Re: Apple 2 benchmark?

["Michael J. Mahon" <mjmahon@aol.com>]

Alex Freed wrote:
> After reading some posts about the different accelerators
> I wonder if there is a program to test how fast a particular
> accelerator really is.
>
> I don't have any commercial accelerators but I want to
> synthesize my FPGApple for a Virtex-4 chip and see what
> kind of an EFFECTIVE  clock rate I can get out of it.

The problem with using benchmarks to gauge speedup is that
all it measures is speedup of that benchmark--and that can
vary pretty widely with different benchmarks and different
accelerator architectures--for example, caching.

The simplest and least demanding benchmark is a simple
FOR-NEXT loop with a large enough count to allow easy
timing.  (Or two nested, if your machine is *that* fast!)

The next step in complexity and specificity is to insert
"typical" computations into the FOR loop.  Some people like
addition and subtraction and IF statements, since they are
the bulk of most processor-bound computation.

If you throw in trig functions and SQR, then you'll really
slow an Apple II down, because not only are the floating-point
software-intensive, but the Applesoft implementations of them
are quite inefficient--but they're compact.  ;-)

Then there's the special case of the Byte benchmark: Eratosthenes'
Sieve, which was dutifully run on dozens of machines in the 1980s,
though there's no reason to expect that any actual computation would
be so heavy on floating-point divides.

You could the graduate to Dhrystone, which is a famous FORTRAN
benchmark used heavily from the mid-1960s to the present (now
mostly on embedded controllers).

Or, you could do what may be the best test:  running a couple
of real applications that have some relevance and measuring the
speedup on those--an Appleworks sort or text search would be one
possibility.  I'd use one of my programs that runs long enough
that I have to wait for it to finish...

The problem, of course, is broad comparability and verifiability.

If you run a popular benchmark, it makes it easier to compare Apples
to "oranges", but at the cost of realism.  If you run a real application
that is meaningful and "real", the results can only be compared for the
machines and configurations that you, yourself, measure.

Computer performance measurement is a many splendored thing, and it
steadfastly refuses to be characterized by a single number--even for
our relatively simple older machines.

-michael

AppleCrate II: An Apple II "blade server"!
Home page:  http://members.aol.com/MJMahon/

"The wastebasket is our most important design
tool--and it's seriously underused."