Jacket MGL

Multi-GPU Option - Single Node

Jacket MGL is targeted at the workstation market. A notebook, desktop or personal supercomputer that contains one (1) to eight (8) GPUs is well suited for Jacket MGL. The base Jacket license supports the 1st GPU, while Jacket MGL enables the 2nd to 8th GPU. There are two modes of operation with Jacket MGL:

■ Simultaneous GPU Sessions: Allow the user to operate multiple MATLAB sessions across simultaneously on the GPUs available in a single computer. A large number of users currently run multiple MATLAB sessions across a number of processors or machines to accelerate their work. This arrangement is also possible with Jacket and the Jacket MGL option without more than standard MATLAB licenses.

■ Multi-GPU Sessions: Allow the user to operate a single MATLAB session that can leverage all GPUs within a single computer. Running multiple GPU sessions with Jacket MGL requires not only a standard MATLAB license but also requires the Parallel Computing Toolbox (PCT). With the simple addition of well known parallel constructs such as PARFOR and SPMD pre-existing code may be dispatched across all GPUs and CPUs in a workstation or personal supercomputer. In many cases, little to no code revision is required to take advantage of this new parallel computing capability.

Executing code such as for-loops or monte carlo simulations with Jacket MGL can dramatically accelerate time to solution on workstations and personal supercomputers. To scale to distributed HPC resources, such as GPU clusters (with 8 or more GPUs), Jacket HPC is available.

多GPU的選項 - 單節點

Jacket MGL是針對工作站市場。一個筆記本，台式機或個人超級計算機包含一（1）至八（8）個圖形處理器非常適合Jacket MGL。基礎Jacket許可證支持第一GPU，而Jacket MGL啟用第二到第八的GPU。有兩種操作模式，Jacket MGL：

■同時 GPU的屆會： 允許用戶的GPU可在一台計算機上同時跨多個 MATLAB屆會。目前大量的用戶運行多個 MATLAB屆會，跨多個處理器或機器來加速他們的工作。這樣的安排也有可能與Jacket和Jacket MGL的選項，而不需要超過標準的MATLAB許可證。

■ 多GPU屆會：允許用戶運行一個單一的MATLAB屆會利用一台計算機內的所有GPU。Jacket MGL運行多個 GPU的屆會，不僅需要一個標準的MATLAB許可證，但還需要並行計算工具箱（Parallel Computing Toolbox - PCT）。由加上於眾所周知的並行結構如PARFOR和SPMD預現有的代碼可派出所有GPU和CPU的工作站或個人超級計算機。在許多情況下，幾乎不需要更改代碼就可以採取這種並行計算能力的優勢。

使用Jacket MGL執行代碼如for-loops或monte carlo模擬，可以大大加快解決方案的工作站和個人超級計算機。要大規模分佈式HPC資源，如GPU集群（8個或更多的GPU），Jacket HPC是可用的。