Serial Execution Evaluation

Serial Evaluation

We evaluate the performance of serial evaluation strategies across systems and compare it to the MXNet runtime performance.

System Tesla_K80

BatchSize 1 on System Tesla_K80

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 2 on System Tesla_K80

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 4 on System Tesla_K80

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 8 on System Tesla_K80

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 16 on System Tesla_K80

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 32 on System Tesla_K80

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

System Tesla_M60

BatchSize 1 on System Tesla_M60

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 2 on System Tesla_M60

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 4 on System Tesla_M60

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 8 on System Tesla_M60

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 16 on System Tesla_M60

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 32 on System Tesla_M60

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

System TITAN_Xp

BatchSize 1 on System TITAN_Xp

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 2 on System TITAN_Xp

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 4 on System TITAN_Xp

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 8 on System TITAN_Xp

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 16 on System TITAN_Xp

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 32 on System TITAN_Xp

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

System TITAN_V

BatchSize 1 on System TITAN_V

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 2 on System TITAN_V

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 4 on System TITAN_V

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 8 on System TITAN_V

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 16 on System TITAN_V

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 32 on System TITAN_V

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

System Tesla_V100-SXM2-16GB

BatchSize 1 on System Tesla_V100-SXM2-16GB

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 2 on System Tesla_V100-SXM2-16GB

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 4 on System Tesla_V100-SXM2-16GB

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 8 on System Tesla_V100-SXM2-16GB

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 16 on System Tesla_V100-SXM2-16GB

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 32 on System Tesla_V100-SXM2-16GB

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

System Quadro_RTX_6000

BatchSize 1 on System Quadro_RTX_6000

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 2 on System Quadro_RTX_6000

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 4 on System Quadro_RTX_6000

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 8 on System Quadro_RTX_6000

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 16 on System Quadro_RTX_6000

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 32 on System Quadro_RTX_6000

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

System Tesla_T4

BatchSize 1 on System Tesla_T4

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 2 on System Tesla_T4

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 4 on System Tesla_T4

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 8 on System Tesla_T4

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 16 on System Tesla_T4

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime

BatchSize 32 on System Tesla_T4

Normalized Against MXNet End-to-End Runtime

Comparing Latency of Parallel, Serial, and MXNet End-to-End Runtime