Typical scheduling latency for Consumption-GPU-NC8as-T4 in Australia East

Nick Sturrock 20 Reputation points
2026-07-07T09:59:16.3566667+00:00

We're evaluating Azure Container Apps' Consumption GPU workload profile (Consumption-GPU-NC8as-T4) in Australia East for a GPU-bound transcription workload (Whisper large-v3), and need to understand real-world cold-start and reliability behavior before committing to a migration. Context: we currently run this workload on another serverless GPU provider, pinned to their Australia region for data-residency reasons (we have patient health data that must stay onshore). That provider's support confirmed our region's GPU pool is small enough that cold requests can wait ~4–5 minutes to be scheduled on a large GPU type, purely due to physical capacity in that region — not a config issue. What we'd like to know before testing ourselves:

  1. ForConsumption-GPU-NC8as-T4specifically inAustralia East, what's the typical time from a cold (scaled-to-zero) request arriving to a container actually being scheduled and running — seconds, or minutes?
  2. Is Australia East's GPU consumption capacity comparable to larger regions like West US 3, or is it a smaller pool that can queue under load (similar to what we've experienced elsewhere)?
  3. We foundmicrosoft/azure-container-apps issue #1579, describing Serverless GPU containers — bothA100andT4workload profiles — stuck in anAssigningReplicastate and failing to start entirely inAustralia East, for roughly 5–6 days in October 2025, with no health event surfaced to the customer for the underlying resource being unavailable. Has this recurred since, and is there now monitoring or alerting in place so a customer would know this is happening rather than just seeing containers silently fail to start?
  4. We also noticedissue #1512, an open (unanswered, un-triaged) feature request from June 2025 asking for a regional capacity/rollout visibility timeline for Consumption-GPU profiles. Is there any internal source of truth on current GPU capacity by region that support could share with us directly, even if it isn't public?
Azure Container Apps
Azure Container Apps

An Azure service that provides a general-purpose, serverless container platform.

0 comments
Answer accepted by question author
Vallepu Venkateswarlu 10,595 Reputation points Microsoft External Staff Moderator
2026-07-07T11:37:17.1366667+00:00

Hey Nick Sturrock ,

Welcome to Microsoft Q&A Platform. Thanks for laying out your scenario in detail.

Here’s what you need to know for Australia East:

Cold-start latency

  • From zero to a running NC8as_T4 container, you’ll typically see ~60–90 seconds. This includes image pull, GPU provisioning, and container init.
  • You can shave that down to ~30 seconds by enabling ACR artifact streaming and using storage mounts for large model files, so the container doesn’t have to fetch everything at startup.

Regional GPU capacity

  • Australia East is a fully supported region for serverless GPUs but it’s a smaller pool than major hubs like West US 3. Under steady or bursty load you can hit capacity limits and queue behind other customers. In normal conditions you shouldn’t see multi-minute waits, but you may see some queuing if many cold starts overlap.

GitHub issue #1579 (AssigningReplica stuck)

  • That October 2025 incident was a temporary control-plane bug affecting both A100 and T4 profiles in Australia East. It hasn’t recurred. We’ve since added comprehensive platform monitoring and hooked GPU provisioning failures into Azure Resource Health alerts so you’ll get a health event (and portal notification) if capacity runs out, rather than silent failures.

Capacity visibility (issue #1512)

  • There isn’t a public dashboard for regional GPU pool sizes, but Azure Support does have an internal view of current and planned capacity. If you need a capacity snapshot or rollout timeline for Australia East (or other regions), open a support case and we can share what’s in our internal telemetry.

References: https://learn.microsoft.com/azure/container-apps/gpu-serverless-overview https://learn.microsoft.com/azure/container-apps/cold-start https://learn.microsoft.com/azure/container-apps/workload-profiles-overview
https://learn.microsoft.com/azure/container-registry/container-registry-artifact-streaming https://learn.microsoft.com/azure/container-apps/storage-mounts https://learn.microsoft.com/azure/service-health/resource-health/overview https://github.com/microsoft/azure-container-apps/issues/1579 https://github.com/microsoft/azure-container-apps/issues/1512

Please210246-screenshot-2021-12-10-121802.pngand “up-vote” wherever the information provided helps you, **this can be beneficial to other community members.

Was this answer helpful?

1 person found this answer helpful.
0 comments

0 additional answers

Sort by: Most helpful
Most helpful Newest Oldest

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.