Resource Usage

Different GPU types have different multipliers based on performance and capabilities:

Here are the Compute Credit costs per minute for popular configurations:

Credit consumption by GPU configuration:

• 1× T4: 30 credits
• 4× T4: 120 credits
• 1× A100 (40GB): 120 credits
• 4× A100 (40GB): 480 credits
• 1× H100: 360 credits

Credit consumption by GPU configuration:

• 1× T4: 120 credits
• 4× T4: 480 credits
• 1× A100 (40GB): 480 credits
• 4× A100 (40GB): 1,920 credits
• 1× H100: 1,440 credits

Credit consumption by GPU configuration:

• 1× T4: 480 credits
• 4× T4: 1,920 credits
• 1× A100 (40GB): 1,920 credits
• 4× A100 (40GB): 7,680 credits
• 1× H100: 5,760 credits

• Check weekly — Review usage weekly to avoid unexpected limit hits
• Track trends — Monitor consumption patterns to predict future needs
• Plan ahead — Upgrade before reaching limits if planning large projects
• Set internal alerts — Create team processes to notify at 80% usage thresholds

Data Rows:

• Review assets for quality before uploading
• Remove duplicates in your local dataset
• Import pre-labeled phrase grounding or VQA annotations to save time

Compute Credits:

• Choose appropriate GPU types for your workload
• Implement early stopping in training configurations
• Cancel underperforming runs promptly

Collaborators:

• Audit team membership quarterly
• Remove inactive or former members
• Use guest access for temporary collaborators (if available)

Anticipate your resource needs:

1. Calculate project requirements — Estimate Data Rows and Compute Credits before starting
2. Review historical usage — Analyze past consumption patterns
3. Plan for scale — Consider upgrading if consistently near limits
4. Contact sales early — Discuss Enterprise plans for large-scale projects

Situation: You're at 95% Data Rows with 200 more images to annotate for phrase grounding.

Options:

1. Upgrade plan — Purchase a higher tier for immediate allocation increase
2. Buy add-on pack — Purchase additional Data Rows if available on your plan
3. Optimize dataset — Review and remove any unnecessary assets (note: this won't refund Data Rows but can help you understand actual usage)
4. Contact support — Discuss temporary allocation increases or plan options

Situation: No Compute Credits remaining with 3 more training experiments planned.

Example calculation:

Completed run: 4× A100 (40GB), 2 hours
  16.0 credits/min × 120 minutes = 1,920 credits consumed

Remaining experiments:
  - Experiment 1: 1× T4, estimated 3 hours = 180 credits
  - Experiment 2: 2× L4, estimated 2 hours = 480 credits
  - Experiment 3: 1× A100 (40GB), estimated 1 hour = 240 credits
  Total needed: 900 credits

Options:

1. Wait for renewal — Pause training until next billing cycle
2. Optimize training — Use 1× T4 (1 credit/min) instead of multi-GPU setups
3. Purchase credits — Buy Compute Credit add-ons if available
4. Upgrade plan — Move to a higher tier with more monthly credits
5. Cancel low-priority runs — Focus on the most important experiment

Situation: All 10 Collaborator slots filled, need to add 5 contractors.

Options:

1. Remove inactive members — Audit and remove former team members
2. Upgrade plan — Select a plan with higher Collaborator limits
3. Share credentials — Use shared accounts (not recommended for security)
4. Rotate access — Add/remove contractors as needed per project phase

Issue: Upload button is disabled or shows error.

Potential causes:

• Data Rows quota reached
• Network connectivity issues
• File format not supported

Solutions:

• Check Data Rows availability on Resources page
• Upgrade plan or purchase Data Row add-ons
• Verify file formats are supported

Issue: "Start Training" fails with resource error.

Potential causes:

• Compute Credits depleted
• GPU configuration exceeds available credits
• Active runs consuming all credits

Solutions:

• Check Compute Credits on Resources page
• Cancel unnecessary active runs
• Reduce GPU count or select lower GPU tier
• Wait for renewal or purchase add-ons

Issue: Invite button disabled or shows limit error.

Potential causes:

• All Collaborator slots consumed
• Pending invitations counting toward limit

Solutions:

• Remove inactive or former members
• Revoke unused pending invitations
• Upgrade to plan with more Collaborators

Data Rows: Persistent allocation that does not renew or roll over. Your total Data Rows are fixed to your plan.

Compute Credits: Typically do not roll over; unused allocations reset at renewal. Check your specific plan terms.

Collaborators: Are persistent slots, not monthly renewals.

Depends on your plan. Some plans offer add-on purchases for Data Rows and Compute Credits. Contact support to explore options.

Your existing data remains safe and accessible. You simply cannot perform new operations that consume the depleted resource until renewal or upgrade.

No. Data Rows are consumed at the time of upload and annotation pair creation. Deleting assets or annotation pairs does not refund Data Rows.

Usage is tracked in real-time and reflects actual consumption. Minor delays (seconds) may occur during bulk operations.

No. Resources are organization-specific and cannot be transferred or shared between organizations.