R*******N
发帖数: 7494 | |
a****l
发帖数: 8211 | 2 what's the surprising part?
【在 R*******N 的大作中提到】
: 哪有Intel Itanium的性能报告呢？
|
R*******N
发帖数: 7494 | 3 It is too low, why not 80%?
Two processors running for the same program ...
【在 a****l 的大作中提到】
: what's the surprising part?
|
a****l
发帖数: 8211 | 4 why do you think two processors running the same program should be 80%
faster on dual-core?
【在 R*******N 的大作中提到】
: It is too low, why not 80%?
: Two processors running for the same program ...
|
R*******N
发帖数: 7494 | 5 Why not? hehe,
Anyway, do you have any citation (report or weblink) about that 30% performence? I read it somewhere, but forget its source...Many thanks!
【在 a****l 的大作中提到】
: why do you think two processors running the same program should be 80%
: faster on dual-core?
|
b*****e
发帖数: 1193 | 6 The truth is:
if your compiler can't support parallel programming, you can't get the 80%
increasement in execution speed.
the intermediate code generator is the bottle neck for the Dual Core
performance.The compiler is not that efficient to work with the Dual Core. |
p****f
发帖数: 251 | 7 极端情况，如果你的程序并行度为0， dual-core比single-core还慢。
performence? I read it somewhere, but forget its source...Many thanks!
【在 R*******N 的大作中提到】
: Why not? hehe,
: Anyway, do you have any citation (report or weblink) about that 30% performence? I read it somewhere, but forget its source...Many thanks!
|
R*******N
发帖数: 7494 | 8 如果是２个线程，放双核和单核上跑，单核执行完２秒，双核估计得多少？能１秒跑完？
【在 p****f 的大作中提到】
: 极端情况，如果你的程序并行度为0， dual-core比single-core还慢。
:
: performence? I read it somewhere, but forget its source...Many thanks!
|
a****l
发帖数: 8211 | 9 I believe most clasic books on parallel computer architecture have answers
to this question.
完？
【在 R*******N 的大作中提到】
: 如果是２个线程，放双核和单核上跑，单核执行完２秒，双核估计得多少？能１秒跑完？
|
R*******N
发帖数: 7494 | 10 你有reference么？主要是share data那部分对多处理器冲击如何？
记得有讲因为数据share的冲突，双核不仅不能接近2倍速，只能达到单核的130%，这样
性价比就不合算了。
【在 a****l 的大作中提到】
: I believe most clasic books on parallel computer architecture have answers
: to this question.
:
: 完？
|
L****a
发帖数: 572 | 11 see Amdahl's law :
http://en.wikipedia.org/wiki/Amdahl%27s_law
it's a law saying that when you stack up CPUs together, you can
only get sub-linear speed-up of the real computation power.
This is because of a few of bottlenecks, e.g. communication bottlenecks,
resource access bottlenecks ... |
R*******N
发帖数: 7494 | 12 多谢！请问哪有实测的数据没？比如因特尔双核之类的。
我记得原来看个文章讲，因特尔也只有30%的提高，很符合这个Amdahl law。但现在想
不起哪的数据了。
【在 L****a 的大作中提到】
: see Amdahl's law :
: http://en.wikipedia.org/wiki/Amdahl%27s_law
: it's a law saying that when you stack up CPUs together, you can
: only get sub-linear speed-up of the real computation power.
: This is because of a few of bottlenecks, e.g. communication bottlenecks,
: resource access bottlenecks ...
|
