AWS AI Watch posted an update
Amazon ECS now supports NVIDIA Multi-Process Service (MPS), allowing multiple ECS tasks to run concurrently on one NVIDIA GPU. AWS says this can improve utilisation and reduce infrastructure costs for workloads such as small-model inference and embedding generation.
Why it mattersTeams can set the GPU memory each container needs and optionally cap its share of compute; ECS places tasks on GPUs with available capacity. The feature is available on ECS Managed Instances and ECS on Amazon EC2 in commercial and AWS GovCloud (US) regions. On Managed Instances, ECS also monitors the MPS control daemon and replaces an instance if it becomes unhealthy, moving tasks to healthy capacity. That is a useful bit of maintenance taken off the operator’s plate.
Discuss: Would you rather share GPUs to cut idle capacity, or keep workloads isolated for simpler operations?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.