ArBB for distributed memory architectures/cluster

ArBB for distributed memory architectures/cluster

We are interested is using ArBB for large scale apps. I was wondering if ArBB will be extended in future to support distributed memory architectures/clusters (e.g. SMP multicore nodes connected via a high performance network like Infiniband).

Thanks.

3 posts / novo 0
Último post
Para obter mais informações sobre otimizações de compiladores, consulte Aviso sobre otimizações.

Thank you for your feedback

Intel ArBB is currently able to utilize a single node, but it does not prevent using an infrastructure (e.g. MPI) to go beyond a single node. We are confident that our programming model is able to also cover the case you mentioned. We will count your feedback when considering this case for Intel ArBB in the future.

Thanks for the reply.

MPI is too low-level (in the sense that it is primarily a data movement layer with added collectives, and has no abstraction to data i.e. arrays). This makes MPI difficult to use in applications when compared to ArBB which has better abstractions and more suitable for scientific applications (and better productivity), where arrays are the primary (or the only) data atructure.

Deixar um comentário

Faça login para adicionar um comentário. Não é membro? Inscreva-se hoje mesmo