All Products
Search
Document Center

DashVector:Vector Retrieval Service product specifications

Last Updated:Jun 03, 2026

Learn about DashVector instance types and specifications to choose the best fit for your workload.

Instance types

DashVector offers three instance types:

  • Compute-optimized: Best for high QPS, low latency, and high write throughput. Choose this type for high-concurrency or latency-sensitive workloads.

  • Storage-optimized (Recommended): Provides 5x the storage capacity of compute-optimized instances, ideal for large data volumes. Delivers strong performance, meets the requirements of most scenarios, and is the most cost-effective option.

  • Free trial: For testing only. Do not use it in a production environment. Valid for one month and renewable after expiration. Subject to Constraints and limits.

Important

A free trial instance expires after one month and is automatically released. All data is permanently deleted and cannot be recovered. Upgrade to a paid instance within 30 calendar days to avoid data loss. Alibaba Cloud is not responsible for data loss caused by expiration of a free trial instance.

Instance specifications

DashVector offers multiple specification tiers for compute-optimized and storage-optimized instances, differing primarily in storage capacity.

Storage capacity reference

Instance type

Instance specification

Doc capacity reference (based on 768-dimension FP32 vector data)

Doc capacity reference (based on 1536-dimension FP32 vector data)

Compute-optimized

P.small

500,000

250,000

P.large

1,000,000

500,000

P.2xlarge

2,000,000

1,000,000

P.4xlarge

4,000,000

2,000,000

P.8xlarge

8,000,000

4,000,000

P.16xlarge

16,000,000

8,000,000

Storage-optimized

S.small

2,500,000

1,250,000

S.large

5,000,000

2,500,000

S.2xlarge

10,000,000

5,000,000

S.4xlarge

20,000,000

10,000,000

S.8xlarge

40,000,000

20,000,000

S.16xlarge

80,000,000

40,000,000

Important

To purchase higher-specification storage-optimized or compute-optimized instances, or for technical support, contact us:

  • Official DingTalk group: 25130022704

  • Technical support email: dashvector@service.aliyun.com

Important
  • This data is verified through testing but is for reference only.

  • The Docs used for the capacity tests contain only a primary key and vector data, with no Fields. The primary key is an auto-incrementing integer converted to a string, starting from zero. In production, Fields consume additional storage, so actual Doc capacity may be lower than listed.

Retrieval performance reference

Instance type

Instance specification

topk=10

topk=100

topk=250

topk=1000

QPS

RT_p99

QPS

RT_p99

QPS

RT_p99

QPS

RT_p99

Compute-optimized

P.large (based on 1 million 768-dimension FP32 vector data)

962.6

<30 ms

429.7

<30 ms

387.5

<45 ms

134.7

<250 ms

Storage-optimized

S.large (based on 5 million 768-dimension FP32 vector data)

297.6

<30 ms

112.5

<30 ms

107.4

<50 ms

37.1

<300 ms

Important
  • Test results are based on the Cohere dataset. Performance varies by data distribution, so these values are for reference only.

  • The Docs used for the performance tests contain only a primary key and vector data, with no Fields. The primary key is an auto-incrementing integer converted to a string, starting from zero.

  • DashVector performs background index optimization, typically completed within 4 hours of a write operation. Performance is optimal after optimization completes.

  • At full capacity, QPS meets or exceeds the listed values and does not decrease as data volume grows. For example, a P.2xlarge instance with 2 million 768-dimension FP32 vectors achieves QPS > 600 for topk=100.

Number of replicas

DashVector supports 1 to 5 replicas per instance. All replicas contain identical data. QPS scales linearly with replica count, and more replicas improve availability. For production environments that require high availability, set the number of replicas to 2 or more.

Note
  • Changing the replica count affects only QPS and availability, not storage capacity.