- Newest
- Most votes
- Most comments
Hello.
Is that still the case? Or are these available on-demand now? And if so, in what regions?
On-demand startup is possible.
However, attempting to launch instances this way often results in "Insufficient capacity" errors, so associating a Capacity Block would likely ensure a more stable startup.
https://aws.amazon.com/jp/about-aws/whats-new/2025/06/pricing-usage-model-ec2-instances-nvidia-gpus/
Based on the available information, Amazon EC2 P6-B200 instances powered by NVIDIA Blackwell GPUs can be reserved through Amazon EC2 Capacity Blocks for ML. With Capacity Blocks, you can reserve accelerated compute instances for future use in cluster sizes ranging from one to 64 instances. These reservations can be made for durations up to six months, and can be reserved up to eight weeks in advance.
However, the sources provided do not specify whether P6-B200 instances are currently available for on-demand usage without capacity reservations, nor do they detail the minimum reservation duration requirements or regional availability for these specific instances. To get the most current information about on-demand availability, minimum reservation periods, and regional availability for P6-B200 instances, I recommend checking the AWS EC2 pricing page or contacting AWS Support directly through the AWS Support Center at https://console.aws.amazon.com/support/.
Sources
Navigating GPU Challenges: Cost Optimizing AI Workloads on AWS | AWS Cloud Financial Management
Reserve GPU instances for ML workloads – Amazon EC2 Capacity Blocks for ML – AWS
Relevant content
asked a year ago
asked 8 months ago
Yep, that's a fine summary of the information I already have. AAFIK me posting this question is contacting AWS Support in a way.