I was not aware of the /LARGEADDRESSAWARE link flag and the /3GB boot switch in XP. I will try those. P.S. I wish I were a computer scientist/programmer. Unfortunately I'm just an engineer.<div>Thanks<div><br><div><br>
<div class="gmail_quote">On Wed, May 4, 2011 at 10:27 AM, John Drescher <span dir="ltr"><<a href="mailto:drescherjm@gmail.com">drescherjm@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div><div></div><div class="h5">On Wed, May 4, 2011 at 11:15 AM, Prashanth <<a href="mailto:prashanth.dumpuri@gmail.com">prashanth.dumpuri@gmail.com</a>> wrote:<br>
> All,<br>
> I'm trying to allocate a vtkImageData of size 512 x 512 x 218 with 3<br>
> scalar components. To save memory I'm storing the scalar values as 'short',<br>
> but 'am still running out of memory. I'm running my code on a Windows 32 bit<br>
> machine. All the previous posts about memory allocation for vtkImageData<br>
> have been answered with "switch to Windows 7 or a 64-bit machine".<br>
> Unfortunately neither of these solutions work for me. Any other suggestions?<br>
> Thanks<br>
<br>
</div></div>It will be hard to use such a large image in a 2GB address space that<br>
is fragmented such that ~1.2GB is the largest memory allocation you<br>
can make regardless of how much ram or swap you have.<br>
<br>
Did you add the /LARGEADDRESSAWARE link flag to your settings and use<br>
the /3GB boot switch in XP or use Vista and above?<br>
<br>
This will extend the limit to 3GB and reduce the fragmentation<br>
problem. Even with this I had serious problems processing my 0.625 mm<br>
LungCT data. If you are stuck with this avoid copying the image buffer<br>
and try to avoid any image type other than 1 component short.<br>
Originally I used a vtkBlender to combine a color overlay with my<br>
LungCT and that ended up using almost 800 MB per view (Axial, Coronal,<br>
Sagittal) I replaced that with separate actors for the overlay and<br>
image and that cut the memory usage down to less than 1/3 so now I can<br>
process 1 case without running into problems. Although when I now<br>
think of it I remember a developer recently telling me if I restricted<br>
the extent of the blending operation to a slice it would have reduced<br>
the memory footprint down to just the memory needed for blending.<br>
<font color="#888888"><br>
John<br>
</font></blockquote></div><br></div></div></div>