olympus.utils.gpu.amd module¶

class olympus.utils.gpu.amd.AmdGpuMonitor(loop_interval, device_id)[source]¶

Bases: object

ROCm smi utility is too basic for us to use. We will use sysfs/KFD api instead.

One problem with KFD is that the node number is not guaranteed to be the same after reboot which sucks. I only have one GPU so I do not care.

Methods

parse_gpu_usage
parse_memory_free
parse_memory_props
parse_memory_total
parse_memory_usage
parse_memory_used
parse_name
parse_temperature
read_props
read_value
run
stop

parse_gpu_usage()[source]¶

parse_memory_free()[source]¶

parse_memory_props()[source]¶

parse_memory_total()[source]¶

parse_memory_usage()[source]¶

parse_memory_used()[source]¶

parse_name()[source]¶

parse_temperature()[source]¶

read_props(file)[source]¶

read_value(file_name)[source]¶

run()[source]¶

stop()[source]¶