Hello Googlers
I have a VM that needs to start as fast as possible! I’ve messed around with optimising my kernel and userspace services and got my TTL to a sub 3s
But when I attached a GPU to this VM, the staging time suddenly changed from a couple of seconds to over 30s! So, I tried a different distro - same problem, I tried a preconfigured Nvidia driver image, same again, different region.
It seems to be consistent across all my VMs no matter what I configure or do.
From the initial description, it is not so clear about the type of GPU that you are using on the GCE VM.
After checking the current restrictions that are applied to a GPU attached VMs, I would recommend you to test the scenario with a different type of GPU. This would at-least confirm whether the issue is specific to a particular GPU model or not. If the issue is persisting even after changing the GPU models, you may then need to follow the help center articles about ‘Optimizing GPU Performance’ & ‘Monitoring GPU performance’ to track and understand the better utilization of the GCE VM especially if they are attached with GPUs.
If the issue is persisting(or reproducible) even after the tuning/optimizing of GPU, then I would strongly recommend you to open an issue tracker report for an additional investigation on the GPU models. While opening the issue-tracker report, try to include as many as details such as reproduction steps, documents you were following, etc.