[Aces-support] Matlab Memory Limits

[Ryan Abernathey]

Sai,

This is precisely the limitation I was referring to in my original  
email. A 32-bit system can only address 4GB of memory, and I needed  
more than that. I was asking whether there exists a 64-bit version of  
matlab on a 64-bit node which would be free from this limitation. I  
do think I've got this whole "exponent" thing figured out. ;-) Bill's  
suggestion would probably work, but I got around the problem by  
converting everything to single precision. The tiling would also be a  
good solution.

Thanks,
Ryan



On Aug 29, 2007, at 7:35 PM, Sai Ravela (MIT) wrote:

> 2^32 = 2^10 * 2^10 * 2^10 * 4 = 4GB (32-bit)
> (back when 80386 came, 4GB was more than the universe (in  
> segments), we've finally caught up to its edge).
>
> 2^64 = 2^10 * 2^10 * 2^10 * 2^10 * 2^10 * 2^10 * 16 = Lots more  
> memory.
>
> Read up on logarithms and exponents, and memory and addressing,   
> that should be a good starting point for figuring out what the  
> inherent limits are.
>
> Why not just tile interpolation with overlap? (Isn't interpolation  
> shift invariant? )
>
> S
>
>
>
>
> William Boos wrote:
>> You could presumably try running octave on one of the Itanium  
>> nodes, which have up to 16GB ram i believe.  though i'm sure this  
>> would require some fiddling to get your scripts to work.
>> -bill
>>
>> --------------------------------------------------------------------- 
>> ---
>> *From:* aces-support-bounces at mitgcm.org [mailto:aces-support- 
>> bounces at mitgcm.org] *On Behalf Of *Ryan Abernathey
>> *Sent:* Wednesday, August 29, 2007 2:44 PM
>> *To:* ACES-support at mitgcm.org
>> *Subject:* Re: [Aces-support] Matlab Memory Limits
>>
>> This is not a real solution, but Matt Mazloff suggested I modify  
>> the original interp3 function to convert its internal variables to  
>> single precision. This was able to reduce memory usage enough for  
>> the script to work.
>>
>> The larger question of Matlab's inherent memory limitations still  
>> stands...
>>
>>
>> On Aug 29, 2007, at 11:57 AM, Ryan Abernathey wrote:
>>
>>> Hello,
>>>
>>> I am attempting to interpolate between two relatively large grids  
>>> ( 360 x 160 x 50 --> 2160 x 320 x 42 ) using interp3. I keep  
>>> getting the OUT OF MEMORY error. I looked into how much memory  
>>> matlab can allocate, and the limit seems to be around 4GB (2^32),  
>>> which is consistent with the limit of a 32-bit system. (http:// 
>>> www.mathworks.com/support/tech-notes/1100/1106.html) I am  
>>> wondering if there is any way around this memory limit, i.e. if  
>>> there is a 64-bit node where I can run a 64-bit version of Matlab.
>>>
>>> Does anyone know anything about this?
>>>
>>> Many thanks,
>>>
>>> __________________
>>> Ryan Abernathey
>>> rpa at mit.edu <mailto:rpa at mit.edu>
>>> MIT Ph.D. Student
>>> Program in Oceans, Atmospheres & Climate
>>> http://web.mit.edu/rpa/
>>>
>>>
>>>
>>> _______________________________________________
>>> Aces-support mailing list
>>> Aces-support at acesgrid.org <mailto:Aces-support at acesgrid.org>
>>> http://acesgrid.org/mailman/listinfo/aces-support
>>
>> --------------------------------------------------------------------- 
>> ---
>>
>> _______________________________________________
>> Aces-support mailing list
>> Aces-support at acesgrid.org
>> http://acesgrid.org/mailman/listinfo/aces-support
>>
>
> _______________________________________________
> Aces-support mailing list
> Aces-support at acesgrid.org
> http://acesgrid.org/mailman/listinfo/aces-support