A hybrid CPU/GPU approach for optimizing sorting throughput

Keyphrases

• Communication Overhead 100%
• Heterogeneous CPU-GPU 100%
• GPU Sorting 100%
• Between-host 50%
• Transfer Mechanism 50%
• Data Transfer 50%
• Global Memory 50%
• GPU Platform 50%
• Memory Bandwidth 50%
• Heterogeneous Approach 50%
• Performance Bottleneck 50%
• Massive Parallelism 50%
• Memory Capacity 50%
• Multi-core Parallel 50%
• NVLink 50%
• Reference Implementation 50%
• Sorting Algorithm 50%
• Out-of-position 50%
• Multi-GPU 50%
• Performance Gain 50%

Computer Science

• Communication Overhead 100%
• Parallelism 50%
• Multicore 50%
• Sorting Algorithm 50%
• Memory Bandwidth 50%
• Memory Capacity 50%
• Reference Implementation 50%
• Performance Gain 50%
• Performance Bottleneck 50%