[comp.sys.acorn] ARM3 upgrade

djhr@crosfield.co.uk (dave redman) (05/07/91)

Hello all,

anyone know anything about ARM3's ?

I bought a 30MHz ARM3 and fitted it to an A340..

It worked fine until I switched the ARM3 cache on and then...

The machine would run VERY fast for about 3 mins ( from cold ) and then all
hell breaks loose with modules dying left right and centre.

It was quite amusing watching the OS modules die one at a time with an
Address Exception error...

Anyway I think I traced the problem..

  I bought the Atomwide 4Meg memory upgrade and it has the MEMC1a fitted
  but it also uses a PEEL18CV8 -35 ( fitted between the CPU and MEMC sockets )
  Now if my maths serves this device can only run at 28.5 MHz ( max ).

  I have since modded the ARM3 motherboard to run at 25Mhz and everything works
  fine, however I WOULD like to go back to 30MHz and I propose to do this by
  replacing the -35 device with a -15.

  If anyone at acorn ( or otherwise ) would like to comment about this solution
  I would appreciate it. Or if anyone would like to post me the ABEL source for
  the PAL I would be even happier. ( I tried to read the device but it has the
  security fuse blown. - must be a BIG market for ripped off MEMC1a control
  PALs. )

I am curious abut something else though.. In my VLSI data book it says the ARM3
runs at 20Mhz.. does this mean a 30Mhz device is just a 20Mhz device that
happens to run at 30Mhz or is my Nov. 1990 VLSI data book already out of date ?

So the moral of this story..

    IF you buy a Watford Electronics 30Mhz ARM3 it may not work..

to be fair.. Watford DID tell me that in SOME machines the 30Mhz beasty doesnt
work.

Here are some performance figs I just ran..

	Machine		Dhyrstones per sec
	ARM2		5289
	ARM3 @ 25Mhz	14883
	SUN3/60		2888
	SUN/Spark1	8000
	SUN/Spark2	16000

And the SUN times were ONLY counting the executing procedures times not the
whole process time.

dave.

##############################################################################
# David Redman,                      #  JANET      :                         #
# Software/Hardware Engineer,        #  EARN/BITNET:                         #
# Crosfield Electronics Ltd,         #  UKNET      : djhr@cel.uucp           #
# Three Cherry Trees Lane,           #  Phone: 0442 230000  ext 3378         #
# Hemel Hempstead, Herts.            #                                       #
# United Kingdom                     #                                       #
##############################################################################

nbvs@cl.cam.ac.uk (Nicko van Someren) (05/08/91)

In article <9887@suns6.crosfield.co.uk> djhr@crosfield.co.uk (dave redman) writes:
>I bought a 30MHz ARM3 and fitted it to an A340..
>
>It worked fine until I switched the ARM3 cache on and then...
>
>The machine would run VERY fast for about 3 mins ( from cold ) and then all
>hell breaks loose with modules dying left right and centre.
>
...
>Anyway I think I traced the problem..
>
>  I bought the Atomwide 4Meg memory upgrade and it has the MEMC1a fitted
>  but it also uses a PEEL18CV8 -35 ( fitted between the CPU and MEMC sockets )
>  Now if my maths serves this device can only run at 28.5 MHz ( max ).

It is very unlikely that this is the problem as the PAL between the ARM and
the MEMC only needs to run at RAM speed which is 8MHz on everything except
540/R260s.

>  I have since modded the ARM3 motherboard to run at 25Mhz and everything works
>  fine, however I WOULD like to go back to 30MHz and I propose to do this by
>  replacing the -35 device with a -15.
...
>I am curious abut something else though.. In my VLSI data book it says the ARM3
>runs at 20Mhz.. does this mean a 30Mhz device is just a 20Mhz device that
>happens to run at 30Mhz or is my Nov. 1990 VLSI data book already out of date ?

Indeed you are right.  30MHz ARM3s are not rated as such by VLSI.  The first
few hundred ARM3s all seemed happy at higher speeds so firms like Watford 
decided that 30MHz was O.K..  Later VLSI changed the fabrication process to
increase the yield and at the same time reduced the top speed.  Since then
Aleph One have stopped rating their parts at 30MHz because a few were found
to be unreliable once they got hot. 

>So the moral of this story..
>
>    IF you buy a Watford Electronics 30Mhz ARM3 it may not work..
>
>to be fair.. Watford DID tell me that in SOME machines the 30Mhz beasty doesnt
>work.

I think that that statement is a bit misleading.  It should read 'Some 30MHz
ARM3 cards will not work at all when hot'.  Unless your machine is suffering
from bad power supply droop (not unknown on early 310s) it is very unlikly that
the rest of your machine will effect the speed of your ARM3.  The reason 
for this is that the high frequency signals are confined to the upgrade card 
and never get on to the main PCB.

If Watford sold you the card as a 30MHz unit, send it back, it clearly is not.

Nicko


+-----------------------------------------------------------------------------+
| Nicko van Someren, nbvs@cl.cam.ac.uk, (44) 223 358707 or (44) 860 498903    |
+-----------------------------------------------------------------------------+

cs9h9tts@cybaswan.UUCP (t.simpson) (05/08/91)

In article <9887@suns6.crosfield.co.uk> djhr@crosfield.co.uk (dave redman) writes:
>Hello all,
>
>anyone know anything about ARM3's ?
>
>happens to run at 30Mhz or is my Nov. 1990 VLSI data book already out of date ?

[stuff deleted]

>
>So the moral of this story..
>
>    IF you buy a Watford Electronics 30Mhz ARM3 it may not work..
>

from my experience : if you buy anything from watford it may not work


Tom.

klamer@mi.eltn.utwente.nl (Klamer Schutte) (05/10/91)

In <9887@suns6.crosfield.co.uk> djhr@crosfield.co.uk (dave redman) writes:

>Here are some performance figs I just ran..

>	Machine		Dhyrstones per sec
>	ARM2		5289
>	ARM3 @ 25Mhz	14883
>	SUN3/60		2888
>	SUN/Spark1	8000
>	SUN/Spark2	16000

>And the SUN times were ONLY counting the executing procedures times not the
>whole process time.

I did another benchmark. We took a program we had overhere, which took real
cpu time on our sun's (SS1 / SS1+). We compiled && ran it also on a 
R140 + ARM3 + RiscIX. And the sun (also running UNIX, just a different flavour)
cleared the job in 10% time of the archimedes.
This was a program with a very small amount of FP (approx 1% on the suns.).

Now i almost only use the archimedes as a B&W X terminal...

Klamer
-- 
Klamer Schutte
Faculty of electrical engineering -- University of Twente, The Netherlands
klamer@mi.eltn.utwente.nl	{backbone}!mcsun!mi.eltn.utwente.nl!klamer

ajdh@stl.stc.co.uk (Andrew J D Hurley) (05/16/91)

In the referenced article cs9h9tts@cybaswan.UUCP (t.simpson) writes:
>In article <9887@suns6.crosfield.co.uk> djhr@crosfield.co.uk (dave redman) writes:
>>    IF you buy a Watford Electronics 30Mhz ARM3 it may not work..
>from my experience : if you buy anything from watford it may not work

All too true, I'm afraid.

Anyone know anything about their 'ULTIMUM' box, ie does it work ?
Unfortunately there are precious few alternatives for someone with a
packed A3000, too many expansion cards and nowhere to put his/her
hard disc.

-- 
Andrew J D Hurley,     ( ajdh@stl.stc.co.uk )
Mail route:   uunet!ukc!stl!ajdh  |  Phone:   +44 279 429531 x. 2535
BNR Europe Ltd., London Road, Harlow, Essex CM17 9NA, UK.