Guidance for the Accelerator Test

Accelerator test quick guide.

This test uses the global accelerator test (gat) script (/opt/cray/nodehealth/default/bin/gat.sh) to first detect the accelerator type and then launch the test specific to that type of accelerator.

The gat script supports two arguments for NVIDIA GPUs:

-mmaximum_memory_size

Specify the maximum_memory_size as either a kilobyte value or a percentage of total memory. For example, -m 100 specifies that no more than 100 kilobytes of memory can be allocated, while -m 10% specifies that no more than 10 percent of memory can be allocated.

In the default NHC configuration file, the specified memory size is 10%.

-r

Perform a soft restart on the GPU and then rerun the test. In the default NHC configuration file, the -r argument is specified.

The gat script has the following options for Intel Xeon Phi:

-M kilobytes or -M n%

This option works exactly as the -m option for the NVIDIA GPUs.

-c

Specifies the minimum number of cores that must be active on the Xeon Phi for the test to pass. If -c is omitted, the minimum number of active cores required to pass the test is the total number of cores on the Xeon Phi.