AccelerEyes has introduced a new version of its Jacket software platform designed for multiple GPU systems. The new version allows Matlab users to leverage the growing popularity, computation power, and energy efficiency of GPU clusters.
The Jacket platform consists of a runtime and language-processing system that automatically optimises existing applications or new algorithms for GPU computing. JacketHPC extends the base platform to deliver the ability to span computation across multiple GPUs, either on a local machine or over a network. JacketHPC enables an unprecedented ability to transparently scale GPU and CPU computing resources simultaneously. It eliminates the need to re-program the applications in lower-level programming languages, advanced application programming interfaces, or parallel extensions such as MPI (message passing interface) – decreasing time to solution and dramatically improving productivity of domain professionals.
JacketHPC is built atop MathWorks’ Parallel Computing Toolbox (PCT) and Distributing Computing Server (DCS). PCT and DCS product licenses are required for executing JacketHPC on network based HPC and GPU resources. By combining Jacket’s GPU data types with parallel constructs such as parfor, spmd, or co-distributed arrays, pre-existing code may be dispatched across all GPUs and CPUs in a cluster or a Cloud service. In many cases, little to no code revision is required to take advantage of this parallel computing capability.
JacketHPC comes at a time when a majority of scientific and engineering users have or will have access to GPU clusters in their organisations, yet few have the time or expertise to fully exploit the capabilities of the technology. JacketHPC is for GPU clusters with eight or more GPUs. The base Jacket platform will support workstations with up to eight GPUs.