory indeed contains zeroes.

The app then moves the content of 'LODR',0 resource to $FS064000. That resource supposedly contains some x86 code, presumably a loader. The size of that code is 8K.

I'll continue my investigations tonight.

It would be helpful to know how the Nubus interface is implemented in a 200 series card.
I especially interested in that board register at 0x40000 (see above).
Where is it implemented?

Then it would be great to know which memory is accessed using the aperture at 0x60000.

While it's possible to figure out a lot of details using SW hacking, I'd appreciate some HW hacking assistance.
With some luck, we'll maybe able to emulate that card.
It would become a really amazing project

 

[Bolle]

You assume that FPGA provides a data path that pushes video data generated by WD90C30A to the Macintosh frame buffer over Nubus, right?

Correct.

The other QFP100 chip obscured by the sticker is an Actel A1020. I think those are antifuse devices so it's holding its own configuration.
Then there's two GAL-like PEEL devices and an AMD MACH CPLD.

 

[powermax]

Correct.

The other QFP100 chip obscured by the sticker is an Actel A1020. I think those are antifuse devices so it's holding its own configuration.
Then there's two GAL-like PEEL devices and an AMD MACH CPLD.

I suppose there is no way to read out A1020 configuration without decapping the chip and doing some time and cost consuming IC RE.
Do you have a list of ICs used in the Orange PC 200 cards handy?
Everything I got so far are two high-resolution pictures of two boards.
The markings of some ICs aren't readable though.

 

[evanboonie]

My cards are packed away in boxes for moving right now. But if you can't get part numbers from others, I should be able to get them within a month or so.