Boosting Performance - for training and inference (run-time)

 

Existing HW --if you have a GPU that supports acceleration --you have to
read about your systems and Tensorflow GPU support --> see requirements (most likely a desktop machine)

• Tensorflow Version with GPU Support (if you have NVIDIA card with GPU)
  
  

  SEE ABOVE for CURRENT requirements --but, they will be something like   Hardware requirements The following GPU-enabled devices are supported: NVIDIA® GPU card with CUDA® Compute Capability 3.5 or higher. See the list of CUDA-enabled GPU cards. Software requirements The following NVIDIA® software must be installed on your system: NVIDIA® GPU drivers —CUDA 10.0 requires 410.x or higher. CUDA® Toolkit —TensorFlow supports CUDA 10.0 (TensorFlow >= 1.13.0) CUPTI ships with the CUDA Toolkit. cuDNN SDK (>= 7.4.1) (Optional) TensorRT 5.0 to improve latency and throughput for inference on some models.

   

• https://www.codingforentrepreneurs.com/blog/install-tensorflow-gpu-windows-cuda-cudnn

 

 

 

 

Improving Training Performance

 

• Cloud: Google, Microsoft, Amazon, NVidia(tensorflow)

• Specialty Machine/ cards and Laptops:

• "Tensorbook" by Lambda
  
  

 

• Nvidia "systems"
  

 

 

 

 

 

Improving Runtime/Inference Performance

 

• Specialty Laptops

• Lots of hardware card options (search the web):

  • Intel's USB "chip" peripheral