I was hoping to find a usage monitor like Windows task manager or Linux htop to see the load of the Myriad X on the Intel Neural Compute Stick 2. I did search this forum and saw a couple of threads saying it's not possible. Is this accurate? If so, does anyone have a reasonable work around? (Touch it to see how hot it is won't work...)
If it is not possible, what is the reason? Does the chip simply not report it's state completely enough?
Thanks for reaching out.
It is correct, there is no way to monitory the Intel® Neural Compute Stick 2 Utilization. You can only monitor the temperature using the Inference Engine Device Query API.
Can you expand on why this is not possible? If the device can report any statistics at all we can write our own basic monitoring software. So any metrics it can return are potentially useful. We're not particularly interested in temperature, that was a joke based on a suggestion in other thread!
Could you also explain or provide a link to explain what other queries are possible? The device query API mentions name and temperature as examples but presumably there are more queries that can be sent?
I do not have an explanation for why monitoring utilization is not supported, but I will take your feedback and provide it to engineering.
The metrics available by the Myriad plugin can be listed using the IECore.get_metric function, see the list below.
['NETWORK_NAME', 'SUPPORTED_METRICS', 'SUPPORTED_CONFIG_KEYS', 'OPTIMAL_NUMBER_OF_INFER_REQUESTS', 'DEVICE_THERMAL']
Are you able to shed any light on what OPTIMAL_NUMBER_OF_INFER_REQUESTS means? When called on the executable network it returns the number 4, which doesn't have any obvious meaning to me. The definition at is just "Metric to get an unsigned integer value of optimal number of executable network infer requests" which doesn't add anything.