jm7e+@andrew.cmu.edu (Jeremy G. Mereness) (02/13/90)
ruzun@pro-sol.cts.com (Roger Uzun) writes: > In-Reply-To: message from rnf@shumv1.uucp > > As a general rule I would say that the 65816 will compute at 1/4th to 1/3rd > times the speed of the same clocked 68030. > -Roger Alright, alright. Enough of this. I would like some real hardware gurus out there settle this once and for all. This whole thing started with the rumor of 20 MHz 65816's coming out. Anyway, the 68000 series is fast at some things, esp. multiplication and division, but NOT at memory movement. Most 68000 instructions take many more clock cycles than the 65816. Finally, as a 32-bit chip, lousy coding would make the thing have to move long words to read an address. Now, of course, there are MMU's (memory management units) built into 68020's and above, and typically they are coupled with 68881 math co-processors which gravely screw the performance scores. So, let's have a fair and objective comparison of the two platforms, their strengths and weaknesses, and what can and is done to circumvent the latter. And remember, different strengths may only manifest themselves in certain applications! A machine that flies through a benchmark may still be a slug for typical user operations!!!!! Then again, why does it take a 2 meg mac 15 disk swaps to copy a floppy? Apple // forever dammit! ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |Jeremy Mereness | Support | Ye Olde Disclaimer: | |jm7e+@andrew.cmu.edu (internet) | Free | The above represent my| |r746jm7e@cmccvb (Vax... bitnet) | Software | opinions, alone. | |staff/student@Carnegie Mellon U. | | Ya Gotta Love It. | ---------------------------------------------------------------------------
ruzun@pro-sol.cts.com (Roger Uzun) (02/14/90)
In-Reply-To: message from jm7e+@andrew.cmu.edu
>> lets have some real hardware gurus settle this.
I have worked with the 65816 for about 3.5 years now, written
a few commercially published applications for the //gs, including
2 sound editors for MDIdeas, and many sound/graphics utilties for them. I
have also written a lot of commercial Mac and Amiga software. I have spent
countless hours hand coding assembly routines for the 65816 and 680X0
processors, in the case of the 68000 on a variety of platforms.
Please recall that the 680x0 has a highly orthogonal instruction set, and
many general purpose registers as well as a much more powerful set of
addressing modes. While for some applications a similar clocked 65816 can
perform at a slightly faster rate than the 68000, for 99% of the applications
that I have coded, the 680X0 wins hands down. When I say that a 20 Mhz
65816 would perform at 1/3rd the throughput of a 20 Mhz 68030, I am being
very conservative in the 65816's favor. I invite anyone with a C compiler
to try some benchmarks I have, and see if this is not true.
-Rogergsnow@pro-freedom.cts.com (Gary Snow) (02/14/90)
In-Reply-To: message from jm7e+@andrew.cmu.edu > Then again, why does it take a 2 meg mac 15 disk swaps to copy a > floppy? Excuse me, but my 1 MEG MAC can copy a full floppy in TWO passes, I do not know what your problem was, but you sure had a problem. Gary _______________________________________________________________________________ UUCP: crash!pnet01!pro-freedom!gsnow | ProLine: gsnow@pro-freedom | Pro-Freedom: (206)253-9389 ARPANet: crash!pnet01!pro-freedom!gsnow@nosc.mil | Vancouver, Wa InterNet: gsnow@pro-freedom.cts.com | _______________________________________________________________________________
shankar@SRC.Honeywell.COM (Subash Shankar) (02/15/90)
> [Lots of messages about benchmarking the 65816 vs. the 68000 deleted.]
Well, to get a quantitative comparison of various processors (8085, 80286,
68000, 68020), there was an article in IEEE Micro (June 87) which identified
instruction mixes for various general applications (scientific, commercial,
systems, and general). I can apply these benchmarks to the 65816, but the only
problem is that they require four operations, which I don't have the time to
code. So if anybody has either code or times (best, worst, and average) for
the following two operations, I'll post the results.
These are the four operations:
Mul16 - C <= A*B where A, B, and C are all 16 bit integers. Note that
the result is only 16 bits, so you don't need to worry about
overflows. All three are stored in memory.
Fadd32 - C <= A+B where all are 32 bit floating point numbers.
Fadd64 - Same as Fadd32, except with 64-bit format
Fmul32 - Same as Fadd32 wxcept with multiplies
So if anybody has any fast code for any of these, mail me either the code or
the execution times, and I'll figure out how the 65816 ranks. All code should
use native mode 65816 insturctions, and be reasonably efficient (i.e. not SANE,
not 6502 code, etc).
Or nobody could have the code, and we could continue arguing til eternity about
which is faster then which.
---
Subash Shankar Honeywell Systems & Research Center MN65-2100
voice: (612) 782 7558 US Snail: 3660 Technology Dr., Minneapolis, MN 55418
shankar@src.honeywell.com srcsip!shankar