64-bit vs 32-bit

Quote from vikana:

...

Processing speed and throughput is a function of clock speed, memory speed, cache and memory configuration, etc, and not the size of the addressable memory space of the CPU/MMU.
...
Vikana,

That is not strictly true. Flushing a 32bit cache is faster than flushing a 64bit cache. Moving 32bit values into and out of 32bit registers is faster than 64bit registers, etc. It is complicated and depends on application. This is worth reading:

http://www.cs.ucf.edu/~wspires/32bit_versus_64bit_amd64.pdf

Generally though, if your application uses lots of memory, as you pointed out the 64bit applications more than make that cost up by dealing with the bigger memory address space.

nitro
 
Quote from nitro:

Vikana,

That is not strictly true. Flushing a 32bit cache is faster than flushing a 64bit cache. Moving 32bit values into and out of 32bit registers is faster than 64bit registers, etc. It is complicated and depends on application. This is worth reading:

http://www.cs.ucf.edu/~wspires/32bit_versus_64bit_amd64.pdf

Generally though, if your application uses lots of memory, as you pointed out the 64bit applications more than make that cost up by dealing with the bigger memory address space.

nitro

Who is the author?
Computer Science department head or freshman student?
 
Nitro,

I agree. On my new XEON system there's a BIOS setting that lets the memory controller automatically load adjacent cache lines to help with this. Was actually turned off by default.

In my case, I'm floating point limited, which is why the savings of 32bit registers doesn't factor in for me.
 
Back
Top