Benanza: Automatic μbenchmark Generation to Compute "lower-bound" Latency and Inform Optimizations of Deep Learning Models on GPUs

Keyphrases

• Latency 100%
• Deep Learning Model 100%
• Benchmark Generation 100%
• Model Execution 40%
• Parsing 20%
• Major Components 20%
• Response Time 20%
• Benchmark Dataset 20%
• Potential Optimization 20%
• Fast Pace 20%
• Analysis Design 20%
• Keep up 20%
• Model Optimization 20%
• Internal Representation 20%
• Profiling Tools 20%
• Micro-benchmarks 20%
• Automatically Generate 20%
• Convolution Algorithm 20%
• Benchmark Results 20%
• Johannes Kepler 20%
• Algorithm Selection 20%
• Latency-sensitive Applications 20%
• Current Profiling 20%
• GPU Systems 20%
• Parallel Layers 20%
• Selection Framework 20%
• CuDNN 20%
• PyTorch 20%
• Latency Metric 20%
• Tensor Cores 20%
• MXNet 20%
• Layer Fusion 20%
• Benchmark Generator 20%

Computer Science

• Deep Learning Model 100%
• Graphics Processing Unit 100%
• Execution Model 40%
• Speed-up 20%
• Algorithm Selection 20%
• Response Time 20%
• Major Component 20%
• Internal Representation 20%
• Profiling Tool 20%
• Sensitive Application 20%
• Benchmarking 20%
• Execution Layer 20%
• Processor Model 20%