Amazon ECS with AWS Fargate added 32-vCPU task configurations on June 5. The new sizes pair 32 vCPUs with 60, 120, or 244 GiB of memory for Linux workloads on x86 or Arm.
More work fits inside one task
The release announcement lists both Fargate and Fargate Spot support, with availability across commercial and GovCloud Regions. AWS identifies compute-intensive processing and AI inference among the intended uses.
For an application that cannot be easily partitioned, a larger task creates room to revisit a previous deployment constraint. It may also let a team compare fewer large workers with a larger number of smaller workers.
Our analysis: that comparison should include the unit of failure. Concentrating more work in one task changes how much progress is affected when that task stops, how much input must be replayed, and how quickly replacement capacity becomes useful.
Benchmark useful work, not allocated CPUs
Measure a representative job from input acquisition through completed output. Track elapsed time, memory pressure, CPU utilization, and downstream bottlenecks. A task with more CPUs will not necessarily finish sooner if the application is serialized or waiting on another service.
Review worker concurrency and queue behavior together. Larger batches can make throughput look better while increasing the delay experienced by an individual job.
Include interruption handling when evaluating Spot capacity, and preserve checkpoints where the application’s work requires them. Compare the measured cost of completed work using current regional pricing rather than treating a larger allocation as inherently efficient. Start with one task definition and a repeatable workload, then choose a size from observed throughput and recovery behavior.
See the original announcement for availability and release details.