Hello everyone,

recently I took a closer look at the BIOS chips of a Diamond SpeedSTAR 24 and encountered some oddities regarding their content. Since no conclusive information was to be found on the web (apart from a thread in a german forum where a user made similar observations), I decided to research a bit further. Apologies if all of this is already known, but if not, it might be useful to someone:

The SpeedSTAR 24 has two BIOS chips labeled "DBIOS-L" and "DBIOS-H". Both chips are M27C256B EPROMS with a capacity of 32 KiB each. This seems a bit odd at first, since the full BIOS itself is only 32 KiB large, i.e. there seems to be twice as much space as strictly necessary.

The next surprise awaits when dumping the contents of the chips: They do not seem to contain anything useful. The label suffixes ("H" and "L") seem to indicate "high" and "low" bytes, but even interleaving the data does not yield anything useable. Additionally, the "H" chip contains the same 16 KiB data block twice.

This is due to the fact that Diamond shuffled around most of the address and data lines while connecting the BIOS chips to the ISA bus, and the contents of the chips compensate for that. The chips' data outputs are connected directly to the bus, while the address lines are routed through two 74LS245 transcievers. The "L" chip's data outputs are connected the the lower 8 bits of the ISA bus data lines while the "H" chip is connected the the higher 8 bits in the following manner:

1DBIOS-L:                  DBIOS-H:
2
3EPROM  ISA Bus            EPROM  ISA Bus
4
5A0     Addr 03            A0     Addr 09
6A1     Addr 02            A1     Addr 05
7A2     Addr 12            A2     Addr 10
8A3     Addr 01            A3     Addr 04
9A4     Addr 11            A4     Addr 11
10A5     Addr 04            A5     Addr 01
11A6     Addr 10            A6     Addr 12
12A7     Addr 05            A7     Addr 02
13A8     Addr 08            A8     Addr 13
14A9     Addr 13            A9     Addr 08
15A10    Addr 00            A10    Addr 07
16A11    Addr 14            A11    Addr 06
17A12    Addr 09            A12    Addr 03
18A13    Addr 06            A13    Addr 14
19A14    Addr 07            A14    (not connected ot the address bus)
20Q0     Data 0             Q0     Data 15
21Q1     Data 1             Q1     Data 14
22Q2     Data 2             Q2     Data 13
23Q3     Data 7             Q3     Data 12
24Q4     Data 6             Q4     Data 11
25Q5     Data 5             Q5     Data 10
26Q6     Data 3             Q6     Data 09
27Q7     Data 4             Q7     Data 08

After unscrambling the chip contents using this information, it becomes clear that the DBIOS-L chip simply contains a copy of the full BIOS. The DBIOS-H chip carries additional copies of the bytes residing at odd addresses (which are repeated once within the chip for reasons unknown to me). Obviously, the card has the ability to feed the BIOS contents to the bus using either both chips in parallel or using only the DBIOS-L chip. (Might this be related to the "Turbo BIOS" setting?)

Note: The above might be valid for other models of Diamond VGA cards utilizing two BIOS chips. I currently don't have access to any other cards to check this, though.

In any case, I have attached the utility I used to process the dump files. It can either unscramble the contents of the DBIOS-L chip (the H chip is not needed since it only contains redundant data) ...

1DBIOS /u scrambled_low.bin unscrambled_bios.bin

... or produce images for both chips from a plain (unscrambled) BIOS file:

1DBIOS /s unscrambled_bios.bin scrambled_low.bin scrambled_high.bin