TensorFlow LSTM: Big-LSTM 1 Billion Word Dataset (words/second)でTESLA V100とRTX2080Ti1枚、2枚の計算能力を比較してみた

→数値が大きい程高性能
TensorFlow LSTM: Big-LSTM 1 Billion Word Dataset (words/second)スコア

RTX2080TiがTESLA V100を超えさらに2枚で大幅改善
参考サイトはここです。
データ詳細
Big LSTM ubuntu16.04.6 Tesla V100 PCIE 16GB CUDA 10.1 18.09-py3 – TensorFlow – Training performance (wordssecond) wps=25343 23151 23241
Big LSTM ubuntu16.04.6 RTX 2080Ti CUDA 10.1 i7 5960x 18.09-py3 – TensorFlow – Training performance (wordssecond) wps=39696 39548 39165
Big LSTM ubuntu16.04.6 RTX 2080Ti x2 NVLink CUDA 10.1 i7 5960x 18.09-py3 – TensorFlow – Training performance (wordssecond) wps=62502 61236 61693
GPU CPU温度の状況
Big LSTM ubuntu16.04.6 Tesla V100 PCIE 16GB CUDA 10.1 18.09-py3 – TensorFlow – Training performance (wordssecond) wps=25343 23151 23241 nvidia-smi
Big LSTM ubuntu16.04.6 RTX 2080Ti CUDA 10.1 i7 5960x 18.09-py3 – TensorFlow – Training performance (wordssecond) wps=39696 39548 39165 nvidia-smi
Big LSTM ubuntu16.04.6 RTX 2080Ti x2 NVLink CUDA 10.1 i7 5960x 18.09-py3 – TensorFlow – Training performance (wordssecond) wps=62502 61236 61693 nvidia-smi