See that monitoring app mentioned above--- that can pull an amazing amount of data from all the computer components and it might give you the answer.
My first thoughts on version 6
- Thread starter heyho
- Start date
You can measure the power at the wall and that will be a bit more than the PSU is providing to the system, but I never push too near PSU limits due to long term component degradation. However I am running two Proxmox hosts with various VMs including VMs for BI and Docker containers for CPAI on each host. Both hosts have 3060 12GB GPUs and the power draw for each host is averaging around 75W and that's doing AI for 14 cameras. This means the GPUs are using a fraction of their 170W TDP. The GPU fans never kick even when its raining and the AI is doing a lot of work and just checking over the last 18 days I've averaged one AI analysis a second on one host.
If you're close to PSU limit you could also use nvidia-smi command line utility to power limit the GPU to avoid any high load spikes.
Thanks for the good analysis. I've measured my overall system at the plug at generally 100w with spikes up to 130w. Assuming 85% max efficiency, the little 240w PSU can handle about 190w. I've lowered my ambtions down to a GTX 1650 which would add about 50w. Is it too much of a risk to draw 180w?
P.S. Good tip on the nvidia-smi utility. They show a chart that a 10% drop in performance can be had with a 33% reduction in power consumption.
Last edited:
My system is a factory-built HP and I've read that HP is notorious for making it difficult to swap PSU's. If what I have will support a GTX1650, then I think that gives me a sufficient boost (upgrading from a Quadro P400)