NVIDIA tutorial: reading data into GPU memory with KvikIO and GPUDirect Storage
NVIDIA published a technical tutorial showing how to use the KvikIO Python API to read a file directly into GPU memory through GPUDirect Storage (GDS). The walkthrough configures storage I/O, allocates a 1 GiB CuPy buffer, performs the read via KvikIO's cuFile interface, and then uses cuFile level-three statistics inspected with gds_stats to check whether the nvidia-fs accelerated path or the CPU-staged POSIX path handled the transfer. NVIDIA stresses that KvikIO and cuFile make separate path decisions, so reaching the cuFile API does not by itself confirm the accelerated GDS path was activated. The post also notes KvikIO is part of the CUDA-X libraries for data science and is used by libraries such as cuDF.
NVIDIA says GDS avoids unnecessary staging through CPU memory and reduces overall CPU utilization, which it frames as increasingly important as model compute and training-dataset sizes grow.
GPU applications are processing ever-growing datasets with increasingly powerful accelerators. According to the 2025 Stanford AI Index, compute for notable AI models has been doubling approximately every five months, while the size of datasets used to train LLMs has been doubling approximately every eight months. As computation and data scale together, moving data efficiently from storage becomes a necessity for keeping accelerators supplied.
NVIDIA Magnum IO GPUDirect Storage (GDS) addresses this challenge by enabling a direct data path between supported storage and GPU memory. This avoids unnecessary staging through CPU memory and reduces overall CPU utilization in the system. GDS is supported by a growing ecosystem that includes many major storage vendors.
At the application layer, Python plays a central role in modern AI development. As of August 2025, nearly half of all new AI projects on GitHub were built primarily in Python while the main application interface to GDS is the cuFile C API.
NVIDIA KvikIO, a Python and C++ library for high-performance file I/O, bridges this gap. It provides high-level, idiomatic Python and C++ interfaces, along with additional controls over how I/O requests are performed.
#NVIDIA#KvikIO#GPUDirect Storage#cuFile#Magnum IO#CUDA-X#cuDF#CuPy