I'm new to AWS and received startup credits. My company has also been accepted into the NVIDIA Inception Program, and we've requested additional AWS credits. We're training our own machine-learning models and need access to GPU instances, ideally comparable to the A100s we used on Google Cloud. What is the correct process for requesting higher GPU service quotas, and what information should we include to improve the chances of approval? Have other startups using AWS credits successfully obtained larger GPU limits?
1 Answer
Use the AWS Service Quotas console to request an increase for the specific GPU instance-family quota in your target region. GPU limits are usually split by family, such as accelerated computing or a particular instance type, so make sure you request the quota that covers the exact instances you plan to run. Include your model-training workload, expected instance types and quantities, region, launch timeline, estimated runtime, and how your startup credits will cover the usage. Credits help with billing, but they don’t automatically grant additional capacity or quota.
I submitted a request but it was rejected and AWS suggested reopening the case with more details. I came from Google Cloud, where we used A100s for training. Is an appeal with a specific workload and capacity plan usually the best next step?