Skip to main content
Version: User Guides (BYOC)

FAQ: Resource Planning

This topic lists the possible issues that you may encounter while planning your resources on Zilliz Cloud and the corresponding solution.

Contents​

FAQs​

What is a Compute Unit (CU)?​

A compute unit (CU) is a group of hardware resources for serving your indexes and search requests. You can simply consider a CU as a fully-managed physical node for deploying search service.

For more details, see Select the Right CU.

What is a vCU? How does it get calculated?​

A vCU is a virtual compute unit used to measure the resources consumed by read operations (such as search and query) and write operations (such as insert, upsert, bulk insert, and delete). The data volume written or read will be converted from GB to vCUs. For details, refer to Understand Cost.

How can I avoid expenses on unused clusters?​

We recommend suspending unused clusters to save computing costs. You can resume them later when necessary.

How many CUs do I need for a given collection?​

  • Performance-optimized CU: Supports up to 1.5 million 768-dimensional vectors.

  • Capacity-optimized CU: Supports up to 5 million 768-dimensional vectors.

  • Extended-capacity CU: Supports up to 20 million 768-dimensional vectors.

These estimates are based on vectors with primary keys only. Additional scalar fields like IDs or labels may reduce capacity. We recommend conducting your own tests for accurate assessment.

Which type of CU should I pick?​

Select the Performance-optimized CU if you instant search results and high concurrent traffic for real-time applications. Choose the Capacity-optimized CU if you need to handle large vector datasets while maintaining reliable search speeds. Opt for the Extended-capacity CU if you need to manage massive-scale datasets where optimizing total cost is prioritized over latency. Please contact sales if you need Extended-capacity CU.

What's the difference between Performance-optimized CU and Capacity-optimized CU?​

The "Performance-optimized Compute Unit" suits low latency or high throughput similarity searches. This option works best for high-search performance scenarios.

The "Capacity-optimized Compute Unit" suits data volumes that are five times larger than the performance-optimized CU option. This option works best for increased storage capacity scenarios.

For more details, see Select the Right CU.