Segmentation fault compile problem Code works one 1 PC but not on another

I got a problem during the compile prozess of the cuda templates which you can get here:
http://cudatemplates.sourceforge.net/

On my pc at college they compile just fine, but my pc at home gets an segmentation fault on nearly every testfile where a more complex parameter is given to the kernel, like:

kernel(Cuda::DeviceMemoryLinear1D::KernelData buf)

When i remove the parameter the compilation works.
Now i would like to narrow the problem down, so I can continue working at home.

Specs of the working Pc:
openSUSE 10.2
Geforce 9800 X2
Cuda 2.0
ITK 3.6

Specs of the non-working Pc:
Ubuntu 8.10
Geforce 8800GTS (changing from device to emulation didnt work if i did it correct)
Cuda 2.1
ITK 3.10

If you need more specific specs like kernel/gcc or whatever just ask, im at college right now so i cant all details at the moment.

Anybody any suggestions what it could be or what i could try to narrow the problem down?

Ubuntu-8.10 is not currently supported. Unless you can reproduce this on a supported Linux distribution, then I’d say this is not a bug.

I was running into the same problem with Ubuntu 8.04 (x86_64)!

@flobbes:

Try installing CUDA Toolkit v2.0 (only the toolkit, no downgrade of the driver is required), this solved the problem for me…

The spend the complete last weekend and 2 more days installing linux ditributions, nvidia driver (screwing the system up in 75% of the cases) and i didnt get it to work.
And now i find out that i just should test the old toolkit, which isnt even supported under ubuntu 8.04 (which im running right now, just for offical support reassons :) ).
So i tested it and it worked directly.

Ram11 thank you so much for you advise!

Ubuntu 8.10 compiles as well without a problem.

So it seems that its a SDK issue.

Thx again for your help.