Lightning

• Data processing using GPUs on distributed platforms
• Program a cluster of GPUs just like programming a single GPU
• Multi-GPU execution of existing CUDA kernels with minimal effort
• Focus on scalability and efficiency

Lightning is a framwork for fast data processing using GPUs on distributed platforms. The framework enables distributed multi-GPU execution of compute kernels written in CUDA in a way that resembles programming CUDA using a single GPU. This enables scaling existing CUDA kernels to much larger problems size, far beyond the memory capacity and compute capabilities of a single GPU.