Pushing the Performance Envelope of DNN-based Recommendation Systems Inference on GPUs

Keyphrases

• Recommendation System 100%
• Performance Envelope 100%
• DNN-based 100%
• Deep Learning Recommendation Models 100%
• Latency 66%
• Embedding Stage 66%
• Growing Model 33%
• World Wide Web 33%
• Occupancy 33%
• Personalization 33%
• Performance Improvement 33%
• Memory Requirements 33%
• Stall 33%
• Microarchitecture 33%
• Large Animal Model 33%
• Long Memory 33%
• Model Size 33%
• Memory Latency 33%
• Low Occupancy 33%
• Play-based 33%
• Personalized Recommendation 33%
• Compiler Optimization 33%
• Model Inference 33%
• Dataset Size 33%
• Plug-and-play 33%
• Software Prefetching 33%
• Inference Serving 33%

Computer Science

• Inference System 100%
• Deep Neural Network 100%
• Memory Requirement 20%
• Microarchitecture 20%
• prefetching 20%
• Memory Latency 20%
• Compiler Optimization 20%
• Inference Model 20%