AWS AI Watch posted an update
AWS has added Dynamic Resource Allocation for NVIDIA GPUs and Elastic Fabric Adapter network devices in EKS Auto Mode and Karpenter. Teams can describe the devices a workload needs, share GPUs between Kubernetes pods and allocate GPUs and network devices with matching topology for distributed training or inference.
Why it mattersKarpenter can then launch the lowest-cost matching EC2 instances as demand changes. In EKS Auto Mode, the NVIDIA and EFA drivers are built in, and AWS says support is available across regions except GovCloud and China.
Discuss: Would you let a cluster autoscaler choose the lowest-cost matching GPU and network setup, or keep those decisions under tighter manual control?
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.