device memory

Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)

device memory

Postby mikesalin » Fri Mar 14, 2014 4:41 am

When I do matrix factorization on GPU, is the matrix size limited to size of the device memory?
(I am new to Magma software. Sorry for stupid question.)
Posts: 4
Joined: Fri Mar 14, 2014 4:37 am

Re: device memory

Postby mgates3 » Fri Mar 14, 2014 10:30 am

Not necessarily. Routines that take the matrix in the GPU memory, e.g., magma_dgetrf_gpu, are of course limited to GPU memory. Some routines that take the matrix in the CPU memory will automatically run in "out-of-GPU-memory" mode if needed, so the matrix can exceed the GPU memory. In particular, these should run out-of-GPU-memory:

magma_xgetrf and magma_xgesv
magma_xpotrf and magma_xgesv

where x is precision: s (single, float), d (double), c (single-complex), z (double-complex).

There are also multi-GPU routines that utilize the combined memory of multiple GPUs.

Posts: 712
Joined: Fri Jan 06, 2012 2:13 pm

Return to User discussion

Who is online

Users browsing this forum: Google [Bot], Yahoo [Bot] and 2 guests