From: Raymond Kan on
I have been getting different answers in my Matlab (64 bit R2009b and R2010a) for two different machines (ones is Intel E8400 Core 2 Duo, the other is I7-920). I narrow down the different answers to some simple vector mulltiplication. I want you guys to confirm this with me before I report the bug. Please run the following

y=[0.03999999999999998 0.48000000000000009 0.71999999999999997 2.5600000000000005 2]';
x=[1 3 2 4 2];
format long
x*y
[0 x]*[0; y]

On my i7-920, I get 17.160000000000004 for both cases, but on my E8400, I get 17.160000000000004 for the first case, but 17.160000000000000 in the second case. Although the difference is small, the error accumalates and after a while I can get very different answers from different machines. Very annoying.

Best,

Raymond
From: dpb on
Raymond Kan wrote:
> I have been getting different answers in my Matlab (64 bit R2009b and
> R2010a) for two different machines (ones is Intel E8400 Core 2 Duo, the
> other is I7-920). I narrow down the different answers to some simple
> vector mulltiplication. I want you guys to confirm this with me before
> I report the bug. Please run the following
>
> y=[0.03999999999999998 0.48000000000000009 0.71999999999999997
> 2.5600000000000005 2]';
> x=[1 3 2 4 2];
> format long
> x*y
> [0 x]*[0; y]
>
> On my i7-920, I get 17.160000000000004 for both cases, but on my E8400,
> I get 17.160000000000004 for the first case, but 17.160000000000000 in
> the second case. Although the difference is small, the error
> accumalates and after a while I can get very different answers from
> different machines. Very annoying.
>
Annoying, yes. Bug? No. It's pretty much the norm expected from
floating point between machines to have such differences crop up.

--
From: Jan Simon on
Dear Raymond!

> I have been getting different answers in my Matlab (64 bit R2009b and R2010a) for two different machines (ones is Intel E8400 Core 2 Duo, the other is I7-920). I narrow down the different answers to some simple vector mulltiplication. I want you guys to confirm this with me before I report the bug.

If you call it a bug, which of the solutions is "correct"?!
If the solution of an equation differs on two computers, the algorithm has a limited stability. That's all. No bug.

Kind regards, Jan