Re: [Starpu-devel] Problems with memory in CUDA

[Samuel Thibault <samuel.thibault@ens-lyon.org>]

Cyril Bordage, le Mon 21 Oct 2013 16:12:59 +0200, a écrit :
> The "out of memory" can occur at different locations.
> I have found one example again : in cudaMemset for the initialization of
> a redux handle.

D'oh, cudaMemset running out of memory...

> #3  0x00007fffec920322 in __GI___assert_fail (assertion=0x4ae460 "_b ==
> cudaSuccess", file=0x4ae435 "utils/starpu.c", line=157,
>     function=0x4ae640 <__PRETTY_FUNCTION__.14205>
> "init_vector_cuda_func") at assert.c:101
> #4  0x000000000043e16a in init_vector_cuda_func (buffers=0x7fff7f244978,
> cl_arg=0x0) at utils/starpu.c:157

And we clearly don't want to make you systematically check for out of
memory conditions in cudaMemset calls...

I guess we don't really have another better option than defaulting to
using only e.g. 90% of the available memory, to let cuda have 10% for
its own uses. You can already do that easily by setting e.g.

export STARPU_LIMIT_CUDA_MEM=3600

if your GPUs have 4GiB memory.

Samuel