Score:0

T4 GPU slows down automatically after sometime with power usage limited

am flag

I had been using T4 GPU with 8 virtual CPUs and 30 GB RAM Virtual Machine for more than a month without any issues.

I have been running some deep learning training jobs on it.

But since yesterday, I have been trying some new models on a larger dataset and I noticed that in the beginning the training runs fast (like it always normally did). But after one or two epochs, the GPU slows down considerably, taking five times more time per step!!!

On checking GPU usage with nvidia-smi, I noticed that initially the power usage fluctuates around 70 W (which is the limit) but then when GPU slows down, the power usage is also fluctuating around only 49 W.

It seems that the VM automatically throttles the GPU performance due to overheating/high power usage? I would like to know if I am doing something wrong that's getting me in to this situation. I couldn't find anything in the documentation on the subject.

Thanks

bjoster avatar
cn flag
Your GPU is most likely beeing throttled due to overheating. This has to be fixed in the physical world, not in your VM.
I sit in a Tesla and translated this thread with Ai:

mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.