24-bit graphics on Wombat's DAFB?

[eharmon]

Picking up from this thread: https://68kmla.org/bb/index.php?threads/supermac-pdq-spectrum-8-24-thunder-roms.42656/post-505422

The video DAC is the main difference. AC842 does 1/2/4/8/24 bit, and AC843 does 16 bit but not 24. That is auto-detected by the driver as well, so if you had a spare DAC from a battery bombed Q700 or 900, you could theoretically drop it into a Wombat (and extend the VRAM to 2 MB) and have 24 bit (the pinouts are the same).

Aaaaah. I didn't realize they switched RAMDACs. Looking at the reference though (https://bitsavers.org/pdf/apple/mac/video/DAFBII_ACDC.pdf):

The AC843 is designed specifically tor Apple graphics needs of the future. The architecture
. . . . /ena5Te~rthe._display o f u p t o 1 1 5 2 x ~7.0-blr-mappesl c o l o r g r a p h i c s ( u p t o 1 6 b i t s pe~.f)ix~I). Tbe·multiple pixel ports and interl)af mul~.ipfe-xmg enabJ.es TTL comp3:~ib~~..!!:l!edaciflg (u~to ,64 MHz)\ to the kame buffer, while maintaining.100 MAz video data ates at 1, 2, 4, 8, or
/16 bits per pixel. 24 bit chunky"planar rtlultiplexirig enables a 24 bit or··t2·bit~hunky pla~r
' frame bufier to maintain 64 ~Hz video data rates~. 24 bit mode enables a 32 bit frame b~ffer
(upper byte ignored) °'to maintain 64 MHz video data rates. Convolved video d~ta can ~ maintain a rate of 16 MHz.,' · .
The AC84~contains t~iple'256 x 8 colbr lookup tables with triple 8-bit video oJA convertJrs. On chip fe~tures inclu~e ,4.pple convoiution, 8 bit psuedo colo,r lookup or triple! 8 bit lookqp for 16 andi24 bit modes, dhunky, and\chunky planar multiple~ing, and built in logic block
observaticr-

And:

The AC843 supports the same MPU signal definition and timing as the AC842, but slightly different internal, control register addressing to accommodate the extended functionality. As illustrated in Table 7, the read and write functionality of the internal address register, the color look up table and the test register remain the same as defined in the AC842 spec.

Unfortunately OCR has done...poorly...in the first paragraph.

I'm a bit confused (and not real familiar with RAMDACs) -- it indicates it's designed for 16-bit graphics at 1152x870, but it refers to supporting 24-bit frame buffers.

So maybe a RAMDAC swap would Just Work™ (which is pretty neat), but the fact that it seems like you can send 24-bit graphics data into the AC843 is curious.

 

[Arbee]

The graphics system is very modular. Each DAFB generation is connected to a different RAMDAC and a different clock synthesizer, for instance. If they'd made a RAMDAC that could do 16 and 24 bpp, it would've worked without changing DAFB itself, in theory. djMEMC seems to integrate a fully intact DAFB II, since the Q800 schematics even show the additional VRAM lines for banks 2 & 3 just sitting unconnected. (DAFB VRAM is up to 4 512K banks).

 

[Arbee]

Here, I'll include some extra teaching aids

 

[Phipli]

Centris & Quadra 800 Video Capabilities

There is a chance that it is more than just the RAMDAC to be honest. If you swapped the RAMDAC, can the hardware video subsystem actually provide the 24bit data?

 

[Melkhior]

Here, I'll include some extra teaching aids

Interesting documents, but I think they tell a slightly different story than "AC843 doesn't support 24 bits"..

The "product description" paragraphs are enlightening.

• The bitsavers DAFB document (p114) talk about the 'AC843' (343S1075-01)
• The '357S001X-A_AC842.pdf' (p16) talks about the 'AC842' (357S001X-A) [X is 0, 1, or 2]
• The '343S1091_AC843.pdf' (p12) talks about the 'AC843 Stripped' (342X1091-A)

For the first paragraph of the product description:

• If you compare the AC842 and AC843, they have pretty much the same description including 24-bits support. The difference is that AC842 supports pixel clock up to 125 MHz (vs. 100 MHz), enabling 1280x1024 (instead of just 1152x870) but the AC842 doesn't list 16-bits per pixels.
• If you compare the AC843 and AC843 Stripped, the later keeps the 100 MHz maximum, but loose both 16 and 24-bits and has a diminished framebuffer interface frequency (33 MHz vs. 64 MHz). Presumably the much lower frequency from the FB controller precludes the higher depth available on the non-Stripped siblings. It also loose 'convolved video'.

The second paragraph is about the CLUT proper:

• f you compare the AC842 and AC843, they have pretty much the same description. The AC843 mentions 16-bits mode, though in the triple 256x8 DAC it's trivial to support in the source (the apple way is to substitute '5bABCDE by '8bABCDEABC with A to E being bits, with a linear CLUT)
• The AC843 Stripped as a 256x24 DAC, which is functionally effectively the same even if programming is a bit different. 16-bits is mentioned there despite not being in the first paragraph. 24-bits isn't.

So there's no variant of the AC84x (as listed here) that will do everything - you need the higher-clocked AC842 (there's 3 speed grade listed, X is 2/0/1 for 64/100/125 MHz) to get 1280x1024, but you need one of the AC843 for 16-bits (from the 2nd paragraph and other part of the document, the Stripped likely supports 16-bits as well) and the AC483 (not Sripped) for both 16 & 24 bits.

As their programming interfaces are (I think, didn't double-check the registers) all different, replacing one with another isn't trivial. And then whatever is pushing pixels from the VRAM to the DAC will need to handle the desired resolution and depth, which may not be easy.

What surprises me in the 'Stripped' variant is that it's likely a cost-cutting measure. Doing a new chip with most of the functionalities was cheaper than just producing higher volume of the full one? Those were different times...

 

[Arbee]

Good points all around.

What surprises me in the 'Stripped' variant is that it's likely a cost-cutting measure. Doing a new chip with most of the functionalities was cheaper than just producing higher volume of the full one? Those were different times...

That and 357xxxx part numbers are for custom-marked off the shelf parts (AC842 is almost certainly a Brooktree part). So doing a new chip was at least thought to be cheaper than just buying them, which seems unlikely when you include design and startup costs. (Per unit production may well have been cheaper). Maybe that’s part of why Apple almost went bankrupt a few years later.

 

[Melkhior]

(AC842 is almost certainly a Brooktree part)

I agree the SW interface makes it look like a BT device. The problem is that none of the BT devices I've ever seen (a datasheet of) have a similar pixel input interface.

All the early eras BT have a fixed interleaving, usually 1:1 (so one pixel pet clock) and sometimes 4:1 or 5:1 (they have 4/5 input to read 4/5 pixels per clock). By '93, they have a couple with a 'flexible' input, but those are large, specialized devices (the 445 and 496, the later is CMYK not RGB) that don't otherwise fit the bill.

The AC842 (and AC843) has a 32-bits pixel bus that can be used to input a varied number of pixels depending on the depth (p15 of the AC842).

My guess is, Apple was used to the BT SW interface, and specified (perhaps designed?) the device (see the note on p1) based on the BT SW interface, which was already cloned at that point. Then the devices were made to Apple"s specifications by other companies, probably including Brooktree as they might have helped with the design as well. So while the AC84x look a lot like Brooktree devices, I think they were never actually catalogued and sold by Brooktree themselves.

 

[dougg3]

(AC842 is almost certainly a Brooktree part)

Very cool documents! This is not my picture of a Quadra 700 logic board, but it looks like you're right.

Random side note irrelevant to the purpose of this thread: all of these documents mention sync on green, but at least in the case of my LC 475, it doesn't seem to actually work with my monitor. I'll probably do more testing later and talk about it in BMOW's VGA adapter thread.

 

[Arbee]

Nice, dougg3! The revision history on the first page of the 842 also says "RECORD CHANGE. BT 8-3-90". Apple used several specified/co-designed parts in that era: at least the AMD Combo and Curio chips, and the NEC NewAge FDC in the Quadra AVs. The IIfx SCSI DMA is a customized NCR 5380, but it's unclear if Apple did the design on that or it was specified. F108 contains a possibly customized NCR53C9x, and MESH is a much more heavily customized NCR53C9x.

 

[Phipli]

Very cool documents! This is not my picture of a Quadra 700 logic board, but it looks like you're right.

Random side note irrelevant to the purpose of this thread: all of these documents mention sync on green, but at least in the case of my LC 475, it doesn't seem to actually work with my monitor. I'll probably do more testing later and talk about it in BMOW's VGA adapter thread.

Note it is Bt manufactured, but copyright Apple.

 

[eharmon]

Here, I'll include some extra teaching aids

Doing some "light" reading tonight and WOW those are handy!

The graphics system is very modular. Each DAFB generation is connected to a different RAMDAC and a different clock synthesizer, for instance. If they'd made a RAMDAC that could do 16 and 24 bpp, it would've worked without changing DAFB itself, in theory. djMEMC seems to integrate a fully intact DAFB II, since the Q800 schematics even show the additional VRAM lines for banks 2 & 3 just sitting unconnected. (DAFB VRAM is up to 4 512K banks).

Wait, Q800 schematics?

 

[Phipli]

Doing some "light" reading tonight and WOW those are handy!

Wait, Q800 schematics?