Is CULA a GPU library or a hybrid CPU+GPU one?

[jgpallero]

Hello,

I can read in this (viewtopic.php?f=15&t=1135) thread of the forum:

"
I'm using culaDeviceSsyev(). This CPU/GPU process is taking twice as long as a CPU-only routine.
"

I'm a bit confused about if CULA uses only the GPU for computations or uses CPU+GPU, as MAGMA does. I've found nothing about in the reference manual... So, uses CULA only the GPU or CPU+GPU.

And I can also seen that CULA provides some BLAS functions. Are these functions wrappers to cuBLAS or are different implementations? If are different implementations, are them used internally in the LAPACK CULA functions or is cuBLAS used?

An finally, can be the CULA functions asynchronously executed? I've not found any information about in the reference manual

Thank you very much, and happy new year to everyone