rapids_test_gpu_requirements

Added in version v23.04.00.

States how many GPUs and what percent of each a test requires.

rapids_test_gpu_requirements( test_name GPUS <N> [PERCENT <value>])

This function should only be used when rapids_test_add() is insufficient due to the rapids-cmake test wrappers not working for your project.

When combined with rapids_test_init() informs CTest what resources should be allocated to a test so that when testing in parallel oversubscription doesn’t occur. Without this information user execution of CTest with high parallel levels will cause multiple tests to run on the same GPU and quickly exhaust all memory.

GPUS

State how many GPUs this test requires. Allows CTest to not over-subscribe a machine’s hardware.

Any integer value >= 0 is supported

PERCENT

State how much of each GPU this test requires. In general 100, 50, and 20 are commonly used values. By default if no percent is provided, 100 is used.

Any integer value >= 0 and <= 100 is supported

Default value of 100