Blast from the Past with Greek Raspberry
LOL Does that make any kind of sense?
$35 Raspberry Pi 3 more powerful than $3,000,000 mainframe, circa 1980.
33 years ago the January 1983 Byte magazine published an article on benchmarks run on a wide range of computers. 8 bit micro-computers had been around for some time and 16-bit machines were not yet dominating the market. There was still a shortage of 16-bit software, the IBM PC AT-286 computer was a year away and the Intel 386 was three years away. But of course no one knew that at the time.http://www.theregister.co.uk/2012/11/23/test_template?page=3
So how powerful were all of those toy computers and the glorius 16-bit machines to come? What was a reasonable benchmark that would run on everything? The 8-bit processors did not have multiply and divide instructions so they would be blown away by any tests using heavy math. An ancient Greek dude came up with a method of finding Prime Numbers that required no computation, called the Sieve of Eratosthenes. Byte provided source code in multiple languages and ran the tests with multiple compilers of the same language when possible. This was done by volunteers all over the country.
The fastest machine was an IBM 3033 running assembly language taking 7.8 miliseconds for 10 loops. Getting an accurate measurement required 100,000 loops. The slowest machine was a Z-80 running Microsoft COBOL, taking 1 hour 25 minutes and 15 seconds. I presume they did not do 100,000 loops. Oh yeah, they would still be looping. An IBM PC running assembly language took 4 seconds. So IBM's mainframe was 500 times faster than their new desktop with a brain from Intel.
I have had my new Raspberry Pi 3 for a week now and crashed it once and had to reinstall the operating system. But ages ago I wrote a C program that duplicated the Byte benchmark. The Pi Linux install comes with a GNU C Compiler so I had to see what would happen. It compiled with two warning messages but produced executable code.
The Pi 3 takes 4.7 ms to run one instance of the program in one core. So 4 versions ran simultaneously, 1 in each core at 4.7 ms each. Running 8 instances of the program slowed things down with 2 in each core taking 8.8 ms each. Something is keeping it from doubling the time to 9.4 ms. So running 8 instances of the Byte benchmark took only a little longer than running 1 instance on the IBM 3033, a $35 machine versus a $3,000,000 machine with a 33 year time dilation.
3 is a Magic Number.
The only other ARM powered device I have tested is the Archos PMA400, a 2005 Linux based pocket computer. It took 52.2 ms and had only one core. So the Pi wins by a factor of 11 with one core and 44 with all of them. So 11 years and 33 years make a big difference. The Archos beat the mainframe in size and price at $800, but not processing power. The Raspberry won all 3 categories.
Beware of Greeks and Moore's Law, they will blow you a Raspberry.