Wednesday, October 13, 2021
|ORNL||Remote Participation Only||1:00 PM – 3:00 PM (ET)|
|NERSC||Remote Participation Only||10:00 AM – 12:00 PM (PT)|
|ANL||Remote Participation Only||12:00 PM – 2:00 PM (CT)|
NVIDIA will present “CUDA Graphs” on Wednesday, October 13, 2021. This event is a continuation of the CUDA Training Series and will be presented by Matt Stack from NVIDIA.
Many HPC applications encounter strong scaling limits when using GPUs sooner than when using CPUs due to higher throughput. The latency associated with submitting work to the GPU can be a challenge to this strong scaling. CUDA graphs are a model for work submission in CUDA that help improve this situation. A graph is a series of operations (such as kernel launches) connected by dependencies, which is defined separately from its execution. This allows a graph to be defined once and then launched repeatedly. Separating out the definition of a graph from its execution enables a number of optimizations that reduce the cost of launching and executing work on the GPU. This presentation will demonstrate how to use CUDA graphs in an application and discuss in what circumstances an application speedup may be obtained by using them.
This presentation will demonstrate how to use CUDA graphs in an application and discuss in what circumstances an application speedup may be obtained by using them, followed by hands-on example exercises.
All remote participants can watch the presentations via web broadcast and will have access to the training exercises on Github. If you don’t already have access to the Summit, Cori GPU, or Theta GPU systems – temporary access to the compute systems will be limited as follows:
- Current NERSC users will have Cori-GPU access temporarily added to their accounts.
- Temporary Summit access will not be available for remote participants.
- Temporary Theta GPU access will not be available for remote participants.
If you have any questions, please contact Subil Abraham (email@example.com)
This event is now over.