32/64 bit AMD question

[Zach]

It is a common misperception that "64-bits" means more/better
performance than "32-bits".  This is not (necessarily) the case.  For
those who don't remember, when we transitioned from the 16-bit 8086
and 286 to the 32 bit 386 in the 80's, the new 32 bit processor wasn't
faster, at least not because of "being 32 bits".  386's were faster
than 286s because of higher clock speeds, architecture optimizations,
and things like this.  However a 32 bit x86 cpu could let you do
things that you couldn't do with the 286, assuming software developers
took advantage.  These included addressing more memory than the 286's
1mb limit, virtualized memory mapping (each process thinks it's the
only one running), multitasking, as opposed to task swapping which the
286 was capable of, and other things.

The same thing applies to the transition to 64 bit x86.  The main
difference here is that systems now can address more than 4gb of ram. 
If your application benefits from more than 4gb ram, then you might
see a performance benefit there.

The Athlon 64 is faster than Athlon XPs due to a completely new
architecture, not because of being "64 bits".   It features things
like the memory controller being right on the cpu die, as opposed to a
chip on the motherboard, and things of this nature.

As mentioned perviously, 754 pin socket vs the 939 pin socket doesn't
cripple the "64-bitness" in any way.  The difference is the cpu can
address two banks of memory at once vs just one.  This is only going
to benefit applications and games that are starved for memory
bandwidth.

Hope this helps clear things up.

On 10/19/05, Lewis Futrell <silicon.vampire at gmail.com> wrote:
>
>
>
>
>
>  Maybe put another way, what is the performance difference in 64 bit mode
> compared to 32 bit, of the 754 pin and the 939 pin AMD Athlon64's for the
> same clock speed?
>
>
>
>  The funny thing is, it is exactly the memory interface that is different
> between the two.  The 754 socket only does single channel memory (64 bit
> path) whereas the 939 CPU's are dual channel (128 bit path).  The 64 bit
> part of the processor is the same.  The performance difference will be the
> memory bandwith.
>
>
>
> Thanks Lewis, does that mean that in 32bit mode the chip only does 32bit
> memory accesses but in 64 bit mode the 754pin chip does 64 bit and 939 pin
> chip does 128bit (memory access not calculations)?
>
>  If true, then 64 bit mode should be faster than 32bit in both chips, but
> the 939 will be faster (by the memory access time) then the 754. Do I have
> this understanding correct or am I still missing something?
>
>  Regards Russell.
>
>
>
>
>
> That is correct.  However, in most tests that I've seen on the review sites,
> the performance increase between 32 and 64 bit mode varies upon the
> application one is doing.  Sometimes 32 bit mode is just as fast, if not
> slightly faster than 64 bit, and vice-versa.  The biggest issue that you may
> want to look at is to see if there are 64 bit versions of the
> programs/applications that you want to run.  For example, there is not a 64
> bit version of Flash that I'm aware of (one can do a chroot environment to
> run 32 bit applications in 64 bit Linux, but that is a little over my head).
>  Also, some drivers do not have a 64 bit version of them, but this is
> thankfully being worked on and the likelihood of having a device that won't
> work on the 64 bit version of Linux is becoming less and less of an issue.
> --
> ubuntu-users mailing list
> ubuntu-users at lists.ubuntu.com
> http://lists.ubuntu.com/mailman/listinfo/ubuntu-users
>
>
>