I am running jetpack 5.1 and I am running some basic object detection and seeing this “System throttled due to overcurrent”. Interestingly… I do not see this issue if I run a stress test that maxes out both the GPU and CPU’s.
My question is, is this a problem with our carrier board 5V rail dipping below the limit of 4.75? Or is it possible that the object detection software is just pushing the GPU too hard? Why would the xavier give a user to run the GPU at a certain clock speed if it will cause overcurrent issues?
Are there any logs or anything that you would like to see for my specific use case?
The message isn’t what you think it is. Jetsons have available different performance models. The reason this is useful is that people sometimes need to enforce power limitations on battery powered devices. The message is just saying that clocks were throttled back to reduce power consumption. A memory model which allows more power consumption will consume more power before clocks are throttled. The MAXN model can still show this message because thermal limits still exist…it won’t allow damage and still has max limits.
I’m trying to prove to myself that our carrier board can properly source the power that Jetson needs to run at its highest power mode.
I am using the Jetson Xavier NX SOM and am operating it in the 20W mode.
According to the power GUI, the Jetson really never gets above 50C.
However, I do see, in the Power GUI, the Jetson is temporarily pulling >20W of power. Around 22W sometimes.
Is this message being posted because the software running on the Xavier is just too GPU and CPU heavy? Or, like you mentioned, it could be related to temperature and therefore a possible fix could be a more powerful heatsink?
I can’t answer if the power rail in question is within tolerance. However, the normal symptom of a power rail with either voltages fluctuating too fast or too far is a system suddenly shutting down. I don’t think the overcurrent is an indication of that. If you don’t have enough thermal dissipation though, you might see that warning…but you’d see it in max power mode, and not in lower power modes.
The message is very very likely just a result of the Jetson trying to keep within the power profile you set. Reducing clock speeds will be the main result of seeing that message. Some combination of CPU power consumption and/or GPU power consumption is being regulated when that message occurs, but I’ve never known it to be due to insufficient power rail voltage (but system suddenly shutting down is a result you’d see from insufficient power).
Unless you are running in max power mode I think it is extremely unlikely that even a modest heat sink would ever be a problem. Most Jetsons, when initially starting, won’t even turn a fan on because they produce so little heat. I suppose running computations could be called GPU or CPU “heavy” in the sense that the CPU or GPU will consume more power if more clock cycles are used in a given time for computations, but this isn’t any kind of “pathologic” issue…this is how it was written, and perhaps the message itself is misleading.
You could run tegrastats while some computation is running to see GPU performance, but I don’t know of anything that would directly tell you some specific reason for that warning. I can tell you that 50C is far below maximum, and there isn’t any chance that 50C is harmful, nor is it a reason for thermal throttling…the message wording is misleading.
If you want to ask someone about specific rails and voltage tolerances, then I’d suggest a thread on that topic.