Date   : Fri, 28 Jun 1996 12:19:35
From   : pnt103@...
Subject: Re: Hardware problems

Mike wrote:

> However, the ram chips themselves are usually very reliable.

I certainly agree with this; out of literally hundreds of Beebs I've fixed,
only a small percentage had RAM faults (cf Sinclair machines, where RAM
failures account for about 30% of faults).  However, the only two truly
pattern-sensitive faults I've ever seen were on Beebs :-)

> Try changing ic14, 74LS245, to the right of the 6502. This is an 8 bit
> octal latch which buffers the data bus to the memory. It's heavily
> loaded and is a common failure point.

Not *that* common, in my experience (it's a bi-directional buffer, not a
latch).  However, it *is* heavily loaded, and you sometimes get timing 
errors, especially with 2nd Processors attached.  The standard Acorn-
approved fix in this case is to replace the LS245 with an ALS245 or F245 
-- but be warned that in a very few cases this itself can lead to obscure 
timing problems.

> Also make sure the heatsink on the video processor is securely mounted,
> and that the power leads which connect to the PCB in the area of the
> DRAM are clean and tightly connected.

Nitpick: video processors (VTi chips) don't use heatsinks, only the original
Ferranti ULAs.

But in machines that have been "enthusiast-owned" and have had things added
it's very common to find the power leads get a bit loose.  As Mike indicates,
this is A Bad Thing, especially on Issue 2 and 3 boards.

> Troubleshooting the BBC's memory is not straightforward, as there are
> two CAS lines for each bank of 16k

Actually, one RAS and one CAS...

>                                    (the Beeb was originally produced as
> a 16k Model A machine), there are three 81LS95 octal buffers which

Three pairs of 81LS95s, actually, one for rows/one for column addresses in
each pair...

> select high resolution (bitmapped) graphics and teletext modes, and
> there are a lot of ram chips to deal with.

It's not that hard; it's a pretty standard circuit.  A useful trick is to
swap link S25, then reset the machine.  This puts it in the same 16K
configuration as a Model A; removing S25 altogether and resetting, puts it
in 16K mode but using the other bank of DRAMs.

Having written all that, it is still the case that most "memory" faults on
Beebs are actually due to bus overloading, or timing faults elsewhere, and
to have a DRAM or 81LS95 fail is relatively rare.

Pete