你好,
最近,我买了2080 RTX gpu。我们目前有2 RTX 2080和1泰坦xp gpu。我希望这些gpu并行工作。但是,我一直得到CUDNN_STATUS_EXECUTION_FAILED错误。我已经添加了cuda缓存系统需求的环境变量为512 mb,但是我仍然得到同样的错误。
培训跨多个gpu。
初始化图像归一化。
| = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = |
| | |时代迭代时间| Mini-batch | |验证Mini-batch学习| | |验证基地
| | |(秒)| | |损失损失RMSE | RMSE | |
| = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = |
错误使用trainNetwork(第150行)
意想不到的错误调用cuDNN: * CUDNN_STATUS_EXECUTION_FAILED。*
错误在segnet_deadpixel_train(第86行)
(网络信息)= trainNetwork (pximds、lgraph选项);
引起的由:
错误使用nnet.internal.cnn.ParallelTrainer /火车(第68行)
错误发现工人1。
错误使用nnet.internal.cnn.layer.util.Convolution2DGPUStrategy /向前(16行)
意想不到的错误调用cuDNN: CUDNN_STATUS_EXECUTION_FAILED._
希望这个错误,当处理一个gpu的泰坦xp工作但RTX 2080,它运行缓慢,给下面的警告。
警告:GPU是低内存,这能减缓由于主存额外的数据传输性能。尽量减少
的“MiniBatchSize”培训的选择。不会再出现这个警告,除非你运行命令:
警告(“上”,“nnet_cnn:警告:GPULowMemory”)。
我试过了MATLAB与windows 2018 a和2018 b版本10 64位。我应该使用哪个版本的MATLAB解决这些问题吗?哪个版本的CUDA和CUDNN支持RTX 2080 ?金宝app我怎么能修复这个错误呢?
