Prev: illumina solexa sequencing
Next: please help with selecting color for graphs
From: Andrew Scott on 13 May 2010 15:47
My PC hangs at seemingly random points when performing fairly long computations (10mins - 12hrs) in 64 bit Matlab. The most common bit of code to hang is a stack of for loops used in a normalised cross correlation I wrote, but it has also crashed during my image reconstruction code (mostly for loops and FFTs). When it hangs there is nothing I can do except switch the machine off with the button. I am quite confident that the problem is worse if I turn down the fan speed in the bios (although my co-workers are not impressed at the turbine sounds coming from under my desk), but the only temperature monitoring software I can get to work says motherboard temperatures are 20-30degrees C and the GPU is about 65degrees. The problem seems to a little be worse if I use a parfor loop for each cross correlation and seems to have got worse recently. It is also useful for me to have several
vncservers running that I can log into and run slightly different tasks, running more than one seems to make it worse but I'm not sure.

Please help me someone! I need to run a lot more of these to finish my PhD and it's crashed 4 or 5 times today.

I have 64 bit Matlab (currently2010a, although the problem was the same on at least the last 3 releases) installed on a PC with 4x3.6GHz Xeon processors and 6GB RAM. I am currently running 64 bit Ubuntu 10.04 but have updated this regularly and changed from Windows XP 64bit sometime ago because of this same problem (no change). I have also tried swapping the machine (no change) as I did have access to an identical one which died (possibly unrelated).
From: John O'Flaherty on 13 May 2010 15:57
On Thu, 13 May 2010 19:47:11 +0000 (UTC), "Andrew Scott"
<a.scott07REMOVE##THIS(a)imperial.ac.uk> wrote:

>My PC hangs at seemingly random points when performing fairly long computations (10mins - 12hrs) in 64 bit Matlab. The most common bit of code to hang is a stack of for loops used in a normalised cross correlation I wrote, but it has also crashed during my image reconstruction code (mostly for loops and FFTs). When it hangs there is nothing I can do except switch the machine off with the button. I am quite confident that the problem is worse if I turn down the fan speed in the bios (although my co-workers are not impressed at the turbine sounds coming from under my desk), but the only temperature monitoring software I can get to work says motherboard temperatures are 20-30degrees C and the GPU is about 65degrees. The problem seems to a little be worse if I use a parfor loop for each cross correlation and seems to have got worse recently. It is also useful for me to have several
>vncservers running that I can log into and run slightly different tasks, running more than one seems to make it worse but I'm not sure.
>
>Please help me someone! I need to run a lot more of these to finish my PhD and it's crashed 4 or 5 times today.
>
>I have 64 bit Matlab (currently2010a, although the problem was the same on at least the last 3 releases) installed on a PC with 4x3.6GHz Xeon processors and 6GB RAM. I am currently running 64 bit Ubuntu 10.04 but have updated this regularly and changed from Windows XP 64bit sometime ago because of this same problem (no change). I have also tried swapping the machine (no change) as I did have access to an identical one which died (possibly unrelated).

It might help to get the machine out from under the desk. You could
also check for dust accumulation on the CPU heat sink. That can make a
big difference. (20-30 C doesn't seem believable, so maybe it's a lot
hotter.)
--
John
 | 
Pages: 1
Prev: illumina solexa sequencing
Next: please help with selecting color for graphs