Page 1 of 1

logical fallacy, demonstrated

Posted: Thu Oct 02, 2003 10:30 pm
by Jonathan
Fact: AMD's Athlon FX 51 is faster than Intel's Pentium 4 Extreme Edition at running the Splinter Cell benchmark.
Fact: The Athlon FX 51 has a clockspeed of 2.2 GHz. The Pentium 4 Extreme Edition has a a clockspeed of 3.2 GHz.
Fact: The Athlon FX 51 is 64-bit capable. The Pentium 4 Extreme Edition is a 32-bit processor.
Erroneous conclusion:
Because a 64-bit chip can process more data per clock cycle, a much lower clock speed can offer more performance than a comparable speed 32-bit processor.
Reporters are the best.
In older 16-bit processors, like the Intel 80286 CPU, the processor can, for example, add numbers from 0 to 65536 (2 the power of 16) and would require two steps for higher numbers.

32-bit processors, on the other hand, can add from 0 to more than 4 billion (2 to the power of 32) in one step.

A 64-bit processor, therefore, can add from 0 to 18,446,744,073,709,551,616 in one single step, which of course means it can process tons more data per clock cycle!
That's like 4 billion times more!

Posted: Thu Oct 02, 2003 10:36 pm
by VLSmooth
Ok, you need to post a link to this article so all can correct the person (although chances are, this has already been done, repeatedly)

Posted: Thu Oct 02, 2003 10:43 pm
by Jonathan

Posted: Thu Oct 02, 2003 10:46 pm
by VLSmooth
Now if I only wasn't too lazy to dissect and reply to that article ^_^;

Posted: Thu Oct 02, 2003 11:51 pm
by quantus
You realize that there's a slight grain of truth to that argument if there are a lot of 64-bit operations and the P4 has to emulate them... Of course, I'm really doubting there are a lot of those operations in Splinter Cell, so that grain of truth got infinitely small.

Posted: Fri Oct 03, 2003 1:31 am
by Jonathan
There's one SPEC CPU2000 test which uses 64-bit ints. Generally, developers needing that kind of dynamic range turn to doubles.