search

NVIDIA / DALI

A GPU-accelerated library containing highly optimized building blocks and an execution engine for data processing to accelerate deep learning training and inference applications.
https://docs.nvidia.com/deeplearning/dali/user-guide/docs/index.html
Apache License 2.0
5.1k stars 615 forks source link

Can I use DALI on Nvidia TX2 #1022

Closed KleinXin closed 5 years ago

KleinXin commented 5 years ago

thx

JanuszL commented 5 years ago

Hi, Now DALI support for ARM platform is rather elementary. You can build the native part using this guide but without python bindings. We tested it on Xavier platform and it passes the basic unit tests (you can also check this GTC19 talk and the blog post). So feel free to experiment and we encourage you to contribute to DALI if you manage to solve some problem with launching DALI on Nvidia TX2.

klecki commented 5 years ago

Just FYI the resize kernels have some troubles working on Xavier (we suspect that using unified memory may be the cause and we are working to fix it).

klecki commented 5 years ago

Updating my previous message, #1026 fixes the issues with resampling on aarch64 platform, this change is available in weekly & nightly builds and will be part of upcoming 0.12 release.

KleinXin commented 5 years ago

@JanuszL & @klecki Thank you for your reply！ What is the fps of Resnet18 on TX2？ According to my test, it is about 50fps.

JanuszL commented 5 years ago

Hi, We haven't checked the actual performance of TX2 with DALI. You can try to check https://devblogs.nvidia.com/jetpack-doubles-jetson-inference-perf/ or other Nvidia materials regarding TX2 performance.