any body have some experience using 64bit Linux? - 未名空间MITBBS历史存档

国际科技财经博客移民网络热点娱乐民生时事公众号

Redian新闻

>未名空间

>Computation - 科学计算

any body have some experience using 64bit Linux?

any body have some experience using 64bit Linux?# Computation - 科学计算

r*d2003-12-20 08:12

1 楼

I am trying to compile program in Linux 64 bit , but it is a kind of hard to do so
because of it is not so mature, any body have any idea what's the best
reference for that?

a*s2003-12-20 08:12

2 楼

why do you want 64 bit computation? any advantage?

【在 r*****d 的大作中提到】

: I am trying to compile program in Linux 64 bit , but it is a kind of hard to do so
: because of it is not so mature, any body have any idea what's the best
: reference for that?

r*d2003-12-20 08:12

3 楼

it is supposed to be much faster than 32bit, anyway, our
group are getting opterons, so sooner or later we will move
to 64 bits...

【在 a******s 的大作中提到】

: why do you want 64 bit computation? any advantage?

X*r2003-12-20 08:12

4 楼

It's only faster (in the sense of CPU) for large integer computation that
originally need 2 32-bits integers to hold, but this situation is really rare.
For other cases, the benefit is ignorable, if any.
The really advantage is that now you can handle >4GB memory naturely
(32bit architectures handle them with the aid of special registers),
so that for example a large mount of data can be mapped into memory directly.
If you have enough physical memory, this would really boost the speed.

【在 r*****d 的大作中提到】

: it is supposed to be much faster than 32bit, anyway, our
: group are getting opterons, so sooner or later we will move
: to 64 bits...

r*d2003-12-20 08:12

5 楼

our group is doing a kind of MC simulation, I find the 32 bit code
can run 20% faster on Opteron vs the same clockwise 32 bit CPU such
as Athlon MP. I suppose this is because of the CPU operation is 64 bit,
so memory operation is faster such as memcpy() can be accelerated
twice faster. Not so sure...but when I try to compile the code using
64 bit compiler, it is tough...

【在 X****r 的大作中提到】

: It's only faster (in the sense of CPU) for large integer computation that
: originally need 2 32-bits integers to hold, but this situation is really rare.
: For other cases, the benefit is ignorable, if any.
: The really advantage is that now you can handle >4GB memory naturely
: (32bit architectures handle them with the aid of special registers),
: so that for example a large mount of data can be mapped into memory directly.
: If you have enough physical memory, this would really boost the speed.

X*r2003-12-20 08:12

6 楼

Unaltered 32-bit code running on Opteron doesn't benefit at all from
the fact that the architect is 64-bit. Opteron itself is faster than
Athlon MP (e.g. bigger L2 cache), and of course speed results from
many factors, e.g. memory bandwidth...
Typical Monta Carlo simulation uses floating point computation and has
a relative small memory footprint. I would say converting your code
from 32-bit to 64-bit won't make a big difference.
What's the problem you meet when compiling using 64-bit compiler?

【在 r*****d 的大作中提到】

: our group is doing a kind of MC simulation, I find the 32 bit code
: can run 20% faster on Opteron vs the same clockwise 32 bit CPU such
: as Athlon MP. I suppose this is because of the CPU operation is 64 bit,
: so memory operation is faster such as memcpy() can be accelerated
: twice faster. Not so sure...but when I try to compile the code using
: 64 bit compiler, it is tough...