Hi all,
When I was running cp2k-7.0-cuda on a gpu, it gave such an error message as below:
* \___/ acc_hostmem_alloc_raw: Could not allocate host pinned memory *
* | *
* O/| *
* /| | *
* / \ dbcsr_acc_hostmem.F:112 *
*******************************************************************************
Does it mean it needs more memory? But I have requested the maximum memory at the GPU. Look at the DBCSR message as show below, I am not sure how many threads was used for the calculation. My question is if I already used all the memory available at GPU, is there an option to limit the threads running in GPU therefore to reduce the memory requirement?
Thanks.
DBCSR| ACC: Number of devices/node 1
DBCSR| ACC: Number of priority stack-buffers 40
DBCSR| ACC: Number of posterior stack-buffers 80
DBCSR| ACC: Number of priority streams 4
DBCSR| ACC: Number of posterior streams 4
DBCSR| ACC: Avoid driver after busy F
DBCSR| ACC: Process inhomogeneous stacks T
DBCSR| ACC: Min. flop for processing 0
DBCSR| ACC: Min. flop for sorting 4000
DBCSR| ACC: Number of binning bins 4096
DBCSR| ACC: Size of binning bins 16