[sdiy] Small MCU MIPS, DMIPS?

[Scott Gravenhorst]

Rainer Buchty <rainer at buchty.net> wrote:
>On Mon, 28 Feb 2011, Scott Gravenhorst wrote:
>
>> I disagree with the premise that it's useless, meaningless or just
>> marketing blather.
>
>Sorry, but it *is* meaningless.
>
>Take two CPUs for instance, both showing 100MIPS.
>
>CPU #1 has a CPI of 8, hence to achieve 100MIPS it will need 800MHz.
>CPU #2 has a CPI of 1, hence to achieve 100MIPS it requires 100MHz.
>
>Which is better?
>
>
>Now let's look at the code:
>
>CPU #1 requires for your application 5000 lines of code with an average 
>size of 1.5 bytes per instruction.
>
>CPU #2 does the same application in just 20000 lines of code with 4 
>bytes per instruction.
>
>Which is now better?
>
>
>MIPS says nothing. You need to have the reference frequency at which 
>these MIPS occur or a CPI value. And then you still have no measurement 
>of resulting code size.

Yes, I understand this.  My point all along was that it's not the only rating to
consider.  It is an attribute and was not meant to say everything there is to say
about a device.  In the DEC world, it was used for comparing one VAX model to the
next.  The MIPS rating of an 11/780 being higher than an 11/750 was particularly
valid in comparing those two machines because they both ran the same instruction set.

I still say that it is not _useless_, just not the value that some would like it to
be.  It is simply one of many ratings to consider -- and my main point is that it is
the particular application running on the devices to be considered that is most
important and that if it is critical, a benchmark is the best way to evaluate
performance.  For example, It is obvious to me that a 1 MHz 6809 (don't know it's
MIPS, but it's less than 1) is simply not going to perform as well as a 40 MIPS dsPIC
in any application, so I would eliminate it from the devices to consider.  The
problem becomes more difficult when the MIPS ratings are similar, hence why I would
want to test with my application.  Of course, if one device is, for example, 8 bit
and the other is 16 bit, there's a good bet that even if MIPS are the same, the 16
bit device would outperform the 8 bit device in 16 bit applications.  But we are now
dealing with two ratings, MIPS and bit width and trying to compare rocks with horse
carts.  So "all things being equal" (which they never are)...  Benchmark.

-- ScottG
________________________________________________________________________
-- Scott Gravenhorst
-- FPGA MIDI Synthesizer Information: home1.gte.net/res0658s/FPGA_synth/
-- FatMan: home1.gte.net/res0658s/fatman/
-- NonFatMan: home1.gte.net/res0658s/electronics/
-- When the going gets tough, the tough use the command line.