FAQ

What is the recommended hardware for CN24?

We recommend a new-ish (Kepler/Hawaii) GPU with at least 4GB of VRAM. Datasets in CN24's Tensor format can be very large, so your machine should either have enough RAM to fit your largest dataset or be equipped with a fast SSD.

If configured correctly, CN24 can use most of a GPU's theoretical performance. Look for cards with high GFLOPS and high memory bandwidth.

CN24 does not support SLI or CrossFireX at this time.

CN24 shows the error message "Cannot deserialize from this stream"!

Please check the paths in your dataset configuration file. Make sure there are no whitespaces in the training= and testing= lines and that the paths are either absolute or relative to the current working directory. Paths relative to the dataset configuration file are not supported!

CN24 shows an OpenCL error code! What does it mean?

Most OpenCL errors in CN24 are caused by memory shortage. All the output and error buffers for each layer have to fit in the GPU's memory. You can try a lower parallel batch size (see this article) if you haven't already. If you don't need maximum performance, you can switch to one of the supported BLAS libaries for CPU training.

This article contains a table with explanations for all the OpenCL error codes. If you are sure that your error is not the result of too little VRAM, please write an email to the maintainer.