Clarification Needed: Does Binary Quantization Provide Up to 40x RPS Improvement Compared to Original Vectors?

[cainzhong]

According to the following documentation: https://qdrant.tech/documentation/guides/quantization/#binary-quantization

It is mentioned that “Binary quantization can achieve up to a 40x speedup compared to the original vectors.”

I want to confirm whether this 40x speedup refers to an improvement in RPS (requests per second) when using binary quantization, compared to using the original vectors.

[timvisee]

Yes, that translates to 40x the RPS. It is of course dependent on your configuration and accuracy needs.

Some more details here: https://qdrant.tech/articles/binary-quantization/

[cainzhong]

Yes, that translates to 40x the RPS. It is of course dependent on your configuration and accuracy needs.

Some more details here: https://qdrant.tech/articles/binary-quantization/

Is this related to the EC2 instance type (arm64 or x86)? We’re using r7g.4xlarge (AWS Graviton3, arm64).
Are there any additional configurations required? Currently, we’re not seeing a 40x RPS improvement—in fact, we’re not even achieving 1x RPS.

[cainzhong]

#7922 (comment)