Hardware and Software Support ‐ Cuda Component - icl-utk-edu/papi GitHub Wiki

The below table serves as documentation for the current hardware and software the cuda component supports. If this documentation seems incorrect or outdated please open up an Issue.

	Legacy APIs (Event and Metric)	PerfWorks Metrics API (MetricsEvaluator)
Supported Architectures	Kepler, Maxwell, Pascal, Volta	Volta, Ampere, Hopper, Blackwell
Supported Compute Capabilities (CCs)	CCs <= 7.0	CCs >= 7.0
Notable Devices	K80, M40, P100, V100	V100, A100, H100, RTX 5080
Cuda Toolkits Supported - Lower Bound	11.4.0	11.4.0
Cuda Toolkits Supported - Upper Bound	12.9.1 (Removed in 13.0.0)	13.0.0

An important note to consider is that depending on your compute capability (cc) you will have an earliest Cuda Toolkit version that will be supported. Therefore, even though we have a lower bound of 11.4 that does not mean the architecture you are running on will successfully work with it. For example, for a cc of 9.0 the earliest Cuda Toolkit version that is supported is 11.8. For more on this topic, see the following post on the NVIDIA Developers Forum.