Share via

Request TPS increase for Computer Vision and Document Intelligence on AIServices resource.

Binjie Zhang 0 Reputation points Microsoft Employee
2026-04-07T07:58:24.9033333+00:00
We are using the Azure AI Services resource merbFoundry (West US, S0 tier) for batch document processing with OCR. Our workload has two OCR patterns:

1. Computer Vision Image Analysis (api-version=2023-10-01, features=read): Documents containing 50-115 embedded images are parsed, with each image submitted as a separate API call in parallel. The current 10 TPS limit causes frequent HTTP 429 errors on image-heavy documents (scanned PDFs, presentations).
2. Document Intelligence prebuilt-layout (api-version=2024-11-30): Entire documents (PDFs) are submitted as single requests for layout-aware OCR. This is less rate-sensitive but we would like headroom for concurrent document processing.

Requested increase:
- Computer Vision Image Analysis: 10 TPS → 50 TPS
- Document Intelligence prebuilt-layout: current default → 30 TPS

Resource details:
- Resource name: merbFoundry
- Resource group: rg-merb
- Subscription: MVEP Workspace Dev (082c86d5-934d-4bf4-bc22-1fb1d5e8e088)
- Region: West US

Business justification: This resource supports our internal MERB (Model Embedding and Retrieval Benchmark) evaluation pipeline, which processes document datasets for search quality evaluation. The batch processing nature means requests are bursty but short-lived — we process 19-100 documents per experiment run, not sustained high traffic. We need higher TPS to avoid 429 errors during parallel image OCR within individual documents.
Foundry Tools
Foundry Tools

Formerly known as Azure AI Services or Azure Cognitive Services is a unified collection of prebuilt AI capabilities within the Microsoft Foundry platform

0 comments No comments

1 answer

Sort by: Most helpful
  1. Binjie Zhang 0 Reputation points Microsoft Employee
    2026-04-07T08:17:11.7366667+00:00

    I create a support ticket for this. Please close this, Thanks.


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.