CODEWITHSUNDEEP
gpunvidiarapidscudf
R_M
R_M

Reputation: 1

Rapids Cudf library not fast enough

I have a server with 1xNvidia T4 and a 32 Cores CPU. Trying to migrate from pandas to cudf. I am performing multiple row wise transformations on my data but I could find the CPU computing faster than GPU(using cudf). Want to understand if using cudf (with single GPU) will leverage all the cores of GPU (I am new to GPU please let me know if there is anything else I need to know about computing with cudf).

I can check the GPU utilisation going upto 100% but memory utilisation is less than 50%

I have tried using dask-cudf but it doesn't fit my usecase well as it doesn't support .groupby operations over the loop.(And as per my knowledge dask-cudf is useful if there are multiple GPU, I have a single GPU so not considering it).

Upvotes: 0

Views: 110

Answers (0)

Related Questions