Large Language Model—LLM—applications, such as Retrieval-Augmented Generation (RAG) (opens new window) apps, are fueling a surge in vector database usage. Many LLM-based applications require massive datasets with high query-per-second (QPS) rates to function optimally. However, most vector database vendors prioritize storage capacity over QPS.

MyScale's MSTG algorithm is a novel approach to vector storage and retrieval. It stores vector data using NVMe SSDs, greatly increasing data density while retaining good search performance. In MyScale, a single capacity-optimized pod size x1 can accommodate 10 million vectors—each with 768 dimensions—surpassing any competitor's capacity.

Compared to our standard pods, our capacity-optimized pods offer double the vector storage capacity together with a 43% cost reduction.

Using Capacity-Optimized Pods

Note:

Capacity-optimized pods are currently offered as part of MyScale's Standard & Enterprise plans.

The Launch New Cluster page now offers the flexibility to choose between capacity-optimized and standard pods.

Capacity

A single MyScale capacity-optimized pod (with an x1 size) boasts the storage capacity for approximately 10 million vectors with 768 dimensions, surpassing the capabilities of other vector databases, which typically max out at around 5 million vectors.

Our largest capacity-optimized pod (size x32) can host 320 million 768D vectors.

Contact us (opens new window) for larger pods or horizontal scaling!

Performance

Our capacity-optimized can efficiently process 10 million vector datasets (768 dimensions) while maintaining a respectable 20 QPS, achieving high accuracy and performance for filtered vector search, regardless of the filter ratio.

This surpasses our competitors, who only manage 10 QPS with 5 million vectors, and has unstable filter search accuracy/performance. Moreover, they struggle to store equally large datasets within a single pod, thereby limiting scalability, query performance and cost effectiveness.

Cost

MyScale’s x1 capacity-optimized pod delivers exceptional value (2x storage capacity with 4x QPS) at just $68 per month on AWS. When compared to our competitors, whose similar offerings typically cost around $80 per month, we provide double the capacity with 15% cost savings.

Use our Price Estimator (opens new window) to determine the most suitable pricing plan for your vector capacity and dimension requirements.