CPU starvation on a gpu_1x_a10 instance - Technical Help - DeepTalk - Deep Learning Community

CPU starvation on a gpu_1x_a10 instance

post by Shimmer on Mar 13, 2025

Shimmer
Mar 2025

I have a long-running, CPU-bound, multi-threaded program running on a gpu_1x_a10 instance. Progress over time has stalled twice for an unknown reason, as shown in this graph: Imgur: The magic of the Internet. During these periods, I can see that CPU usage is near 0% via htop.

I assume that I’m running on a VM and sharing the CPU with other users. The two plateaus in the graph seem to indicate that my program is being starved of CPU for long stretches. (It’s theoretically possible that my program has some sort of bug, but it would be difficult to explain such a drastic drop-off and recovery.)

The same thing happened to me yesterday on another instance. Is there anything I can do to prevent this from occurring?

52 views
1 link

Related topics

Topic Replies Views Activity
High CPU “steal time” on 8x A100 machine 0 692 Jun 2023
Running instances 1 376 Jan 2024
Folding @ Home Help 1 1.4k Sep 2020
How long does it usually take for GPUs to become available? 1 2.4k Nov 2022
Booting takes long time and then “alert” status for gpu_1x_h100_pcie 4 2.9k Sep 2025