NVIDIA/DALI v1.45.0 on GitHub

Key Features and Enhancements

This DALI release includes the following key features and enhancements:

Added support for CUDA 12.8 (#5711).
Optimized (zero-copy) transfer of outputs in JAX and PaddlePaddle plugins with dynamic executor. (#5703, #5715)

Fixed Issues

Fixed passing inputs transferred from GPU to CPU with .cpu() call as keyword arguments (#5732)

Improvements

Add CUDA 12.8 support
Update libjpeg2k (#5742)
Update VERSION to 1.45.0
Remove the 'build tag' from the DALI wheel name (#5736)
Update CV-CUDA 0.8->0.12 rapidjson (ToT), google benchmark 1.9.0->1.15.1, black 24.4.2->24.8.0 (#5733)
Return TensorLayout by const-reference. (#5730)
Factor out DALIDataType. (#5729)
Improve printing in hw_decoder_bench.py (#5724)
Replace all parameter references from double backticks to single (#5716)
Enable runtime and sphinx-level signatures for ops API (#5722)
Move to CUDA 12.6 U3 (#5719)
Remove the unused max_num_stream Pipeline parameter. Deprecate max_streams in Python. (#5720)
Remove default_cuda_stream_priority from native code and deprecate it in Python. (#5717)
PaddlePaddle zero copy (#5715)
Add handling of parameter references in Sphinx documentation (#5707)
JAX zero copy (#5703)
Use FMA in separable resampling. (#5711)
Use exec-dynamic in RNN-t pipeline. Minor fix to exec2. (#5706)

Bug Fixes

Bump the cap for numpy version in tf tests (#5741)
Remove TFRecordParser dependency from backend_impl (#5737)
Fix coverity issue (#5734)
Fix passing the results of .cpu() to argument inputs. (#5732)
Use absolute addressing for parameters (#5725)
Correct nvimagecodec version in conda and in installation instruction message (#5714)

Breaking API changes

There are no breaking changes in this DALI release.

Deprecated features

Pipeline arguments max_streams and default_cuda_stream_priority are deprecated. Passing them has no effect, but triggers a warning.

Known issues:

The most recent nvImageCodec (0.4.0) is currently incompatible with DALI. Python wheels starting from DALI 1.44 pin the dependency to 0.3.0, but older releases do not specify the required version explicitly. Users of previous DALI releases may need to manually install older nvImageCodec in order to use fn.experimental.decoders.image.* or, for DALI 1.39 and 1.40, fn.decoders.image.*. The compatible version can be installed with pip install nvidia-nvimgcodec-cu12~=0.3.0.
The following operators: experimental.readers.fits, experimental.decoders.video, and experimental.inputs.video do not currently support checkpointing.
The video loader operator requires that the key frames occur, at a minimum, every 10 to 15 frames of the video stream.
If the key frames occur at a frequency that is less than 10-15 frames, the returned frames might be out of sync.
Experimental VideoReaderDecoder does not support open GOP.
It will not report an error and might produce invalid frames. VideoReader uses a heuristic approach to detect open GOP and should work in most common cases.
The DALI TensorFlow plugin might not be compatible with TensorFlow versions 1.15.0 and later.
To use DALI with the TensorFlow version that does not have a prebuilt plugin binary shipped with DALI, make sure that the compiler that is used to build TensorFlow exists on the system during the plugin installation. (Depending on the particular version, you can use GCC 4.8.4, GCC 4.8.5, or GCC 5.4.)
In experimental debug and eager modes, the GPU external source is not properly synchronized with DALI internal streams.
As a workaround, you can manually synchronize the device before returning the data from the callback.
Due to some known issues with meltdown/spectra mitigations and DALI, DALI shows best performance when running in Docker with escalated privileges, for example:
- privileged=yes in Extra Settings for AWS data points
- --privileged or --security-opt seccomp=unconfined for bare Docker.

Binary builds

NOTE: DALI builds for CUDA 12 dynamically link the CUDA toolkit. To use DALI, install the latest CUDA toolkit.

CUDA 11.0 and CUDA 12.0 builds use CUDA toolkit enhanced compatibility. 
They are built with the latest CUDA 11.x/12.x toolkit respectively but they can run on the latest, 
stable CUDA 11.0/CUDA 12.0 capable drivers (450.80 or later and 525.60 or later respectively).
However, using the most recent driver may enable additional functionality. 
More details can be found in enhanced CUDA compatibility guide.

Install via pip for CUDA 12.0:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-cuda120==1.45.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda120==1.45.0

or just:

pip install nvidia-dali-cuda120==1.45.0
pip install nvidia-dali-tf-plugin-cuda120==1.45.0

For CUDA 11:
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-cuda110==1.45.0
pip install --extra-index-url https://developer.download.nvidia.com/compute/redist/ nvidia-dali-tf-plugin-cuda110==1.45.0

or just:

pip install nvidia-dali-cuda110==1.45.0
pip install nvidia-dali-tf-plugin-cuda110==1.45.0

Or use direct download links (CUDA 12.0):

Or use direct download links (CUDA 11.0):

FFmpeg source code:

This software uses code of FFmpeg licensed under the LGPLv2.1 and its source can be downloaded here

Libsndfile source code:

https://developer.download.nvidia.com/compute/redist/nvidia-dali/libsndfile-1.2.2.tar.gz

NVIDIA/DALI v1.45.0 DALI v1.45.0 on GitHub