Re: MACH chip question

From: Magnus Kollberg <Magnus.Kollberg_at_emw.ericsson.se>
Date: Tue, 23 Jun 1998 08:47:14 +0200

Hi,

I managed to delete your other message about Centurbo 2....

> The AB040 take 9 cycle to acces the falcon mother board because the Mach
> have to rebuild a 32 bits word with 2 16 bit one so They need 2 board acces
> and each of them take 4 cycle plus one in the mach for concatenation. The
> seems also that interupt are filter by the match but i don't know exactly how!

You said that the Centurbo 2 accesed the motherboard in 4 cycles and that it should
be faster. The difference is not very big if you are not only reading/writing
16-bit words. A blocktransfer should be much the same.

Also you said that the Cent2 could handle a transfer at 14MB/s to ST-RAM. How
would that be possible when a normal Falcon can only do 5MB? It does also
differn depending on which operation you do.

Have you tried Nembench on the Cent2?

Another small thing, you said that they choosed the 030 mainly for compatibility
reasons. I don't realy see this as a "reason" as I have not got any problems
with compatibility, the problem is Fast-RAM and not the CPU. Atleast I have no
problems with compatibility, but I guess I'm rare.

Now, I would like to add that I think the Cent2 sounds like a very nice card
and I would probably have got one if I hadn't got my AB040. But now when the
possibility might turn up to add a 060 with L2 cache to my AB040 I wouldn't
switch it against anything! I sure hopes it will work. :-)))

> I don't think tha AB040 Mach PLDs alters boot instruction. If i remember
> well only tos 4.04 work with the AB040 so i thing that there was a
> modification in the begining of the rom so that AB040 mmu doesn't crash at
> boot (because i thing this is the problems). I will take a look at an old
> tos and make comparision between this old one and my 4.04 to chek what are
> the differences.

Have you looked at this?

> >As the AB040 hasn't got the fastest RAM access in the world, maybe it
> >would be possible to change these 3 PLDs and make it faster. If someone
> >just could get hold of those damned equations for the PLDs. It could be
> >a interresting project for someone who has got lots of spare time...
> >
> I've got Mach databook so will take time to look if there is a solution to
> read the internal code (may be not because of fuse protection)

It would be interresting. It would also be interresting to know what is
limiting the AB040 from using more than 64MB RAM. Maybe they didn't add logic
for decoding all address lines?
 
> >Also, do you happen to know anything about the expantion bus on the
> >AB040? Is it just a plain CPU bus from the 040? Pinning?
> I'm doing it and haven't finich it yet. This expantion bus have the upper
> part of adress bus, and the lower part of data bus, There is some signals
> that comme from one of the Mach that i haven't decode yet. In order to have
> a real 32 bits bus we've got to use both trhe falcon expansioin port anf
> the AB040 expansion bus. As soon as i've finish this work i will release
> the pinout in this mailing list. I' could do this work but i haven't
> enougth free time to buil any hardware for it.

Great. Is this how the Cent2 works as well, "doubling" the old expansion bus?

If you need an idea for a great project, it would be very nice with a PCI
interface connected to this 32-bit bus.... ;-)

//Magnus Kollberg
Received on ti. juni 23 1998 - 20:40:00 CEST