On Samstag 06 Juni 2009, Vince Fulco wrote: > Attempting Hua Wong's simple squaring kernel with timing posted to the > list Thu, 28 May 2009 12:55:13, I get a cuMemAlloc failed: out of > memory error referring to gpuarray.py line 81--> self.gpudata = > self.allocator(self.size * self.dtype.itemsize)
If your card has less than a gig of memory, then that's simply accurate--Hua's code uses 1e4**2*4*2 = 800M of memory. > ValueError: couldn't parse C declarator ' float *intpart ' Good catch, fixed in git master. The release branch doesn't even have the code in question. > A number of the undistributed are problematic as well. Can send errors > if list wishes. I think it's simply that they require certain minimum amounts of memory. And meh, that's why they're undistributed. :) I did a quick test--they work fine on a 280 with 1G, but fail on my 260 with 800-some M. If you want to see them work, just edit them and have them use smaller max sizes. No big deal IMO. Andreas
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ PyCuda mailing list [email protected] http://tiker.net/mailman/listinfo/pycuda_tiker.net
