Website • Slack • Docs

Cost-effective serverless computing at scale

Cortex is a serverless platform for compute-intensive applications.

Use cases

• Data processing - run image processing, natural language processing, and more.
• Machine learning in production - train and serve machine learning models in production.
• Large-scale inference - efficiently scale realtime and batch inference workloads.

Scalable

• Cluster autoscaling - configure Cortex to spin up instances when load increases and spin them down when load decreases.
• Workload autoscaling - customize the autoscaling behavior of each workload to ensure efficient use of cluster resources.

$ cortex cluster info

region: us-east-1
instances: [c5.xlarge, g4dn.xlarge]
spot_instances: true
min_instances: 10
max_instances: 100

Flexible

• Any workload - define custom Python functions or containers and deploy them as realtime, async, and batch workloads.
• Any pipeline - chain workloads seamlessly to create custom data pipelines.

$ cortex deploy apis.yaml

creating text-generator (realtime API)
creating image-classifier (batch API)
creating video-analyzer (async API)

all APIs are ready!

Observable

• Structured logging - automatically stream logs from every workload to your favorite log management tool.
• Metrics aggregation - monitor your workloads with pre-built Grafana dashboards and add your own custom dashboards.

$ cortex get

API                TYPE       REPLICAS
text-generator     realtime   32
image-classifier   batch      64
video-analyzer     async      16

Affordable

• Spot instance management - spend less on EC2 instances by letting Cortex manage spot instances.
• Multi-instance type clusters - configure resources per workload to run each workload on the right hardware.

$ cortex cluster pricing

RESOURCE                       COST PER HOUR
1 eks cluster                  $0.10
2 network load balancers       $0.02 each
10-100 g4dn.xlarge instances   $0.53 each
10-100 c5.xlarge instances     $0.17 each