ImageJ 64bit memory limitations

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

ImageJ 64bit memory limitations

Alexander Brandt
Hello!

We have quite big 5D image sets that regularly lead to OutOfMemory
Exceptions in ImageJ. I was therefore thinking of upgrading to a 64-bit
WinXP since the 64 Java runtime is available now from Sun.
Since its quite some work to upgrade the system and upgrading will lead
to other limitations: Does anyone has experience with a similar setting?
Will the 64bit environment with some GB (>4) of main memory successfully
remove the ~1800MB memory limitation of ImageJ or is there a limitation
within ImageJ itself? Does anyone use ImageJ under a 64bit Windows with
large memory?

Thanks in advance,

Alexander
Reply | Threaded
Open this post in threaded view
|

ImageJ 64bit - another round

Michael Weber-4
Dear list,

I would like to open again the discussion about running ImageJ on 64bit
systems. There were some open question, maybe there's already a
solution. Since we have now a Quad-Core Xeon system with 8 GB RAM
available (and not only weak Dual Cores with 4 GB...), it was time to
test this machine for ImageJ.


The first thing that was new to me was that the current version of 64bit
Java is now also running on Intel CPUs. The last time I tried it only
worked on AMD. So if are facing this issue, try the JDK 5.0 Update 9.
With this, it's possible to get rid of the ~1.6 GB limitations. But, to
my surprise, there's still a limitation at 3999 MB (out of memory
message, checked with memory monitor), even if you can config ImageJ for
more and it starts. Any ideas why this is happening?

Second question would be a multi-thread feature. Every process in ImageJ
gets another core for calculation, but often the program is blocked
already by one running process. By starting ImageJ two time in parallel,
I was again facing strange out-of-memory messages. Is there a more
clever way to use multiple CPU cores?


That's it for today, I'm waiting for your input ;).

cheers,
Michael



Alexander Brandt schrieb:

> Hello!
>
> We have quite big 5D image sets that regularly lead to OutOfMemory
> Exceptions in ImageJ. I was therefore thinking of upgrading to a 64-bit
> WinXP since the 64 Java runtime is available now from Sun.
> Since its quite some work to upgrade the system and upgrading will lead
> to other limitations: Does anyone has experience with a similar setting?
> Will the 64bit environment with some GB (>4) of main memory successfully
> remove the ~1800MB memory limitation of ImageJ or is there a limitation
> within ImageJ itself? Does anyone use ImageJ under a 64bit Windows with
> large memory?
>
> Thanks in advance,
>
> Alexander
Reply | Threaded
Open this post in threaded view
|

Re: ImageJ 64bit - another round

Prof. Dr. Karl-Heinz Kunzelmann
hello all,

thank you Michael for starting another round of this discussion.

I have a similar problems here using microct images with around 2 GB
size. I want to do 3D euclidean distance transformation and I am
peramently out of memory.

I use SUSE 10.1 and SUSE 10.2 64 bit with AMD 64  and Core2duo
processors. Unfortunately I could not get more then 3 GB RAM so far due
to budget constraints at the end of the year.

The hint to get the JDK 1.5 update 9 is very valuable. I have the 1.7 GB
memory limit now, too.

I would like which configurations are working before I invest more money
in hardware.

Thanks to all with more experience than I have.

Karl-Heinz
Reply | Threaded
Open this post in threaded view
|

Re: ImageJ 64bit - another round

ctrueden
Hi,

I recently upgraded my system to a quad (2x dual core) MacPro with 5
GB of RAM. I briefly tested ImageJ and found that the maximum memory
allocation Java would accept was 2108MB. I am running OS X 10.4.8 with
Java 1.5.0_06. Supposedly 10.5 (Leopard) will include Java 1.6 and
provide full 64-bit support. If I get my hands on Leopard when it
comes out next spring I'll try ImageJ again and report back to the
list then.

I also tested ImageJ performing operations on multiple images
simultaneously. I loaded two large (542 slice) QuickTime stacks, and
performed smoothing, edge detection and convolution on both
simultaneously, while monitoring CPU usage. To my surprise, I found
that load was distributed pretty much uniformly across all four
processors, rather than one being used. Some research indicated that
it's OS X itself handling this load distribution
(http://lowendmac.com/musings/05/0929.html) -- even a single operation
at once gets distributed evenly. In any case, I could not duplicate
Michael Weber's claim that "often the program is blocked already by
one running process." There should not be any reason to launch
multiple ImageJ instances at once on a single machine (and ImageJ even
guards against doing so by default).

-Curtis

On 10/19/06, Prof. Dr. Karl-Heinz Kunzelmann <[hidden email]> wrote:

> hello all,
>
> thank you Michael for starting another round of this discussion.
>
> I have a similar problems here using microct images with around 2 GB
> size. I want to do 3D euclidean distance transformation and I am
> peramently out of memory.
>
> I use SUSE 10.1 and SUSE 10.2 64 bit with AMD 64  and Core2duo
> processors. Unfortunately I could not get more then 3 GB RAM so far due
> to budget constraints at the end of the year.
>
> The hint to get the JDK 1.5 update 9 is very valuable. I have the 1.7 GB
> memory limit now, too.
>
> I would like which configurations are working before I invest more money
> in hardware.
>
> Thanks to all with more experience than I have.
>
> Karl-Heinz
>