Raven. A homemade Atari-like computer

[agranlund]

@PhilC , I received my 2MB cirrus and it scrolls through the entire 2mb vram as it should in the test program.
Perhaps there is something wired up wrong on your card making it scan out the first 1mb region twice (always accessing your first, orginal, bank of ram perhaps?)


And since it was relatively "cheap", at least compared to the crazy ebay prices, I can confirm that GD5434 is really quite fast.

Both Cirrus cards are running at default speeds set by the bios.
I haven't checked how much the '34 can be overclocked. My '26 tolerates a little overclock but nowhere near what you did with your '29.

All the cards are bus-bound on cpu->gpu writes so I picked three Gembench results which are more or less helped by hardware acceleration:

(all three are _very_ snappy compared to unaccelerated cards, you can't go wrong with either of them, or the WD90C31 which was left out of this test)

Code: Select all

Dialog     640  800 1024
W32i      2034 2034 2015
GD5434    2960 2919 2823
GD5426    2481 2332 2167

Code: Select all

Blitting   640  800 1024
W32i      4441 4347 4256
GD5434    5675 5238 4441
GD5426    3295 2670 2232

Code: Select all

Scroll     640  800 1024
W32i      3227 3095 3169
GD5434    3780 3600 3396
GD5426    2633 2242 1946

The fillrate is very good.
Though I suspect it would end up in a tie with GD5426-28 for game sprites since it can't do single-pass colorkey transparency like those two older models.
And the W32i would blow them both out of the water in that scenario (assuming the W32i can do colorkey transparent blits.. I haven't actually looked, I'm just assume it can based on how well Starcraft runs on PC with an ISA W32i).

GD5426 got 40Mb/s fill, 20Mb/s blit.
WD90C31 got 75Mb/s fill, 17Mb/s blit.

IMG_5145.jpg (73.93 KiB) Viewed 328 times

[Atarian Computing]

Yeah, I've reported this issue years ago but it never got fixed. This is not a Raven specific problem. I've replaced the problematic files from the original installation.

I think I found the problem.
These .SLB files needs to be post-processed by a tool (which is included with the MagiC sources) after compilation and it looks like the ones in the open-source snapshots aren't. I found and resurrected and old thread on atari forum where the slb issue was discussed.

Excellent work. Thanks for looking into it.

[PhilC]

@agranlund the card reports having 2mb in the test program but loops back as you've seen on the memory test. I've got an identical card and a speedstar pro 5429 1mb. I'm just waiting for the ram to arrive to upgrade them. I think the Diamond card should work right away as it's got a couple of resistors you need to alter to change ram types etc which I've already changed. Once that's working I'll fix the other two with the new info I'll have.

Where did you get the 5434? I've not been able to locate one at anything other than silly money.

[agranlund]

small release package update:
https://github.com/agranlund/raven/rele ... .A1.latest

Code: Select all

20250914:
  rvbios:   prevent showing boot screen twice under MagiC
  software: added 3rd party fpupatch
  software: added 3rd party xboot

tools/xboot32e : patched 3.20 in english and working on 060.
Can be a hassle finding out which version to use so I included this one.

tools/fpupatch : utility for patching out SFP-004 detection from programs that were compiled with Pure-C and softfloat (PCFLTLIB.LIB)
Raven is unable to catch and bus-error on attempts to hardware-detect this old addon card, and if a program then attempts to use it chances are very high the machine will hang. This program can patch such programs so they won't attempt to detect or use an SFP-004.

I don't really think there's that many programs where this is an issue but Raven rev.A2 will have the necessary hardware change to prevent incorrectly detecting SFP-004. rev.A1 doesn't have the necessary address signal going to the cpld but can probably be sorted with a patch wire if one wanted to (to be tested)

The only program I've personally found troublesome, and had a use for that patcher program, was Pure-C itself as it would hang when compiling code containing floating point. But then again, I haven't tested tons and tons of programs.

[PhilC]

I've got the extra memory for my second Siig 5429 but thought I'd test if this card runs with the same over locking as the previous one and so far so good.

So for me it's looking like I'm going to stay with a 5429 unless I can find a 5436 for peanuts.

So on the test program, the filltate option crashed on the 2mb card but with the 1mb overclocked I get some amusing results.

Here's filltate on the OC 1mb

17579437589972897939550706993655.jpg (376 KiB) Viewed 207 times

And for an evening more amusing results, at 640x480 it's nearly as fast as the standard 5434 same results for all resolutions.

17579439154831068219156069779159.jpg (356.74 KiB) Viewed 207 times

I'll probably be the only person running this overclocked card of all the Raven owners but it'll at least add to my fun.

Incidentally @agranlund I've been checking the 2mb upgrade, schematics etc and the card appears to be set up correctly, so I'm wondering if it's driver related as there are quite a few specifics in the extension registers for just the 5429. I'll have to see if I can have a mess around with the drivers, might need a specific config for this card.

[agranlund]

Incidentally @agranlund I've been checking the 2mb upgrade, schematics etc and the card appears to be set up correctly, so I'm wondering if it's driver related as there are quite a few specifics in the extension registers for just the 5429. I'll have to see if I can have a mess around with the drivers, might need a specific config for this card.

The driver doesn't do very much direct register manipulation (apart from the blitter stuff).

Initialisation of the card is done by the cards bios, and so is setting video modes.
It does switch to linear framebuffer instead of the default 64kb banked stuff using register manipulation after the vgabios has switched gfxmode so that would be where I'd look if looking at the software.

Alternatively, if possibly there are some software registers that needs setting for 2MB to work and the bios on your card doesn't because the manufacturer never meant to ship a 2mb version (seems incredibly far-fetched, I'd imagine most of the bios is whatever reference bios they received from cirrus plus some minor oem tweaks)

Have you checked how the hardware configuration settings are wired and cross-referenced with the different DRAM configurations shown in Appendix B7 ?

Some configuration options are not changeble by software, the chip reads them from the data bus during reset and there would be pullup/down resistors on the board that sets these. The interesting ones appears to be:

Code: Select all

CF-13 (MD29) (pin 99)  0 = Asymmetric DRAM,  1 = Symmetric DRAM
CF-12 (MD28) (pin 100) 0 = CAS3:0, single-WE 1 = WE3:0, single-CAS
CF-11 (MD27) (pin 102) 0 = Extended RAS,     1 = standard RAS

Special case for CL-GD5429:
CF13 CF12
   0    0  OE pin becomes RAS1 for 2-Mbyte memory option

[PhilC]

@agranlund yes I spent a while checking the card against B7. It's the symmetric 256x16 x 4 chip setup. I need to try and get a scope on it and see what it does but the lines I checked seemed set up correctly, although I didn't do all of them yet. The Diamond card is set up like the last config in B7

I've got some other bios versions to try but would love to try and disassemble the current ones so I know exactly what it's doing.

[agranlund]

I'll probably be the only person running this overclocked card of all the Raven owners but it'll at least add to my fun.

I find it amazing that your card is able to go that high! That's on par with the 64bit GD5434.

Now I had to test what it can do and mine apparently goes to $2B, beyond that it'll start vomiting pixels all over the screen

GD5434_2B.jpg (95.07 KiB) Viewed 167 times

[PhilC]

Nope, changed my mind, I need a 5434

What speed ram does it have @agranlund ?

[agranlund]

@PhilC , these are the ram chips.

IMG_5147.jpg (67.57 KiB) Viewed 161 times

The SEC chip would be 60ns

That other one is labeled 40 like the RAS time but min cycle time says 75ns:
They're socketed so perhaps I should try and locate a 60ns like those soldered ones.

Screenshot 2025-09-15 at 23.39.13.png (218.97 KiB) Viewed 161 times