Embed presentation
Downloaded 75 times


























The document discusses out-of-core programming using NVIDIA's CUDA within a high-performance computing context, highlighting the architecture and bandwidth characteristics of GPUs and CPUs. It presents a solution to utilizing idle disk space in computer clusters as a form of RAM, proposing that disks can effectively serve as shared memory. Various applications benefiting from this computational approach are outlined, alongside challenges such as disk latency and the necessity for parallel programming.

























