#CPU upgrades
6 messages in this thread
Howdies!
I am contemplating upgrading my A-2000's micro-processor. I would like
to know if any of you people have attempted (and succeeded?) the same.
Initially, I was leaning toward the 68020/68881 combination, but now I am
having other thoughts. What about this 68000 "accelerator"? I have seen it
advertised, but not in operation. And are there really THAT many programs
which, despite C-A's warnings, use "move <ea>" instructions? Also, will there
be any apparent increase in processor speed, if I use a 68020, without 32-bit
RAMs? And where does the 68030 belong in this discussion?.
If anyone has any thoughts, comments or experience in this arena,
responses would be appreciated.
Charles Conlow
74116,3067
Charles, It all depends on what you want. Without 32 bit RAM the 68020 will run
only marginally faster than the super speed 68000 adapter. this is due to the
revised (i.e. faster) instructions and if turned on, the instruction cache. Now
if you want a 68881 math co-processor, that is a different story. On a 68020
card the 68881 will run 90 to 100 times faster with in-line FFP code. A program
has to be specially written and (compliled or assembled) to have in-line FFP
code. In Workbench 1.3, CBM is releasing a math library that will detect and
use the FFP with any software that uses IEEEMATHLIB.BAS. It has more overhead
and will be in the 15-25 times faster range.
A 68020 without 32 bit RAM will run 1.3 to 1.8 times faster @ 14.3 Mhz
depending on instruction mix. With 32 bit RAM it will run from 2.5 to 4.0 times
the speed of the 7.16 Mhz 68000.
The 68030 will eventually dominate as it has both an instruction cache and a
data cache. It is more efficient going to 32 bit RAM and can operate in burst
mode. Best of all it has a built-in memory management unit. The 68851 MMU for
the 68020 adds a wait state, and hurts throughput. The 68030 MMU does not add a
wait state.
The MMU allows neat tricks to get better performance: like a program to move
the Kickstart ROM to 32 bit RAM and and access to KS get re-directed to the 32
bit area. With all the gimicks, the 68030 may speed things from 3 to 5.5 times.
<continued>
Greg:
What I do *NOT* want is a money_pit! I can see uses for a 68030 with math
co-pro and a mmu, 32-bit memory in many places, but not at my desk. There are
certainly many Universities, Research Labs and Software Development Houses that
can use (maybe even NEED) such power, and they are probably the only ones who
can afford it.
What I need is simply less contention for the cpu between various processes
that are running in the machine. I can ChangeTaskPri for all that's there, but
there is the inevitable time when one process or another needs to figure
something out. I am looking to reduce the amount of time one process waits for
the cpu from 4-5 seconds to so)QPing like a second or less.
Now, p!aps I am asking the wrong question. I am not a computer scientist by
any means. But from what you have stated, it appears that without 32-bit rams
or a math chip, a 14 Mhz 68000 (as obtained from an "accelerator") seems like
the way to go. Or am I missing something?
Charles Conlow
74116,3067
Charles, This is a cost/performance decision you have to make. A 68020 board
without a FPU and 32 bit RAM will give a varaiable performance
boost. From 1.3x to about 1.8x. The 16 Mhz 68000 will be just fractionally
slower. However, the problems you described are best solved with more
horsepower. If you spend several hundred dollars (I have no idea what the 14
Mhz 68000 costs) and find the performance unacceptable then the money is wasted
as it cannot be speeded up any faster. A bare bones '020 can be expanded to a
level that may be satisfactory to you. Again, this is a choice you have to
make. A simple enhancement is to replace the stock 68000 in your Amiga with a
68010 which costs from $15 to $35. It offers a 4 % improvement, but on some
programs that can be felt. I have one in my home A2000 and I can tell a
difference in screen updates speed in microemacs. The work machine is slugish
compared to my home A2000.
A 14 Mhz 16000 board with MMU support electronics that was recently advertised
in one of the Amiga Magazines is $150 and doesn't include the MMU chip. I
guess that could come later.
<continuation>
Any program that is tied to the CHIP RAM will kill performance as each access
will result in a large number of wait states (as many as 8). for each memory
access. The 20Mhz 68030 design I am prototying will use 11.
As you can see there is a range of choices in speed verus cost.
-Greg