Wednesday, April 23, 2008

Computing

Couple quick items related to computers.

First, has anyone gotten Neuroshare to work on a Mac in Leopard, or has everyone given up and decided to run it in VMWare or Parallels?

Second, does everyone have incredibly shitty performance on Mac OS X Leopard in Matlab?

I ran the same set of analysis routines on our dual quad core 3.2GHz machine and my 2.4GHz laptop under Vista x86. The Mac took 170 seconds, my laptop took 140 seconds. The code is "embarrassingly parallel" so I can run 8 instances of Matlab and get the same performance, more or less (about 190 seconds per run), and on my laptop (dual core) get 150 seconds for two instances, but the difference in performance is pathetic when you take into account that the laptop has a crappier bus speed, almost a full GHz lower for serial computation, is running in friggin Vista which screams of overhead (oh, and I had Outlook, Firefox and a two other programs open on the laptop, while the Mac had nothing else going on). I am considering installing XP on the system now under Boot Camp (Parallels limits you to one CPU, VMWare to two, not sure if having multiple virtual machine instances is just asking for trouble, but will probably try this first with a demo version).

Next, Ubuntu 8.04 is being released tomorrow. Just sayin'.
I'll prob try installing it just for shits and giggles. Be sure to get the AMD64 version for your Intel Core 2 based systems. I didn't know this, but apparently AMD64 is used as the moniker because 64-bit builds of Linux were for the first 64-bit desktop systems running AMD processors, but the instruction set is included in Intel's processors. And they say Linus isn't ready for the desktop... bah! With logic like naming releases for a specific processor when it isn't required or expected, I can't imagine why!
- Also, I have had success using EasyBCD for dealing with booting issues. My current system has Vista, XP, and Ubuntu 7.1 all living in harmony.
- I plan on trying the option of installing from Windows and wiping my old install, which I never used anyway. This also takes care of downloading the install via bittorrent.

5 comments:

Anonymous said...

are you running a matlab version >= 2007a? (ie not a ppc version of matlab in emulation...)

Brandon King said...

Yup, Intel version. Right now I have 2007b installed (the campus hasn't updated their group license to 2008a yet).

Anonymous said...

I'm trying to get Neuroshare's MATLAB interface working on 64-bit XP, but I can't get MATLAB to load the DLL properly.

Unfortunately, I'm new to MATLAB, and I'm not sure what it wants in a header file for the DLL. Apparently MATLAB calls a particular function exported by the DLL, and that function decodes the information passed by MATLAB in order to determine which C function should be invoked, and what the arguments are.

Do you know of anyone using the MATLAB interface in 64-bit Windows? The postdoc I'm helping has too much data for 32 bits.

Thanks,

jonhendry_usenet at mac dot com

Erk Subasi said...

Hi Brandon,

I can not see anyone replying to your first question..

Did you finally manage to get Neuroshare work in native Mac env. or decided to go the VM path.?

Cheers,
Erk

Brandon King said...

We had a few people try to tweak it and recompile the dylibs nut in the end it was unreliable. We have enough PCs in the lab that commandeering one to run a quick file transcoding isn't an issue.

So, the short answer, nope.

As for the Matlab performance issues in OSX, those issues still exist for GUI construction. In fact, both our Windows and Mac computers suffer from the ridiculous Java heap space errors when using GUIDE. A med school student working with me and myself crash Matlab 2009b an average of 5 times per day with any moderate GUI work. If we want to do a major retooling of a GUI with GUIDE (we've been putting together our own analysis and visualization package), that number could rise to basically once every 45 minutes. Even with the increased heap space java opts file. It's painful. I don't see many of us staying with Matlab for the long run, but we don't exactly have time to teach everyone a new language.

Post a Comment