An increasing number of organizations are considering moving toward the use of hybrid systems. They often don't know how to estimate the speed up they could get for their particular applications, nor having a real view of the skills and effort needed to port these applications on such a system. They also would like to size the investment they would have to make, should they decide to acquire an hybrid system for production purposes.

CAPS Compute Lab has been designed to offer a practical answer to the above questions. This is a fully equipped hybrid cluster designed by BULL and powered by NVIDIA Tesla GPUs:

• 42 Teraflops with BULL NovaScale R422 E2 compute nodes (Intel Nehalem) accelerated with NVIDIA Tesla S1070 servers
• CAPS HMPP programming environment with NVIDIA CUDA automatic code generator
• CAPS support & expertise

In addition to accessing this cluster to port their applications using HMPP, customers will benefit from CAPS support during the access period, as well as training related to HMPP CUDA and expertise of CAPS. It is a unique opportunity to benefit from an holistic approach of hybrid systems and quick start a future usage.

CAPS Compute Lab package includes:

• 2 months of experimentation on the cluster with an upper limit of 400 computational hours
• Level 1 support from CAPS during this period
• 2 days of training at CAPS or on customer site
• 5 days of expertise from CAPS consultants