GPU card metrics for monitoring.

We do not parse values in int or float* type and keep them original, end-user should take care type by itself. For example, Open-falcon will parses all values into JSON float64 finally.

Tags

KEY	TYPE	VALUE	NOTES
gpuid	string	0~n	Intel GPU card is hard-coded 0
vendor	string	"intel" "nvidia"

Metrics

Intel GPU Card Metrics

KEY	TYPE	VALUE	NOTES
renderbusy	GAUGE	0~100	in percent

NVIDIA GPU Card Metrics

KEY	TYPE	VALUE	NOTES
fan	GAUGE	int
temp	GAUGE	int
util	GAUGE	0~100	in percent
memory_usage	GAUGE	int	in bytes
memory_total	GAUGE	int	in bytes
memory_usage_percent	GAUGE	memory_usage/memory_total*100	in percent

Build

Collect Intel GPU cards stats requried intel-gpu-tools /usr/bin/intel_gpu_top

sudo apt-get install intel-gpu-tools

NOTICE: Intel GPU card stats collecotr required root privilege.

Collect NVIDIA GPU cards stats requried nvidia-352 /usr/bin/nvidia-smi

sudo apt-get install nvidia-352

Build example

go get -v github.com/MonitorMetrics/redis
cd $GOPATH/src/github.com/MonitorMetrics/redis/examples/json.gpu
go build -o json.gpu.bin

Start in debug mode

sudo ./json.gpu.bin -debug

For more detail, see source code.

See also

github.com/NVIDIA/gpu-monitoring-tools