Any way to use DeepLabCut with a GPU with only 4GB RAM (NVidia Quadro 2200)?

I only have a NVidia Quadro 2200 available for experimental testing before deciding for a GPU upgrade. Is there any way to make DLC run on a GPU with only 4GB? Should still be better than CPU, right?

That should still work better than a CPU. Check out Check out https://github.com/AlexEMG/DeepLabCut/pull/458

I added TF_FORCE_GPU_ALLOW_GROWTH=true to the environment but I’m still unable to get the GPU-DLC’s train_network to run. This time around I get the errors:

2019-11-04 10:43:46.825806: E tensorflow/stream_executor/cuda/cuda_driver.cc:981] failed to synchronize the stop event: CUDA_ERROR_LAUNCH_FAILED: unspecified launch failure
2019-11-04 10:43:46.836367: E tensorflow/stream_executor/cuda/cuda_timer.cc:55] Internal: error destroying CUDA event in context 00000239702993F0: CUDA_ERROR_LAUNCH_FAILED: unspecified launch failure
2019-11-04 10:43:46.843669: E tensorflow/stream_executor/cuda/cuda_timer.cc:60] Internal: error destroying CUDA event in context 00000239702993F0: CUDA_ERROR_LAUNCH_FAILED: unspecified launch failure
2019-11-04 10:43:46.853958: F tensorflow/stream_executor/cuda/cuda_dnn.cc:194] Check failed: status == CUDNN_STATUS_SUCCESS (7 vs. 0)Failed to set cuDNN stream.

I’m not sure if this is related to my initial problem. Before installing DLC, I installed two versions of CUDA: 9.0 and 10.0. Could the parallel installation cause trouble?

I’m on Windows 10 by the way.

Yes, you can only have 1 version of CUDA installed, and the cuDNN is a different version between them, so you should settle on one :wink: