I am trying to do an evaluation of DAAL and like to know what parameters we can use to control the per node parallelism. For example, say I've a node with 2 sockets each with 4 cores - totaling 8 cores. If I run something like KMeans, is there a way to exploit these cores and if so how to tune it?
In the Beta release of the library we have a difference of level of control in C++ and Java API. On C++ side there is a choice of linking threaded or sequential library. Linked with sequential library DAAL will use one core and in threaded case - all available cores. On Java side there is only threaded version of the library at the moment, which uses all available cores.
At the same time we are very interested to hear what level of threading control would you be interested to see in the product?” Please feel free to reply us by Private message if any Private project information.