Greg Freemyer wrote:
On 10/30/07, Aaron Kulkis <[EMAIL PROTECTED]> wrote:
Greg Freemyer wrote:
On 10/30/07, Aaron Kulkis <[EMAIL PROTECTED]> wrote:
Robert Lewis wrote:
I recently bought a 64 bit machine and have 10.3 loaded . It seems very fast.
I am trying to appraise if the advantages of having some software not work
because of driver issues is worth it.
Has anyone run benchmarks on an identical system with 32 bit vrs 64 bit?
Do we have a list of software that is known not to work on 64-bit?
Any defined group tracking this and championing gettin the remainder
fixed.
I know this is a generalized question but I am just trying to appraise
the gain vrs pain of deciding to stay with 64-bit ?
Most data isn't 64-bit. In fact, a large chunk of it
is still 16-bit, and even 8 bit (ASCII text, for example).
Most of the speed increase you are seeing is from:
1) the faster clock speed of your new CPU compared
to the last one you had.
2) wider data buses into and out of the on-chip cache.
3) floating point operations (where applicable).
You won't see a significant difference between
32-bit and 64-bit system performance until you
are running some software which uses lots of
64-bit integers, or is very intensive in
floating-point operations.
Surprisingly (to me at least) we are seeing a speed improvement with a
specialized version of dd (dcfldd) going against raw disks.
Why is it surprising that dd (which just moves large
chunks of data) would be improved by a movement from
4-byte natural word size to 8-byte natural word size?
That's like being surprised that thicker pipes can
move more gallons of water per hour.
Because everyone knows that the disk
drive/SATA-bus/controller/PCI-bus/DMA are the limiting factors in raw
dd speed, so who cares how the kernel/app are compiled.
That's true for unsophisticated disk I/O sub-systems
like in Microsoft's products.
But dd doesn't write to disk... it writes to buffers
held in RAM, and then forgets about it, letting the
system flush the buffers out to disk when needed.
And my tests were with a PCI-bus controller. With a PCI-Express
controller the 32-bit/64-bit issue for kernel/userland is likely even
more important.
Greg
--
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
