Hi - New Here - Cant get past training errors - Info included for suggestions

[punch420]

• 
  I'm a bit of a newb when it comes to NVIDIA and the terms CUDA CUDNN etc.

• 
  I hope if you can get me up and running that ican be a valuable asset to the community 

• 
  I probably need robe told exactly what version etc to be running.

• 
  I have tried a couple of AVX versions of DF LAB and installed CUDA 10.1 separately.

• 
  Someone suggested uninstalling CUDA and CUDNN and i do not know how to do that or if they meant uninstall and re-install.

• • 
    Here are my specs:

• 
  nvidia GEOFORCE RTX 2070 : DRIVER 419.17

• 
  iNTEL I7-8700 @3.2GHZ

• 
  16GB RAM 

• 
  WINDOWS 10

• 
  E tensorflow/stream_executor/cuda/cuda_blas.cc:464] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED

• 
  E tensorflow/stream_executor/cuda/cuda_blas.cc.373] failed to create cublas handle: CUDNN_STATUS_ALLOC_FAILED

• 
  Error: failed to get convolution algorithm

• 
  <more messages appear after this> . cant get an attachment to work to show the rest

• 
  thanks for any help and especially for supporting this cool software. sorry im so dumb. lol

• 
  P420

 

[dpfks]

Try uninstalling, restarting and extract a fresh DFL version. DFL has everything you need to use already.

If not, make sure you don't have any other versions of CUDA installed besides 10.1

 

[Cerciusx]

Try a non AVX version. I have a fairly high end CPU running DFL and it starts and runs just fine so I'm sure it will for you too with a non AVX version

 

[fresh_gumbo]

I would put money (almost haha) on it being that you have installed CUDA separately, uninstall that, reboot and re-extract latest DFL... I just realised I basically repeated @"dpfks", I agree, I had the same issue as yourself because I had installed CUDA for something unrelated, got rid of it and now no issues!! If not @"Cerciusx" pumps out some good results so worth checking that option as well