Resource Allowances vs. Limitations

Restrictions related to resource usage in HyperAI are divided into two main categories:

1. Resource Allowances / Balance: Determines whether you currently have available compute hours, storage resources, or account balance.
2. Quota Limitations: Determines the maximum number of resources your account can create or run concurrently.

You can view your remaining allowances for different resources by navigating to the left sidebar and selecting “Settings” - “Quota Usage”.

The system prioritizes deducting fees from the compute resources displayed on the “Quota Usage” - “Overview” page, billed hourly by default. Once this compute hours quota is exhausted, the system will begin deducting from your account balance. If your balance is insufficient to cover the running compute resources, the system will stop the running container.

What is a Quota Limitation?

A quota limitation controls the maximum concurrency of resources under the current account, as well as the total number of various resources you can own. Its purpose is to cap the account's usage, rather than modify your current balance or the cluster scale.

To ensure stable platform operation and fair usage for all users, HyperAI applies limitations on resource creation and concurrent execution. You can view the primary limitations for your account under “Settings” - “Quota Usage” - “Quota Limit”.

Common limitations are categorized as follows:

1. Concurrency Limits on Compute Resources

Each compute resource type, such as rtx-5090 or rtx-pro-6000, has its own concurrency limit. This limit specifies the maximum number of containers of that resource type that can run simultaneously under your account.

Once the limit is reached, creating another container of the same resource will result in an error, for example:

Resource [rtx-5090] exceeds personal limit

If a resource's limitation is set to 0, the account is not allowed to create containers using that resource. For example:

• If the quota for rtx-5090-4 is 0/0/0, the account cannot create containers using the rtx-5090-4 resource.

2. Total GPU Type Limits

In addition to individual concurrency limits based on resource names, the platform also sets an overall cap based on the GPU Type (e.g., rtx-5090, rtx-pro-6000).

These limits work together with the resource concurrency limits mentioned above:

• Resource limitations restrict how many containers of a specific resource configuration can run concurrently.
• GPU limitations restrict the total number of GPU cards of that type that can be occupied across all containers.

For example, suppose a container requires rtx-5090-2:

• If the concurrency limit for rtx-5090-2 is 3, you could theoretically run up to three containers.
• However, if the overall rtx-5090 GPU limit is 4, you can actually run only two rtx-5090-2 containers simultaneously.

Therefore, the effective maximum number of running containers is determined by whichever limit is more restrictive: the resource limitation or the GPU limitation.

3. Quantity Limits on Datasets, Containers (Projects)

Beyond runtime concurrency, accounts are subject to "ownership quantity" limits, which include:

• Number of Private Datasets
• Number of Public Datasets
• Number of Private Containers
• Number of Public Containers

These limits apply to the number of resources currently owned by the account, rather than the number actively running.

4. Organization Limits

Personal accounts have a limit on the number of organizations they can create. Once this limit is reached, no additional organizations can be created.

5. Organization Member Limitations

In addition to account-level and organization-level limitations, organizations can set individual limitations for their members.

• Member limitations are primarily used to restrict the concurrent containers of various resources that a member can use within the organization.
• Member limitations cannot exceed or override the organization's overall limitation; both take effect simultaneously.
• Consequently, the actual upper limit for a member depends on whichever is stricter: the member limitation or the organization limitation.

For example, if an organization has an rtx-5090 concurrency limit of 4, but a member has an rtx-5090 limitation of 1, that member can run at most one corresponding resource instance at a time, even if the organization still has remaining quota.

For more details, see Organization Resource Management.

6. Concurrent Snapshot Limits

When creating workspace snapshots, HyperAI limits the number of snapshots that can remain in an unfinished state simultaneously.

You must wait for existing snapshots to complete or delete them before creating additional snapshots.

7. Single Upload Size Limits

The "Quota Limit" also displays an "Upload Resource Limit (MB)", which represents the maximum file size allowed for a single data upload by the current account.

Unlike the previously mentioned quantity and concurrency limitations, this limit is heavily tied to your remaining storage space. If your available storage is insufficient, you may be unable to upload larger files even if other limitations have not been exceeded.