AWS Quota Management¶
AWS imposes service quotas (formerly "limits") on how many resources an account can run — most commonly a cap on total vCPUs per instance family per region. Hitting one produces an opaque launch failure like "The requested configuration is currently not supported" or VcpuLimitExceeded. Prism can check the relevant quotas up front and generate an increase request so you find out before a launch fails, not after.
Common quota types¶
| Quota | Common default | What it limits |
|---|---|---|
| Running On-Demand Standard (A, C, D, H, I, M, R, T, Z) | 32 vCPUs | Total vCPUs across standard families |
| Running On-Demand G and VT | 8 vCPUs | GPU instances (G, VT) |
| Running On-Demand P | 8 vCPUs | GPU instances (P family) |
| Running On-Demand X | 8 vCPUs | High-memory instances |
| EBS gp3 storage | 50 TiB | Total gp3 volume storage per region |
Example: with 24 vCPUs already running, launching a p3.8xlarge (32 vCPUs) needs 56 total — over the default 32 vCPU standard quota, so the launch fails.
Commands¶
Quota tooling lives under prism admin quota:
# List the AWS quotas relevant to Prism launches (current region)
prism admin quota list
# Show quota headroom for a specific instance type
prism admin quota show --instance-type t3.medium
prism admin quota show --instance-type g4dn.xlarge
# Generate guidance for requesting a quota increase
prism admin quota request --instance-type p3.8xlarge
prism admin quota show reports the quota that governs the given instance type and how much of it is already in use, so you can tell whether a launch will fit.
Requesting an increase¶
prism admin quota request --instance-type <type> produces the specific AWS Service Quotas request to file (the quota name, region, and a suggested new value). Submit it via the AWS Service Quotas console; increases for standard families are often automatic, while large GPU/FPGA requests may need AWS review.
Required permission¶
Quota checks need servicequotas:GetServiceQuota and servicequotas:ListServiceQuotas on the credentials Prism runs with — included in the recommended IAM policy.