[wellylug] how slow

[Pete Black]

Where high precision or large numbers are involved (e.g. when numbers 
don't fit into 32 bit values) - 64 bit can be a lot faster than 32 bit, 
since you can now fit much huger numbers, or much more precise (e.g. 
more numbers following the decimal point) into a single word, and don't 
have to waste cycles splitting the number into 2 words, doing an 
operation on each word and then re-assembling the answer out of the 
fragments.

Obviously the ability to address more memory helps too.

However, I agree that there are almost no current 'consumer' 
applications that will see a significant speedup due to a 64 bit 
architecture.

It might be interesting to see some type of 32-bit 'dual threading' - in 
which 2 32-bit values are packed into a 64 bit word and an operation 
preformed on them in parallel - SIMD style, but i think a 'proper' SIMD 
unit e.g. Altivec would be a better option anyway.

-Pete







>Quoting "E.Chalaron" <e.chalaron at xtra.co.nz>:
>
>  
>
>>How about that ?
>>
>>
>>******************************
>>64bit is generally slower than 32bit. The only benefit of 64bit is 
>>non-segmented addressing of several gigabytes of data. If you don't 
>>need that then 64 bit adressing is just overhead.
>>
>>It's an urban myth that 64bit is faster than 32bit, like people assume 
>>a 2GHz computer is twice as fast as a 1GHz computer.
>>
>>That is also why Apple have been reluctant to make a 64 bit Mac OS X, 
>>since the people who actually need it already buy AIX instead. Ditto 
>>for Windows.
>>********************************
>>
>>    
>>
>
>http://www.mail-archive.com/mjpeg-users@lists.sourceforge.net/msg05364.html
>  
>
>Post the WHOLE discussion, then ;)
>
>
>  
>